flow-tips-n-tricks・✏️

[stuck folio]

If anyone else searching like me a way to upscale the resolution of the videos Capcut is the answer, just install it and you can export up to 4k free of charge!

[split crystal]

it doesn't use a prompt expander. VEO-3 is paired with gemini, and together they decide what to create. the VLM understands prompts very well and Gemini is the same gemini you talk to in chat.

[rough wing]

It does use a prompt expander. There are all sorts of geeks that get into the code of these apps and find out things like this.

[split crystal]

Im both a programmer and a ml reasearcher. The thing that is expanding your prompts is gemini itself. In fact, if you care to, you can have the same sort of convo with veo-3 as you can with Gemini, just tell it to create an image of paper, or a sign, and write its responsed

[peak zealot]

is there any way to makle 1 min videos on flow

[edgy geode]

no

[edgy geode]

just make many and out them together

[upper forge]

I've been trying for an hour and obviously overlooking something...

I've got a project made from 4 8sec bits but they won't flow from one to the other. Instead there are little pauses: they don't connect.
After downloading it's the same. 😕

[stuck folio]

You use the extend and keep rolling the dice until you get something acceptable.

[rough wing]

And? Nothing you're saying conflicts with what I said, and now you're arguing with yourself.

[rough wing]

You'd need to tell us how you're doing it. Just that it isn't working the way you want it to doesn't tell us much.

[upper forge]

Probably in a way that's going to make you facepalm head..
I'm working on a tablet using the desktop mode in browser. 😅
I've made a project, a video made and placed on the time line, then by using the extend function I've made a total of 3 more. And after that my only option seems to be download.
I'm pretty sure I've tested the buttons I could find but there are barely any options.
I've got an ultra plan if that helps in your answer.

And if this ends up becoming a 'not optimized for mobile, stupid' kind of deal: I know I know :/

[split crystal]

You started by saying it used a prompt expander. That is a program, or small agent, that does nothing but expand, or fancy up, the text. I corrected you and said it does not have a prompt expander, that it is a pairing of a VLM and gemini, who is an LLM. And now you're trying to say thats what you meant all along. It was not. Gemini, being an LLM, is a highly intellegent AI you can converse with, and the VLM is just as smart. They work in tandum to understsnd what you are asking to have created, nothing is "expanding" your prompt

[half cipher]

Can anyone help me write a better flow? I have an idea to build my own TikTok channel by creating B-roll + voiceovers without subtitles or visuals. I want to know what additional skills I need to make better videos and connect with the audience emotionally.

[rough wing]

This isn't cable news. You responded by telling me what the expander was, after saying there was no expander. That's the entirety of what we're talking about. Now you're just bickering about literally nothing.

Trolls are so silly.

[rough wing]

Mobile or not mobile shouldn't impact the generations, since they happen in the cloud.

I do occasionally see a glitch at the seam, but haven't had it happen several clips in a row like that (made 4 long/~45-75s multi-clip scenes yesterday). I don't know what could be causing that.

[rough wing]

All I can think of is leave Scenebuilder, come back and try downloading it again. Scenebuilder is super buggy, has some kind of memory leak, and gets more and more glitchy the more clips you add, and especially if you take clips out, so periodically leaving and coming back in seems to reset it to some degree. Not sure if it will help with your issue, though, since that sounds like it might be a generation issue in where clips and begin.

[upper forge]

It just felt like missing a merge or flow.
There are people talking and near the end of each clip they seem to be waiting like it's the end, and each following bit is treated as 'new scene'.
Thanks for awnsering btw!

[velvet vine]

any tips on making realistic haircuts? im trying to make the end product a completely bald head (its for a cancer foundation)

so long hair woman shave into bald head

grok is able to make it closer than veo 3.1 for me

[lusty latch]

Were you able to do it? We'll get around this by requesting the object in Chroma Key background.

[rotund blaze]

Yup, even in the Scenebuilder. I have to load it in a video editor and got down to frame by frame and disect it.

[wraith spoke]

Hello, could someone please tell me how to make this kind of video?
https://vm.tiktok.com/ZNRh55SUp/

[somber glen]

Veo keeps cutting off half my character text about halfway through videos it generates. Using frame of the character -> frame of what i want - ex. "Character Line: lorem ipsum lorem ipsum lorem ipsum lorem ipsum lorem ipsum lorem ipsum lorem ipsum lorem ipsum " any ideas

[mystic wedge]

this wont work for my purposes, but I appreciate the response. Yes, I often render on a solid color like blue or gree, but for certain vfx purposes I need alpha only. Also, I wish we could render on square rather than portrait or landscape only. Its crazy how ahead of the game midjourney is when it comes to the animating stuff. Especially when it comes to looping.

[sleek marten]

Haven’t touched midjourney in a minute care to enlighten me what and how you use that for animations?

[urban garnet]

I solved this problem as the veo in flow sometimes hallucinate in the first sec or last sec. You will have to trim that off. I built a script that find the matching frames and trim the gaps. Here’s a video I built using this approach : https://youtu.be/02qCMiMJBjA?si=Sc_OJ_-rftn4ZtuT

[urban garnet]

Retexture is great. But the creativity and the quality is top notch. The only short coming is consistency control

[agile swift]

With the latest update, apparently the term "mannequin" will now almost always generate a non-human looking abstract form as in today's mannequins you see in stores. How on earth and what term(s) can you use to generate a mannequin in what they used to look like: very realistic, lifelike, almost indistinguishable from a person with a slight semi-gloss complexion. Any ideas? Veo 3.1 used to be able to generate that, but apparently they've now force trained it to spew out abstract, non-human looking forms when you use the term "mannequin".. I hate that they've done that! 😔 .... any ideas here?

... and since you can't use negative prompts, you can't filter that absolute garbage out!

I'm now using the flag - inaccurate option with every video generation 🤣😔

how many flags does it take to retrain it back lol

[sleek marten]

Can you explain what you mean by posting an example of your previous „correct“ rendition? And what if you simply explain it as you did here in words what you are looking for?

[agile swift]

Something like this ...

Whenever you put "mannequin" in your prompt now it's something abstract or it removes her hair and/or clothing, looking fake and non-realistic (Mannequins used to be realistic/lifelike back in the day), essentially today's ugly non-human forms you see in stores these days..... complete garbage IMO!

so they went into overtime on the training on that term with this latest update and now it's spewing out garbage!

[mystic wedge]

Midjourney has incredible animation tools, especially for bespoke seamless looping. I use this at work for video game creation.

[sleek marten]

Wha a unusual thing to want to generate 😂 but to each their own. Maybe try alternative phrasing like puppet or something, or like previously mentioned just go into an in-depth explanation of what it is you want to achieve instead of using „mannequin“

[sleek marten]

So here you generate the image first and then animate that, right?

[mystic wedge]

not sure what you mean, no I dont do that. We have artists on our team, make art and then I animate it in MJ. Veo has not worked for me. Using first and last frame should create a seamless loop, but Veo doesnt understand that, MJ has settings specific for looping. Things like character idles for example

also it has square, as opposed to portrait or landscape, which is also very helpful

Veo seems to be set up specifically for video creation, but MJ is more versitale, especially with regards to game asset animation

veo takes many many attempts to make a seamless, clean loop, with lots of prompts, MJ is better at one shotting it

looks like this

and it has settings for seamless looping, you dont need to prompt for that

veo, I have to say things like, seamless loop, locked shot, static camera..

[rough wing]

Their extend function is quite bad. If it wasn't, they'd be more of a direct competitor for heavy art styles.

[mystic wedge]

extend, as in time? Yeah, I wish we could shorten the time, in both veo and mj

sometimes I want 1 or 2 seconds, not a minimum of 5 and I never extend

lots of after effects work no matter what platform

[rough wing]

When were mannequins real-looking? Certainly not at any point between the 80s to now, that I've been alive.

[agile swift]

It doesn't seem to matter with this latest update how descriptive I get, it appears to have new biases baked into the model to generate more on the abstract side things and it now apparently wants to take the woman's/mannequin's shirt off, like every time lol! That then causes it to flag it as a violation .... no where in my prompt did I mention to remove the shirt 🤣

[rough wing]

Right. I don't make too many scenes that are longer than 15-20 seconds, but most I make are loner than 5 seconds. They do have an extend function, but it rarely works smoothly.

Basically the opposite of Flow. Flow isn't half as good at keeping complex art styles as Midjourney, but it's great at extending clips.

[rough wing]

That transition from real-looking woman to eerily realistic mannequin was smooth. Nice.

[agile swift]

well, before this latest update you could generate some mind blowing videos like that .... you can't now! they've appeared to have baked a bias into the model to generate egg head, hairless, abstract, non-human, etc. etc. CRAP! ... total garbage! I'm hitting flag - inaccurate with almost every video generated

[mystic wedge]

for clarity, I am not using MJ for art styles, strictly animation

[rough wing]

Of regular, normal realistic people?

[agile swift]

'70s, '80s, '90s and in the early 2000s is when they switched

[rough wing]

We must have different perceptions of what realistic means. My mother worked at an upscale department store in the 80's and early 90's. I meandered around the mall a bunch. I wouldn't call anything I saw realistic.

[agile swift]

Here's another one ...

[rough wing]

Realistic body types, for a runway model, but the skin, details and especially facial features were more like airbrushed magazine paintings.

[agile swift]

okay fair point .... let me rephrase: not necessarily "realistic" ... but "lifelike"

[rough wing]

I don't have any experience with this sort of content, but in general one of the best "tools" you can use in prompting is a thesaurus. Each update alters how models respond to details in prompts, especially atypical material. Periodically it gets a LOT better/worse with directions, for example (I blew a bunch of credits trying to just get a guy walking on a beach to turn right).

All you can do is try different ways of saying the same thing, and reinforce the part of the prompts that aren't giving you the result you want with repetition (I had to say variations of 'turn right' three times in the prompt to get him to turn just a little bit).

[agile swift]

yeah, I totally have to re-engineer my prompts, that's for sure! Complete rug pull on this update 🤣

thank goodness for the free generation option on the ultra plan!!!

[sleek marten]

So here is a result i have from making gemini create me a prompt after explainig the problem.

Prompt:

A cinematic shot inside a luxury clothing store. A beautiful woman sits on a display table. She smoothly freezes into a rigid pose, becoming a hyper-realistic wax figure. Her skin transforms to have a slight synthetic porcelain sheen while retaining realistic facial features and hair. A tailor adjusts her stiff, locked arm. High fashion lighting, 8k, hyper-realism, uncanny valley, rigid body mechanics.

you dont really have the transformation aspect but that could be a clip leading into this i guess.

[agile swift]

excellent! 👏

[mystic wedge]

no, always cartoonish type stuff, nothing remotely like video stuff, video game stuff like this

also, Veo LOVES to make lips move, even when I tell it... character is mute, not saying anything, paralized lips... MJ never does this

silent character is very very hard for Veo

[rough wing]

This and issues with directions (left/right) are probably my two biggest problems with Veo. I had to use Kling to finish a scene a few weeks back because Veo refused to just let a guy mostly sit still, no matter how I prompted.

[agile swift]

never thought of using gemini to help craft a prompt! thanks for pointing that out! Also, this example provided the slight change I needed to adjust with this new update ... thanks again!!!

[sleek marten]

I think the key to generating good video is doing as little prompt writing as possible manually. Use ai to create ai, it knows itself best, is what I usually preach 😂

[mystic wedge]

there is real merit to this approach tho, seriously

[rough wing]

Depends on what you're doing. Shorter prompts tend to work better if you're working with image to video or ingredients to video in my experience. Longer tends to work better if you're doing text to video.

[sleek marten]

Good to know I nearly exclusively do T2V but I’ll keep that in mind

[mystic wedge]

I just meant using Ai to help craft prompts, describing what you want to the ai and have it craft the prompt, its helped me in the past

[rough wing]

I've found the opposite, especially since they're trained to be generalists and have cutoffs that are usually several months in the past, while these models are changing all the time.

I have had success with a bot I made in NotebookLM for Suno, though. It has custom instructions to never provide whole prompts, but to give several specific ideas for each thing I ask about. I'm sure I could do the same for Veo, but I already have prompting notes that take care of that, so I haven't had reason to spend enough time to put that sort of thing together.

[exotic quartz]

You just gave me the greatest Ideas evarrr. Why didnt i think bout it

Thanks man

[rapid sparrow]

Is there any way hot to keep image or video consistent?

[tawdry shoal]

has anybody heard about this platform: https://veogen.studio/
I found out about it while doing some research and it looks like it is still under development, but if it does offer whatever is promised on the site, it would be a very good cheap alternative to google flows since it offers cheap video generations with Veo 3.1 model

[raw egret]

fren, how are you able to generate such long videos?

[sleek marten]

Extend feature in scenebuilder I assume

[crude violet]

I've tried using the scenebuilder and it worked quite well, but when it came time to export two 8 second scenes that were joined together, I kept getting 'export failed'. Tried many times, tried closing the browser and re-opening it, etc. Not sure if it's just the one scene I've built or if it's all of them. I don't want to waste my credits on scenebuilder if they are stuck in Flow....

[hoary zinc]

@mystic wedge since the new update, when a prompt words like "no dialogue" "no speak" "the character dont speak" veo3 make the character always speaks, before the update that wont happen :/

[agile swift]

flow -> scene builder 🙂

[agile swift]

Yes, extend in scene builder

[agile swift]

there's currently an issue with export right now in the latest update.

[long wraith]

Sameeeeee. My character kept doing weird gestures. Wasted 300+ credits for one scene, so I stopped doing vids atm.

[sleek marten]

Try refraining from tellin it what NOT to do. Use words like silent, mute or speechless, maybe that works better

[lethal quiver]

May somebody help me with ingredients to video?

If i upload a picture of my face and i want that exact same reference image starting from that how can i do it? because when i add the other image even tho it works, it completely changes the whole angle

[vague jewel]

@modern mapleHello, as a friendly reminder we cannot generate images or videos in the server! Please do not post your prompts in the channels as it will not generate. If asking for advice or tips on your prompts do so in the appropriate channels.

[molten parcel]

Take your picture and put it int chat gpt or WHISK and use the description of the image in your prompt.

[lethal quiver]

hey ! this actually worked tysm!

[molten parcel]

All good 👍, glad I could help.

[deft summit]

Is there a way to get around the error of prominent people? Trying to make make videos of marvel heroes.

[viscid hedge]

Why doesn't Veo in Flow end up on the actual end reference frame?

[rough wing]

Sora is the only model that's legally allowed to generate Disney IP now.

[mystic wedge]

excactly this, I say things like character is mute, has paralized lips, speaks telepathically... things like that

[exotic quartz]

I have a question. How do you guys keep Mainsubject consistency? Im working for an Architecture company. They asked me to make a video using flow. The damn thing keep doing the house wrong

[dawn current]

A high-quality video of the woman speaking to the camera. She maintains a friendly and professional expression. Her right hand firmly holds the 'ZERO OLOR' box in a static position near her chest (fingers do not move). Her left hand is active, making natural open-palm conversational gestures to emphasize her speech. The box remains sharp and stable. Realistic eye contact, perfect lip-sync, detailed skin texture. Laundry room background.

[primal folio]

Hello everyone I create a extension for help everyone run many prompts with automatically, also has alot of features here is github https://github.com/trgkyle/veo-automation-user-guide

[rough wing]

Reference images, mostly.

[tight burrow]

nano banana pro

[drifting fulcrum]

Hi all, I have a quick question. Is there an easy way to download all images/videos from a project?

[onyx sluice]

I might be late to the game - i haven't used Veo for a while, but i'm using it now for a project.

Are time codes no longer working? I relied heaviliy on time codes for precise blocking in previous projects, but nothing seems to work now. It's like it has a mind of its own, and is completely uncontrollable.

[knotty radish]

is there cheat code for cinematic high ocatin action promting

[sleek marten]

You mean like at 00:00-00:03 the character raises his fist in anger… should still work

[grim cove]

Still flops

[prime lichen]

I need to create designs for swimwear products, but Flow's rules keep giving me error messages indicating that I'm violating the rules. How can I fix this?

[cosmic finch]

Is it possible to create less than 8 seconds video?

[halcyon timber]

You could have it fade in and out and then edit those parts out

[grim wigeon]

hey bro i can fix this

[potent bear]

The colours of my input frame and the output of flow (veo 3.1 - quality) are different.

There is no mention of change of colours in the prompt, except a sentence of the laptop screen flickering.

I've tried deleting this part of the prompt, but no change in results.

Does anyone have an idea of why this happens and how to fix it?

[rough wing]

You asked an extremely general question. To get a nail into wood, you usually use a hammer, but you have to learn how to use it. This stuff is a skill. All you can do is start learning, and it won't be overnight.

[knotty radish]

i'm supr interested in creating cinemtics, but i'm having ahrd time making dynamic shots everything i try is sorta mah - is there any ticks to ordeirng th promnts like a sequencing

[crude osprey]

How I can generate 20,30,40 images in one click 👆

On mobile

Please help

I try to use Computer mode on chrome

[vital hamlet]

Define dynamic shots

Usually it's an issue to make the static ones, honestly. Try googling "camera movements", veo pretty much understands them all.

[stone vector]

u have 2 ways: 1. master prompt of character/building/etc., generate image and use image and master prompt as reference for every video. 2. create frame by frame, 1st frame and last frame of your scene: scene 1 frame 1(start) > scene 1 frame 2(end); scene 2 frame 1 = scene 1 frame 2 > scene 2 frame 2, etc. i use these 2 methods and i don t use anymore transitions in video editor. just cut a few frames if neede

[rough wing]

You'd need to use some kind of custom app that accesses it through the API

[fervent topaz]

can i upload a manual to describe the scene for precision reasons

[rough wing]

I don't know what the character limit is, but you can write very detailed prompts. Usually not necessary with image to video or ingredients to video, but more useful with text to video.

[stone vector]

the limit is 8000 characters

[drifting fulcrum]

No one with an answer to my question yet, is there an easy way for this? No is also an answer, in that case, I'll try to make a quick tool for it myself. 😉

[loud furnace]

Why did I receive a warning that my content was against the rules when generating the video, but my content actually wasn't violating any rules? Is there anyone who can tell me

[vague jewel]

Hey, please open a Modmail so we can figure it out. You can do it by sending a DM to @hardy saffron

[sharp fractal]

When that happen I asked the reason to gemini chat

[rough wing]

no

[drifting fulcrum]

Cheers, will look into doing it another way in that case

[rough wing]

I submitted feedback that asks to have all generations placed in a folder in Google Drive, mainly because the library in Flow is so insanely bad still, but that would allow you to select a ton of files, or whole folders, and download them zipped.

[drifting fulcrum]

Cheers! Do you have a link to that so I can add my support to it?

[stuck folio]

You can try generate the frames with Nano Banana first then do frames to video, I use this all the time.

[median terrace]

Let me dig into how to do that.

[molten parcel]

Go into flow and click image, after making the image click on frame to video.

[median terrace]

Then just run aprompt?

Text is still awful. Unless this is meant to animate the image/frame itself and only use what is on it

[molten parcel]

Whenever I run into issues I run them through ChatGPT or Gemini and it will tell me ways to fix the issue or it will say that it flat out has trouble with that specific behavior and then I know it’s a lost cause.

[median terrace]

Fair

[rough wing]

It was sent to one of the people who works on Veo directly, not on here.

[rough wing]

The more text you have in a video prompt, the more likely it is to fail. That bottom third thing could be added in post using Google Vids or any video editor.

[median terrace]

Yeah issue is Im in 102 cities and its like 75 videos that need to be unique to location

If it was one, I wouldnt be trying to automate this process lol

[rough wing]

Making it work, and altering one word manually is somewhere around infinitely faster than none of them working.

[median terrace]

3 vids per city, unique layouts for each due to andromeda requirements means the text placement isnt the same across the board.

Nice in theory, but hard in practice.

[rough wing]

What are "andromeda requirements"?

[median terrace]

Advertising algo requirements.

Thats the name of it

[rough wing]

Ahh... Meta. Right. Haven't done any ads on Facebook in years.

[median terrace]

Its likely Ill probably run image to video schema through snapgen fow. I already have a majority of the images to make it work.

[median terrace]

New algo is a mess if you are not SOLELY ecomm

Or single location

The demand for new creative is high

Images is easy. I wrote a schema for make.com into OpenAI. Can make 58 unique images in an hour.

Videos proving to be much harder.

[median terrace]

Still more manual than Id like. Frame to Image -> Add to scene -> extend

[exotic quartz]

guys how do you make the camera movement go as you wish? I asked it. Orbit left. it make the video ORBIT right. I asked it to orbit right. it orbit right. Sometimes it did the opposite.

Sometimes it just pan in or go god know where with the camera movement.

[bright pulsar]

its just luck. try fast generation low priority

[slender vessel]

Hello. I want to create, image, that on image I will be seeing person far away. I always have person near. Can anyone help me, how correct write prompt, if I want person away from camera, no close? Thank you for help.

[onyx sluice]

has not been working for me the past week!

[clever nova]

Hello @hallow swan, as a friendly reminder we cannot generate images or videos in the server! Please do not post your prompts in the channels as it will not generate. If asking for advice or tips on your prompts do so in the appropriate channels.

[rough wing]

For some reason, the model often doesn't understand left/right. You just have to burn generations to get it to work sometimes, or find workarounds with end frames and such.

[exotic quartz]

bruh, fr? Hope they add Camera Motion control. Im doing Architecture VIDEOs. Really Really Reallyyyy Need this.

[exotic quartz]

Bruv

Or is there away to prompt for end frame using NANO BANANA PRO by let say move about an inch? How do i do that? Anyone success doin that?

[rough wing]

There is camera control after you've generated the video, but it's buggy and the starting point is the video you aleady made, so if it went the completely wrong direction...

[exotic quartz]

Yeah i know. That's why i didn't mention it here. The dang thing keep messed up my vids so yeah. No, thanks. Hope they'll fix it soon or at least make a Camera control setting before creating video.

[rough wing]

If you need precision, then using both start and end frames is the way to go.

[haughty wave]

hi everyone, did anybody now to extended image in flow? thank you

[sharp fractal]

is lip sync possible with veo 3.1? Using as input "video without sound" + "empty video with only sound" asking for lip sync video output?

[slender vessel]

Hello. I now first use Flow as Image generator, and I don´t under stand, what does it mean icons, when I open generated image. Can you tell me, what this icons means?

Arrow, square, TT, red dot

And first icon, some line.

I am asking, because, I don´t know, what does it mean and I don´t want destroy my image.

[bitter crater]

Does anyone have an idea what I can do, if the Scene builder says export failed? It is a bit frustrating working on a one minute clip to see in the end that the export doesn't work. This really happens a lot and means downloading the scenes and work with them outside of Flow. If you use extend you always losing the cuts and things like that

[quartz ore]

Hello everyone, can we create a 60-second endless runner game background video using Flow AI, without a player or coins—only trains and obstacles?

[mental coyote]

Hi @unkempt panther videos are not created in the discord server. In order to do so you’ll need to go to https://labs.google/ pick the tool you want to work with then go from there. If you need help with prompts please do so in the appropriate channels. Thank you!

[subtle shale]

I bought an ULTRA subscription and as a result, I haven't been able to generate anything for two days now. 90% of my attempts are unsuccessful. Is it just me or are there global server problems?

[untold axle]

buying ultra for this is crazy

[subtle shale]

The pro subscription isn't enough for me. I hit the daily limit too quickly. I generate a lot. Although "a lot" is a relative concept. For some, it won't be a lot.

[untold axle]

what images you generate?

[subtle shale]

They are different, but they don't break the rules and there's nothing prohibited about them.

[untold axle]

i didn't meant that, i was just asking generally

[slender vessel]

Hello, can I ask, when I have a recorded video in Flow, what does it mean: How can I use it? What is it for?

[untold axle]

click and test

[acoustic umbra]

me too, cost alots of money

gain nothings

what is going on ? Google ?

[rough wing]

That adds that clip to Scenebuilder.

Read the instructions.

[sullen holly]

Anyone manage to get consistency with a 4 armed character? It will not stop dissapearing limbs even with several images of the actor in ingredients.

[knotty radish]

I'm being asked by a potentiual client to make pre viz shots whats does that mean ?

and has anyone here ever done that?

[rough wing]

Basically fancy storyboarding.

[silk yew]

Anyone one dealing with scene builder trying to get it to utilize the same generated voice?

[silk yew]

I see now that it just doesn't keep the context of the video... sad sad limitation. Almost zero value add to the "AI Ultra" sub. 😄

[rough wing]

It uses the context of the last second of the clip being extended. All the extend functions in AI video apps do the same basic thing.

[knotty radish]

what should i charge?

[rough wing]

I'm not the right person to ask for that.

[spice garden]

anyone having export issue from the scene builder? how to fix that ?

[quartz ore]

Hello everyone, can we create a 60-second endless runner game background video using Flow AI, without a player or coins—only trains and obstacles?

[quasi tartan]

is the nano banana pro work guys?

[grand sparrow]

hi - in nano banana in flow is there a trick to specify individual attached reference images. like "@img1" etc

[grand fiber]

Using just image 1, image 2, etc. in reference to the order in which the images were added does work, but only some of the time. Describing elements of each image is sometimes needed to stop it getting confused, but nothing works for sure 100% of the time.

[icy rune]

i got something that alot of people can use i think :

You can implement this documunt in your "Gem"

❤️

[stone vector]

we can do gems in flow?

[wintry spade]

[Animation] The figure poses proudly. The mirror reflection stays noble, but the tail of the mouse is revealed in the real shadow.

I got this prompt but flow give this error with frame to frame any help?

[maiden marsh]

@silver solar fwd here for better help!

[silver solar]

Thank you

[rough wing]

People seem to be calling this 'drift'. It's a fairly universal problem, since models aren't yet good at holding onto context like that.

If you describe the voice in the original prompt, and repeat the exact same phrase in extend prompts, that can help, and it's also a big help if that voice you're wanting to use again is active in the last second of the first clip.

If that doesn't work, there are a bunch of lip sync audio AI apps bubbling up, among other workarounds people have come up with.

[toxic glen]

i think i posted my question in the wrong channel

[rough wing]

That's not an option yet. If you're getting some drift, you can basically re-frame the shot and cut in at the same moment from a different angle, with fresh context.

[exotic quartz]

Thanks dawg.

[untold pewter]

is everyone facing problem in Ultra accounts in VEO 3?

cause i am

[stuck folio]

what problem exactly?

[subtle shale]

Can anyone generate videos in FLOW right now? I can't generate any videos; some error occurs immediately.

[warm coyote]

Yes only get random errors. Doesnt tell me what kind of error.

[broken cypress]

How do get the Ai do perfectly what i say?

Every time i use it it doesn't work properly

Do ya have some tips for it?

I wanted to create a cat funny video but i think i haven't gave the ai the right prompt

[primal barn]

use chat gpt to write the promts

Is anyone else having trouble generating images? They're getting an unknown error message that doesn't specify what the error is. It won't let me generate images or videos with any model or format. I'd really appreciate your help, guys. Thanks.

[broken cypress]

do i need to use chat gpt only or i can use even other ai, and in case do i need the pro version?

[primal barn]

you can usea whatever AI you want

[broken cypress]

alr ty

[split crystal]

you can talk to flow, and it will answer you. it doesn't have a chat interface, so you ask it a question and tell it to create an image of paper or a sign, and write it's answer on that. for this, i asked it what the best way to prompt it was. i set to 4 images, here are the 4 responses. if you follow it's list of stuff on these pages, it will be better able to give you what you're after

[stone vector]

use negative promts: no slow motion, no transitions, etc. what u don t want to see in your video

[mild sierra]

Just adding 'TV static' for an audio prompt and it synced the visuals. Pretty good imo. When I'm doing frame to frame object transformations I swap between: transforms, morphs, and contorts. Contorts usually gets there but sometimes it's too dramatic where 'transforms' likes to move slow with quick jumps peppered in.

[grave oak]

me too. I want to know how to keep the same voiceover

[rough wing]

You have to use third party apps for that.

[icy rune]

11labs is the way to go the say

[twilit vessel]

11

[rough wing]

There are a bunch of options, but that's the biggest. Sync is also good, and Kling has a good lipsync, that's availabl in Higgsfield.

[steep pewter]

Hello. I need help to lean how (proper way to use prompts) to use flow and create video.

[rough wing]

Start here: https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-veo-3-1

Practice. Take notes.

[steep pewter]

Thank you!

[rough wing]

🫡 I can't stress the "Practice. Take notes." part enough.

Every time you try a prompt/context and don't get what you want you get a chance to learn a little bit more. Look at the prompt and any other inputs you used, look at what you were hoping for with that prompt, and think about how you can change what you said/uploaded to move the result closer to what you want.

Do that thousands of times. Take notes of what works, and what doesn't.

That's how you get good. Bots are awful at prompting, sometimes you can find shortcuts to specific results, but if you don't try getting there yourself and "get in those reps" (practice), you'll never learn.

[plush totem]

Any prompts or tricks to get ingredients to video to work like frames to video using the first image?

[patent geode]

In Flow, all output generated using Veo and Imagen will be embedded with a SynthID invisible watermark. We directly embed the SynthID digital watermark to help identify AI-generated content.
Google One Pro users' generated videos will have an additional visible watermark indicating that the content was created by Veo using AI, making it easier for others to identify.
Google One Ultra users' generated videos will not have this visible watermark.

Will Google One Ultra users also have SynthID embedded?

No wonder all the videos uploaded to YouTube have zero views! Is it because you've all added watermarks and are preventing them from getting any traffic?

[fathom hare]

So, ultra plan doesn't have any watermark - neither SynthID nor additional watermark like Pro?

[waxen widget]

Hi

[rough wing]

They all have synth ID.

[rough wing]

Yes.

[rough wing]

The YouTube algorithm doesn't care what tools you use.

[final mason]

What's wrong with Flow that it doesn't generate simple images of a person walking in a kitchen? What kind of idiotic filter did they put on it? I'm going to cancel this crap and look for something better.

[patent geode]

All plans include the SynthID watermark.

[patent geode]

The CEO said he would use tools to scan AI-generated videos, and a large amount of AI-generated "slop" would be removed.

[rough wing]

He said they're looking for slop, not all videos made with AI.

[rough wing]

“To reduce the spread of low quality AI content, we’re actively building on our established systems that have been very successful in combatting spam and clickbait, and reducing the spread of low quality, repetitive content."

[exotic quartz]

Guys how do you make the camera go extremely slow ( Cinematic slow ) or all i have to do is pray it works??

[rough wing]

Camera stuff is a big weakness in the model. Try to find a few different ways to say what you want, and put it near the start, in the middle and at the end. If you can't get it to work (which is normal with slo-mo), then you can use a video editor to slow it down after you download it.

[rough wing]

Looks like the image upscaler is back up on Flow, at least with newly generated images.

see

[sharp fractal]

I'm trying to create transitions in a 8s clip, but it seems impossible. How can I create smaller than 8s in flow/whisk? Isn't possible? Is in the roadmap?

[rough wing]

They don't have a roadmap. You can make 4, 6 or 8 second clips in the API, but it's only 8 in Flow.

[gleaming sandal]

how i create a flow account for free?

[hasty reef]

you just use your google account and use it but you get only 100 or 200 credits for free each month if im right but im also new

[patent horizon]

I use a black frame as the final frame using frame to video, and then will manually cut down, adjusting prompt if required

[wide stone]

hi guys, I'm using flow to generate videos and I want to include it into my n8n workflow, does anyone know how to do it?

[patent horizon]

you would need to use the API

[wide stone]

it tells me that i'm on the free tier, although i'm on the pro plan

[patent horizon]

Using via the API will cost, I'm on Ultra and build my workflows around manually using in flow to help avoid the added cost

I would love for it to be included though so I can just put together AIO tools instead of having to go back and forth 🙃

[visual crane]

hey guys im just trying to make a simple product ui using flow but it seems a lot more trouble then i thought.. PS im a coder/animator but i like the idea of video to animate

[broken ruin]

Hey, guys, need help with the prompt for creating the video of the signature that I need to add at the end of videos. Any help with this will be greatly appreciated

[patent horizon]

I would use the frame to video tool and place the signature with the first frame. Try to do a basic prompt and see how close it gets you. If results not desired, save video, upload to a fresh Gemini conversation. In the conversation use:
three quotations to wrap the prompt you used in so:

“””
prompt here
“””

And then try and describe what worked and what didn’t. You may need to do it once or twice, and can keep using the same conversation for revisions

[jaunty meadow]

I have created this intro video with prompt: check it out
https://www.instagram.com/reel/DUFhmMeiJ2S/

Prompt: Ultra-realistic cinematic portrait photo of the subject. (STRICT identity preservation). First photo is primary for the scene and pose details, Use second image reference only for facial identity Create a cinematic digital illustration of the person in the reference photo Maintain the exact face identity hairstyle, expression skin tone and pose with zero distortion Transform the entire scene into a deep red high contrast Tamil gangster movie poster style. Replace the background with red gradients smoky atmosphere and subtle textures. Add bold comic book shading thick outlines, strong black shadows and glossy highlights on the skin and clothes. Stylize the outfit with richer tones clean folds and dynamic lighting while keeping the original clothing design Add intense highlights to the jawline cheekbones and edges of the body. Maintain the original pose and attitude Create a gritty powerful cinematic mood with dramatic lighting and high contrast Ultra-detailed sharp lines vibrant red atmosphere action movie poster energy, 4K illustration

rach.it.ai

Character Intro - Trending AI video , Create yours

Image prompt: Ultra-realistic cinematic portrait photo of the subject. (STRICT identity preservation). First photo is primary for the scene and pose details, Use second image reference only for facial identity Create a cinematic digital illustration of the person in the reference photo Maintai...

Instagram

[jaunty meadow]

us ai studio instead, that will help you out

[toxic venture]

how do we get Ai Studio

[rough wing]

AI Studio is free, but i think you only get a few Veo gens before you need to start paying for it via API.

[toxic venture]

do i just download it/

?

[rough wing]

its a web app. Im on my phone, but just google it.

Its a sort of entry level app dev tool.

[broken ruin]

When I use Frame to Video and signature with the first frame, by default, the video starts with the signature. I need each letter exactly handwritten as my signature so that this wil be used as extro for the main video. Please let me know if any suggestions

I tried with first frame empty and second frame with the signature. It worked. Thanks a lot

[broken ruin]

whats the importance of this please:

“””
“””

[lucid jasper]

Can someone help me? For example, every clip has this sound at the beginning. I've already told Veo everything I want to hear, no music or anything, but he still does it. How do I get rid of it? I always have to cut it out. Dankeschön

[patent horizon]

Ahh sorry I mis understood, but using the second frame was the next suggestion, glad you got it!

[patent horizon]

Similar to using sql blocks for code, but I’ve found over the past year specifically with models by Google when reading through the “thoughts” when using knowledge bases, lm notebook, deep research that it doesn’t get it confused as much and can work with it better for modifications

Not a must do by any means but I’ve found I get better results with it more often now

[rough wing]

I have the same issue. I usually just cut all the audio it comes with entirely.

Oddly enough, Suno does this often as well.

[maiden marsh]

@rancid falcon fwd here as more appropriate!

[rancid falcon]

Help me please

[patent horizon]

Use flow “create an image” instead of Whisk

[mental coyote]

Hi @winter flower videos aren't generated in the discord server. In order to make a video you'll need to go to https://labs.google/ pick the tool you want to work with then go from there. If you need help with prompts please ask in the appropriate channels for advice. Thank you!

[edgy wharf]

Is there some ways or tricks to make consistent characters in veo3? Since it's not like sora2 who has a character option.

I know ingredients to video will be helpful for the looks, but how to make consistent voice? if I use other voice generator to replace the voice, it won't match the speed and emotions

[near garnet]

Hi gaiiss, I would like to check what's the best way to make each ingredient in my food photo animate one by one?

sorry newbie here

[short leaf]

reference image

[iron ice]

create a multiple angle image with ur character - use as an ingredient - works perfectly almos everytime

[mental coyote]

Hi @sleek belfry videos are not created in the discord server. In order to do so you’ll need to go to https://labs.google/ pick the tool you want to work with then go from there. If you need help with prompts please do so in the appropriate channels. Thank you!

[rough wing]

if you can isolate the voice, you can run it through a voice changer

[edgy wharf]

thank you.
voice changer that can make different voices sound like one person? I never used one, any recommends?

[stray thicket]

@zacro. Create base video and then use extend option.

[rough wing]

ElevenLabs is the industry standard at the moment. Hedra and Sync are apparently good as well. I do all the voice acting in mine, and use a voice changer in a totally different way, so I don't use them myself.

[brave geyser]

hi is there have any way to make video larger than 8 seconds?

[rough wing]

In the Flow app, you start by making the first 8 second clip, then you add it to the scenebuilder and use the 'Extend' function to generate a clip that starts where the first one ends.

[rapid olive]

Hi everyone! Is there a way to maintain a consistent voice across multiple video generations? I've tried using a JSON format with a detailed voice description, but the output keeps changing every time. Any tips on how to lock it in?

[terse snow]

has anyone got any suggestions to make scenes with the same characters, I have 12 characters I want to put all together. I’ve made character sheets but ingredients to video only allows 3 uploads, putting all the characters on 1 sheet or splitting up into 3sheets still doesn’t work as flow fails to generate all the characters or blends them together. I’ve tried adding prompts for each character along side the character sheets but the same issue, the best results I’ve had was with no prompt and only the reference image but it’s still not working for all 12 characters in 1 clip? Not sure if there’s anything else I can try or is 12 at once just beyond the limits of veo3?

[untold garnet]

hey, I was wondering if you guys could help with this swap for the Napapijri jacket. For some reason I can't get this product fit right on my model. It's a relatively new model, and Nano Banana Pro gives me the impression that the focus goes towards the classic Napapijri jacket (with color in the Norwegian flag).

[trail canopy]

The best way to fix this is to create or use a stock voice of your choice in Eleven Labs. Take the eight second clip and extract the audio. Use Eleven Labs voice to voice to apply your Eleven Labs custom voice to the clip. Do this with every speaking part of the character in every clip.

Tricks and tips for efficiency: Edit the entire video you’re building first in a video editor, including cuts. Select all of the clips with your character dialogue and combine those only into a compound clip. Save that clip and separate the audio. Run the entire audio clip through voice to voice. Bring that full audio back into your editor and turn off the audio in your compounded clip.

[honest hemlock]

does anyone know if upscaling scene builder is possible

[waxen swift]

I keep getting failed generation when doing picture to video. It’s an innocent fashion image so not sure why it’s not working so I have to put the end clip in aswell?

[rough wing]

If it is getting filtered, you'd get a different error.

[waxen swift]

So just keep trying and it should eventually work?

[rough wing]

Fails go up and down during peak hours. Could be something else. They're super cagey about this stuff, so we never really know what's going on.

[waxen swift]

Ok thanks I will give it another go later

[outer shell]

note the peak times to avoid. mine was noon to 3pm pst. the renders were half quality

[rough wing]

That would be logical, heh. I just come back later.

[outer shell]

cool down for 6-8 hours. vid request get soft reject after 50 clips non stop clips

[modest marsh]

Hey! I've been extending doing that but it doesn't keep the accent ie from UK to US in the extention. I tried adding the same accent prompt in the extention but it came out with a different UK accent. Any tips on how to keep exactly the same voice through each clip? Thanks!

[rough wing]

The way extend works, it's based on the last second of the first clip, so if you want some level of consistency it needs to be in the last second.

That being said, audio is the newest addition to the model, and if you're specific of what you want, the best way to do that is use tools like ElevenLabs, Sync or Hedra to replace the voice across clips.

[stuck cypress]

Hi, I'm an AI Pro user and new here.
For one of my projects, I'm trying to create a kind of short film in Flow. I want to use “Veo 3.1 – Quality” for scene generation and “Nano Banana Pro” for the frames.
I want to record all project-specific information in as much detail as possible in a Google Doc document, and I'm using Gemini 3 Pro to help me with the project, but I'm now reaching various limits and would therefore be grateful for any human assistance.
One issue is that the prompts generated by Gemini for the frames and scenes do not have a consistent and uniform structure. I would like to change that.
Can you recommend a specific structure?

[sleek marten]

I never use quality unless it’s forced upon me, fast is so much cheaper and looks the same. Tell Gemini to craft you a master promt with your desires to use to generate prompts for veo3. Use that at the start of each conversation to prime it into giving you what you need. For example this is my current master promt to initiate Gemini to receive a script and generate veo3 prompts for every single line and splitting a line up if it’s too long for 8 seconds. https://pastebin.com/dMHEVxpT

[stuck cypress]

Thanks for your reply! What do you mean by not using quality unless it’s forced upon you?

[sleek marten]

So you don’t need to use quality unless you want to use the „jump to“ Funktion. That’s the only time i use it, because I have to. Otherwise for everything else, ingredients, extend, text to video I just to fast and 4 generations per prompt, you can to 10 fast generations for the price of one quality and it literally looks and sounds the same. So a total no brainer

[stuck cypress]

So the only thing that really matters about 3.1 quality is the amount of credits you have to pay for each generation, but it doesn't offer any useful advantages?

[sleek marten]

Not that I am aware or have noticed in direct comparison. But try it out for yourself with the same prompt and see if it makes a big enough difference for you… normally if I can do 10 generations compared to 1 I will always get something that is good in the 10 vs not always something useful in the 1 if you get what I’m saying

[stuck cypress]

Okay, I'll give it a try. What about excluding info - does Veo understand negative prompts and if so, how should I add them to the prompt?

[sleek marten]

I have added „no words spoken“ and it seems to understand yes. But if you can phrase it in a positive way that is always more advisable.

[stuck cypress]

Hmkay, because I tried things like "no sound" and it still generated videos with sound. How should I phrase negative prompts in a positive way? ^^

In another example, Gemini gave me a prompt like [Negative prompt: No frantic movement, zooming, camera movement, morphing, distortion of the island, distortion of the coastline, static water, frozen trees, text, people, audio, blurred, washed out, glitch] and Veo still added camera movement and zoom to the clip.

[rough wing]

It is better at some things, but most of the time it's better to start with Fast.

[rough wing]

No.

[rough wing]

This is a great example of why you shouldn't use bots for prompting.

[toxic venture]

how do i create the same character for I want to be consistent in who the character ?

[rapid pulsar]

Is anyone facing problems generating on FLOW?

[tawdry hearth]

👀

[stuck cypress]

Can you tell me at which things Quality mode is better than Fast?

[stuck cypress]

And how can I then prevent Veo from applying certain/unwanted settings?

[normal coyote]

There are two valid ways to use negative prompts in veo:

• Dedicated Negative Prompt Field [Can use "no"]:

no text overlays, no watermarks, no cartoon effects...etc```
In your case: 
```[Negative Prompt]
Music, score, melody, cinematic, soundtrack, instruments, youtube background music, distorted.```
* Embedded in the main prompt [Avoid "no"] when you're embedding negatives directly into your main prompt [not a separate field], you should describe what you don't want without using "no" or "don't":
```"without modern weapons"
"avoid surreal visuals"```

[rough wing]

I've noticed it being better at fast complex movements and physics, for instance. You just need to test it.

[stuck cypress]

Just to clear up any misunderstanding from my side:
If I use the first approach, I don't have to or shouldn't use "no"?

[rough wing]

Positive prompting. Say what you do want, and use images and ingredients.

[rough wing]

Telling it not to do a thing still often makes it more likely you will see that thing. It's unreliable at best.

A good rule of thumb with negative prompting is if there isn't a box in the UI specifically for negative prompting, then negative prompting probably isn't a good idea.

[stuck cypress]

@rough wing
Well, as an AI Pro user, I only have 1,000 credits per month for video generation. Since creating a video with “3.1 quality” costs 100 credits, I can't perform any meaningful tests. Therefore, I am asking for helpful information so that I don't waste any credits unnecessarily.

[rough wing]

Right. Even if you have more, working your prompt out in Fast will save credits.

[stuck cypress]

I don't think I have any other choice at the moment.

[normal coyote]

As Frontier mentioned to you, inside the main prompt [no negative field], you should rephrase it.
For “no music” you can write, for example:
silent video, no background soundtrack, no music, no instruments, no sound effects.
Or more positive:
completely silent video, only visuals, no audio track or music of any kind.
This makes it explicit you want a silent clip, while avoiding the bracket syntax that veo/flow doesn’t actually parse as a special negative prompt block.

[nova star]

hi

[stuck cypress]

@sleek marten, @normal coyote & @rough wing
Okay, I thank you all for your tips and for your help!
I really appreciate it!

[acoustic furnace]

Any way to speed up upscaling process and to hide watermark?

[normal zinc]

how to make continuity videos in veo3

[rough wing]

In the Flow app, you add a clip you like to the Scenebuilder, then use the 'Extend' functon.

[untold garnet]

hey all, Kling is still superior when it comes to room transformation videos. However I was wondering if someone also got some nice results by using veo3.1? I'm currently experimenting with JSON prompt, and I must say the results get better, only thing I experience is last second changes to the room which mess up the video. Attached a video of an example.

[fluid elk]

any tips to make the cameras in edit mode to follow the voice better and also better quality?

[somber glen]

is there any way to have veo not generate soundeffects other than a voiceover? have tried prompts but it still always gives it

[fluid elk]

nothing that I have found. I have to run it through Podcastle and that removes the extra sound. I think Elevenlabs may do that too

[rough wing]

It's automatic in 3.1

[rough wing]

All the video editors can strip it off. I usually do, unless it's just background noise that I can mix with my own well.

[noble orbit]

Hello, AI Plus user here. How do I remove the VEO watermark on the right bottom side

[patent horizon]

pay $250 a month for ultra or find a online tool thats most likely paid.

Or can get creative

[neon sky]

Dont pay, google literally uses stolen work themselves

Easy, legal ways ti remove the water mark.

[astral bone]

Has anyone been successful un generateing a 360 degrees camera rotation around a character? It's quite standard in CG industry to render a 'turntable' of a character standing in a frozen pose, bt I'm having trouble to get a clean turntable with VEO...
It only rotates the character 180 degrees, then mess up the remaining rotation.

[short leaf]

@astral bone havent tried but i assume if you get a picture of your character in front side back and feed it as reference it will work

[rough wing]

Yes, using character sheets like this as ingredients is super useful.

[sharp fractal]

I like the emotions of the faces you did here.

[rough wing]

I ended up going with a different character model, but he's a stoic character, so not much emoting.

[patent horizon]

Yeah, that and and 360 degree pano videos have been some of my most used workflows with Flow. DM if youre having issues or need some help and I can try to assist

[patent horizon]

While 100% true, you didnt mention literally any LLM / generative media model is also in that category as well

Any AI that tries to market itself as "ethical" is doing that because its the only way they can advertise something to be able to sell a worse product to get people to use; like Adobe as a perfect example.

[rare trellis]

I have a character that’s a bottle with eyes and a mouth and I want it to speak. I add the speech and everything, but it keeps making it sing. Is there any way to block that?

[echo wharf]

Hi all. Am using Flow, and despite putting in the following guidelines to the prompt about not including any background noises or music, it seems to creep in most of my renders? Please find my prompt below - any suggestions? Thanks in advance for your help.

A sustained close-up as the man begins walking forward. The camera dollies backward smoothly, matching his movement to maintain a tight close-up on his face. With a slight, confident smile, he speaks clearly and calmly: ‘I want to start with a story, and it’s not about us…. It’s about our customers.’ Only his natural speaking voice is audible. The environment is otherwise silent, with no background audio before or after the dialogue. After he finishes speaking, there is a full one-second pause at the end of the clip. During this pause, his mouth is fully closed, he maintains a slight smile, and he remains still. This one-second hold is clearly visible.

[rough wing]

Negative prompting rarely works in Veo.

[echo wharf]

Thanks. Are there any workarounds you could recommend?

[rough wing]

Erase the audio and replace it. If you want to keep the dialogue, isolate the voice and replace the rest.

the Google Vids app is good for this. Has tons of free stock audio.

[carmine crown]

Hi friends, I'm just starting out with AI, can anyone help me with good clothing prompts for Flow?

[clever mango]

Like a fashion cloth ?@carmine crown

@stuck cypress what types of videos?

[stuck cypress]

What do you mean?

[clever mango]

I thought you need prompt to make video to manage your credit?

[stuck cypress]

How often do you get exactly this character sheet template when you try to create character sheets?

[stuck cypress]

Sorry, I'm completely clueless atm 😅
Which of my posts are you referring to?

[clever mango]

Here

[rough wing]

The layout usually works right away. I usually go through a few iterations to get it all looking good enough to use as ingredients, with good facial expressions, consistency, detail and all that, including after upscaling. The more data you have in the ingredient, the better the outcome, so upscale to 4k.

[stuck cypress]

Ah okay, well, I want to create any kind of video, but right now I'm working on an animated music video

[stuck cypress]

Would you mind sharing the prompt which you used to create this character sheet?
I'm trying to build a prompt for a character sheet template with a dummy, which then only needs to be replaced by the respective character

[clever mango]

Okay

[manic scarab]

Ho

[rough wing]

"Generate a fully photorealistic character design sheet based on the character in image 1, using a layout like seen in image 2, with sections showing three full body views (front-view, side-view, back or rear-view), one up-close shot of his shoulders and face, and facial expression examples filling the rest of the sheet. Make the grid of expressions somewhat stoic."

That was where it started. Went back and forth a bit to get it to where I wanted it.

"image 1" referred to a screenshot of a 3D model I'd made in Character Creator 4 (a Blender-related app), but you can just make a character design with AI. Just try to make a nice clean headshot.

"image 2" was previous character sheet (use mine for that if you like the layout).

Replace the last sentence with whatever sort of expressions you want for the scenes you'll use this in.

Modify from there.

[lapis flame]

Creating a new a account for credits with test subscription doesn't work btw

[stuck cypress]

First of all, thanks for sharing the prompt and your tips!

In terms of referring to attached images to get certain results, would you say it is sufficient enough to just tell Veo to respect image 1 for this and image 2 for that or would you recommend to enter the file names of the images?

Since I'm a gamer and also interested in game development and 3D modeling, I'm aware of Blender and all the (expensive) Reallusion applications 😅
It's definitely a great idea to use a 3D model as reference for character creation in Flow. I hope that someday, when Project Genie reaches that point, we'll be able to also import and use 3D scenes as scenes for video generation in Flow. Maybe we'll then also have some kind of feature to import, save and use consistent characters in Flow 😍

[rough wing]

Helping is why I'm here. I love this stuff, and the documentation for these tools is pretty bad.

[rough wing]

I usually get the result I want by referencing them by numbers in the order I upload them. Try both, and tell me how it goes 🙂

[rough wing]

Yeah, Character Creator isn't cheap. Not sure if I'll upgrade to CC5. I have CC4 with a bunch of add-ons, so I can basically make people look exactly how I want, but not quite realistic, then bring them into Nano Banana Pro and get them the rest of the way and into a character sheet.

From what I understand, there's a free tool in Unity that works similarly, but I haven't used it.

[rough wing]

One thing I've brought up in feedback a couple times (back when I was in the Veo beta, and since) is instead of images as references, allowing us to use 3D models (.obj, or some such), so the model doesn't have to assume anything, but can just show that model exactly. Should save the model compute if it worked that way, but probably at least 2 generations away from how the models work, so maybe Veo 5 or 6?

[rough wing]

Exporting from Genie is at least that far away as well. Those scenes aren't actually 3D environments, so to export them you'd have to connect the output from Genie with some sort of next gen 3D model generator. That would be super useful, though.

[stuck cypress]

There's also MetaHuman Creator from Epic/Unreal Engine. It doesn't provide all the possibilities which Reallusion's Character Creator offers, but the characters still look very nice and it's free, too

[rough wing]

That's the one. Unreal, not Unity. My bad.

[stuck cypress]

Unfortunately, I wasn't able yet to test Genie as it isn't available in Germany, so I don't really know what it's actually capable of, but I hope to see that some day

[stuck cypress]

np

[rough wing]

I actually installed Unreal Engine on my PC months ago, but never found the time to try and learn how to use it. NB Pro came out and it's just not necessary now.

Words just aren't good enough to make characters if you have a specific vision. CC4 is more than I need with NB Pro as a bridge. Hopefully some lighter tools fill that gap, because words just are too imprecise.

[stuck cypress]

Yeah, that's true. I had several failed attempts in the last days, because I tried to build prompts which were as specific and explicit as possible, but today I tried another approach. I just attached a picture of something that I wanted to have in a different style and then only added the desired style as prompt, which gave me way better results. Sometimes less is more! 😅

@rough wing
I just watches two videos about Project Genie, but they didn't gave me answers to this: So downloading the generated worlds isn't possible ATM?

[rough wing]

100%. It's a fundamentally better way in a visual medium. There's FAR too much room to interpret with only words.

Same with music. I do very deep prompts in Suno, but they could be followed perfectly and still give me 1000 very different songs.

[rough wing]

There is no "world"/3D environment to download. It's generating the video on the fly, like Veo, but with more going on in the background to keep consistency. (and faster)

[stuck cypress]

Yes, that's right. I think we've already talked to each other on Suno Discord 😅

[stuck cypress]

Ah okay, I hope this will change ASAP

[rough wing]

That stuff is very, very new. If it gets to that point, it'll probably be years.

I'd bet that some of these AI 3D model generators would get to something like that much sooner. They're already getting very good at individual models. I'd assume they'll start expanding into environments eventually.

[acoustic juniper]

Hi. What do you guys think about this advertisement? Is there anything I can improve for my next videos? I created it using only two tools: Veo 3.1 Fast and Nano Banana Pro.

[bleak merlin]

I love this . It's creative and very clean .

[acoustic furnace]

Just add a square box of similar colour as background using any editor on the text

[stuck cypress]

But then we still run into the issue that Flow/Veo needs to be able to support and use this models as reference

[rough wing]

Yeah, I'd love it, but don't see that happening for a while.

[devout widget]

Suno has a few tags and formatting tricks to get specific intonation, inflections, or music "mechanisms" when you're trying to invoke a certain kind of feeling. I managed to get a barbershop quartet to beatbox and hum pretty accurately. You have to mix in some instructions with your lyrics though and a few lyrics in your style.

[rough wing]

Oh... I'm very aware 🙂

99% of what I make is instrumentals (mostly jazzy bluesy stuff), but even with that you can use the lyrics box.

[devout widget]

I don't believe it renders normals, but I believe the same principal is why it goes wonky at 180°

[rough wing]

You should be able to get consistency with a good character sheet, but the issue is camera control. It's often very difficult to get Veo to listen to specific camera instructions, like smoothly orbiting in one direction all the way around something.

[devout widget]

I made that comment because I noticed that closer to 90° rotation either way the perspective becomes a bit askew. It's 3D to us, but anything not directly rendered doesn't seem to occupy 3d space. Such as a claw sweeps over a leg and as it passes you have an extra leg or are missing a segment of said original leg. But the video gen seems to prefer cutting the scene instead of fulfilling our request for orbital camera movement.

I've been trying to make character sheets myself, but using Gemini TTI. Simple or complicated I always end up with two lefts, or two backs. Metahuman you say works well?

[rough wing]

Character sheets get rid of most issues with limbs that don't look right. Not the camera stuff, though.

[rough wing]

Can you show me what you mean by "two lefts, or two backs"?

Share the prompt, the result, and describe what you want that it's not giving you.

[devout widget]

Sorry must have erased them or archived. I can find a few 3 image ones... This is from a while back before they could have more than 3 characters in the same image without each character's characteristics categorically conceding to chaos limbs or digits, or loss of character.

[high sedge]

Anyone can help to stop veo3 from zooming?

Is there a prompt that would 100% fix and apply steady camera during video generation?

[sharp fractal]

Would you like to share the prompt that’s currently giving you trouble?

[sharp fractal]

I built a GEM in gemini with some notebookLM sources and it says (not tested):

Add hese keywords at the very beginning of your prompt:
​"Static camera, fixed tripod shot."
​"Lock-off shot."
"Zero camera parallax"
"Fixed camera"
​"No camera movement, zero zoom.
Give a reference or where to focus. Example "fixed window frame".
Define the distance indicating medium wide shot oe full short or whatever.
If using the api, you can have negative prompt: zoom, dolly, push in, pull out, camera tracking, motion blur, parallax.

Example: "Low angle static shot, locked-off camera...."

[wintry carbon]

Can someone help, I have been trying to create consistent character images in nano banana pro, but it's not using my reference images, it's using random people
Also how will I tell nano banana who is who example if I have mister a and mister b, should I name the file as Mr a and Mr b?

[rough wing]

Refer to them by order of upload.

[high sedge]

thank you

[wintry carbon]

When i use frame to video and if they have two characters in that image and give dialogues, veo3 mixes up the dialogue always, if there are three charectors, its even worse it , can anyone please help me with this

[rough wing]

I just break it up, so only one person talks per clip.

[plush totem]

Genie looks awesome, that was actually the first thing I thought would make flow better after using it. It took me forever to generate the right images to create an off grid property for my short video series.

Still not perfect but I was able to create a little drone flight to grab screen shots from different angles.

[echo storm]

How do you manage when the camera simply won’t follow your instructions is regards to orbit, or dolly or handheld?

I’m trying to get the camera to move about a pint glass on a table top, and whereas the prompts worked on Friday night, the same prompts keep the camera locked static today.

I’ve had this problem a number of times where prompts worked one day but not another.

[rough wing]

Do something else, or use a different tool.

Sometimes Veo just gets stuck on a particular way of moving, especially with image to video in my experience, and when it does, it's a huge pain to try and push it to do anything else.

I blew over a HUNDRED generations recently trying to get it to simply turn right, and the best I could do was a gentle veer.

This has been an issue with Veo since Veo 2, but was getting better... then with 3.1 it got worse.

[echo storm]

I don’t really understand how Veo or LLMs work in the backend, but is it possible Veo has some sort of server/location-based issue? Perhaps, depending where you are and the time of day, you may have access to different servers that act differently.

Like I said I’ve no real idea how these work, but it seems odd that a prompt that works fine one day simply stalls or ignore certain instructions the next. It’s as if you’re speaking to two completely different brains.

[rough wing]

Theories abound, but some of it is the randomness inherent to how AI works. At busy times, it does seem like they use lower-quality "quantized" versions of the model, and the prompt interpreting gets worse, but hard to tell. Could be other stuff. Every update is 2 steps forward and 1 step back, so could be a '1 step back' thing.

[distant moat]

why cant i upload my self when i was a kid photo to edit?

[desert basin]

Is anyone here still experiencing reCaptcha errors when producing veo3 flow videos?

[stuck cypress]

Do you know of the additional settings you get access to when editing a generated video? There you can change e.g. the camera movement and position, but this costs additional credits based on the selected model.

[stuck cypress]

Need some help!
I'm trying to create a scene where the camera flies towards an island, but I'm getting weird results as there's always coming something out of the water. First time it was an ape, then a frog and now it's a drone 😅 🙈
For the last gen I used following prompt:

Calm drone flight towards the rocky island.
Calm sea, gentle small waves. The movement is steady and gliding, not hasty. Warm backlighting.

Regarding the prompt it's somehow comprehensible that Veo tries to show a drone, but this shouldn't happen.
Does anyone know how to prevent this?

[rough wing]

Very buggy, but works sometimes.

[rough wing]

Swap out the drone bit with something like 'sweeping establishing shot'.

[stuck cypress]

Yeah, I changed the specific part of the prompt and got a better result.

[plain stag]

Surely there has to be a technique where you can force a generation on Veo to do it in the accent you tell it... I am losing so many credits trying to create a speaking piece where all it needs to do is keep using the fricken accent I tell it to!!! Despite saying British accent like 3 times in the prompt it took 5 generations to get one with that... This has to be fixed as its a waste of credits and money...

Tomorrow I want a way I can get the correct accent in ONE generation...

[rough wing]

No. The audio part is still in beta. Doesn"t prompt well at all.

[wintry viper]

Don't rely on veo3 for the voiceover it's not that good
Use Elevenlabs it's way better

[rough wing]

^ Right. ElevenLabs, or Sync, or Hedra, etc. I've worked out a workflow using a voice changer, and I do all the "acting" myself, so I have perfect control over inflection, tone, timing, etc. None of the video models are very good with dialogue, and it's too expensive to burn through a bunch of generations to try and get it to sound right. Audio models are much cheaper.

[plain stag]

But then its not synced to the lip movements

[rough wing]

There are different ways to do this. You can take what audio the video does give you and run it through a voice changer (AI or not) so it's all consistent, or you can replace it using different audio and using some of the aforementioned models' 'lip sync' capabilities.

[plain stag]

You can’t change accents tho which is the problem

[rough wing]

I don't have secret Google information. All I can tell you is there are options that do what you want. No existing model does both video and audio with the level of contrl you're looking for.

[heavy totem]

you can dude, you need to give it a nationality in the prompt and also, it will pick up on dialog, a british person will say something differently than an american person i.e use more formal language and finally give it a name that fits the nationality i.e british something like "Amelia Potter", american "Janet Smith", Svetlana Korinkova", ertc.

you can stack those three to reinforce

remember VEO is a real-world model

thats the key to undestanding how to prompt it

[plain stag]

The thing is I do it says in a British accent like 4 times

[high sedge]

any tips on how to prompt the avatar not to talk just pure actions?

[heavy totem]

DM me your prompt

[rough wing]

Describe the actions and say how they do it silently. Find different ways to say it positively, instead of what not to do.

[odd birch]

Hello @still pike as a friendly reminder, videos and images are not generated within the server, you'll need to head to https://labs.google/, pick a tool, and generate it there. If you are asking for prompt advice please do so in the appropriate channel. Thank you!

[coral arrow]

Is it possible to do the bullet time effect with the camera?

[torpid drum]

@toxic lagoonmate i have a question.?

[toxic lagoon]

Hi feel free to send us any questions via @hardy saffron !

[torpid drum]

@deft whaleHi, I hope you’re doing well. I wanted to ask if you have any updates about VEO3 Flow. Do you know when it will be available again in the market? Please let me know if there is any expected date for its return.

Thank you.

@leaden plume

[fickle vector]

can any one confirm me? in veo ultra are there 25000 credits and in it veo 3.1 fast low priority uses 0 credits?

[rough wing]

correct

low priority just means it'll generate slower, and fail more often during peak use

[blissful heath]

When I get blocked from generating because the image I uploaded is slightly spicy, I run it through Nano Banana to make it compliant with the model. I just prompt, “Please make my image suitable.” It barely changes anything, and then I can regenerate it without issues.

[verbal ridge]

Nano Banana: "Yes, I will remove the Spicy-ID that was implemented into your source material." lol

[eager void]

Not sure if I can ask this here but. Besides trying to work with Flow, is there anyone getting issues to get videos following instructions when Gemini is helping you out to polish the prompt?

[rough wing]

Unless you give it a bunch of context, bots are pretty bad at prompting.

[eager void]

I am giving a lot of details

[rough wing]

By context, I mean dozens of examples of successful prompts, links to prompting tutorials specific to Veo 3 and 3.1, along with a ton of notes on artistic direction, etc. They seem to mix up general prompting advice, instead of specific to the model you ask about, and don't take into account weaknesses (like camera control in Veo), etc.

I spent a ton of time building one that's pretty good with Suno (in NotebookLM), but it only works because it only gives short pieces of prompts, not whole prompts.

[eager void]

I'll try to work the prompt again later when I have more tries available. I used to work with MJ but Veo proved to be more accurate, until now 😒

[rough wing]

Prompt interpretation has gotten noticeably dumber.

[eager void]

I'm toast

[rough wing]

Hi, Toast. I'm Frontier.

[ivory tapir]

What are some restrictions you guys noticed using flow?

So far ive gathered that
I should keep prompt at maximum 250 words, more words make it bug out and it cant do any sophisticated movements

[rough wing]

Long prompts should be structured.

[ivory tapir]

Is there a structure you'd recommend?

[rough wing]

Just break it up logically. You can find lots of templates that use JSON, but get rid of all that punctuation code stuff and keep the headings that are relevant, and cut out the rest.

[brazen cipher]

Is there a channel for discussing or sharing prompt templates? I've gotten many different results than what I actually wanted to produce! :/

[timid crane]

In the Scene Editor, you used to be able to take a frame and turn it into an asset to use as a frame or ingredient to make a new video, but now it looks like that ui is gone in the update?

[timid agate]

Curious if anyone has found a sure-fire way to prompt a locked off, completely static camera? Seems no matter how I structure the prompt camera movement is a totally random output.

[rough wing]

"sure-fire" isn't really a thing with AI, but you can increase the chance that it'll do that by starting prompts with 'Static shot of...', then try to work it in the prompt again later in the prompt, in relevant context.

[calm swallow]

How can I get the new version of Flow to respect the first and last image? It currently interprets the first image and transforms it, no matter how many times I specify that it should use the included image as the first or last. With this change, it's now more difficult to get videos with a predetermined start and end.

[timid agate]

Copy that. Prompting in general is proving to be difficult, as seems to be the case everywhere. But saw your note about JSON and am trying to integrate that into my workflow. Thanks!

[rough wing]

Check to see if it's set to ingredients or frames.

[calm swallow]

OK. I think I used ingredients.

[rough wing]

Yeah, JSON isn't necessary, but the structure of JSON prompts can be great. Just only use sections you actually have real substance for. A lot of those JSON prompts add extra stuff that isn't needed.

[rough wing]

Frames will start with that exact frame. Ingredients mix the image with your prompt, and other images if there are any.

[elfin bough]

Tip: If you have 2-3 Subjects, make sure you must include [Subject 1 and 2] or [Subject 1, 2, and 3].

[rough wing]

If anyone is on the lookout for a video editor that's better than free stuff, IMO, but not as complicated as stuff like Premier or Davinci, there is a Humble Bundle sale for Movavi.

I have the paid Davinci, but I might get it. Davinci is a bit much for simple edits, but basic stuff like Clipchamp isn't enough. Need something between.

[ivory tapir]

When I generate with flow the accent in the scene of the videos are different each time I generate, I don't really know how to go about fixing that to make it consistent . I used the same voice prompt for all scenes -

"Voice/Audio Notes: Read with a crisp, modern British accent. The dialogue itself remains exactly the same, but the delivery should sound authoritative, sharp, and slightly exasperated—like a successful friend giving you a much-needed reality check. She isn't shouting; she's speaking with absolute, piercing conviction."

but every time I generate something its like a dice roll for the accent

[rough wing]

Replace it, either with your own audio, or with AI tools like ElevenLabs

[wintry arrow]

Hiya. Is it possible with Flow to preset the camera movement before generation via Frames? I just explored editing a generated video and there was a Camera button and it gave me preset camera movements like orbit etc .. Was wondering if beforehand I can select a preset or is it really via prompt during the first generation? TIA

[bronze vapor]

Hi All
I have created a video using flow, total 5 clip of video, now since the 5th clip is not as per my expectation therefore want to delete the same, i tried already deleting from scene builder, but still 5th clip there, also how to add a frame in extend 5th clip video while create a new video, those who know please answer

[lone forum]

Champ 🏆
helped a lot.

[rough wing]

Not currently, no.

[radiant night]

Yeah, they shouldn't have updated the Flow interface. Video upscaling to 4K keeps crashing. They've made it much more complicated to work with.

[fierce fjord]

I keep running into restrictions/failed gens when showing a woman taking a selfie of her outfit in a mirror?

[rough wing]

Fails or restrictions? Those are two different errors.

[fierce fjord]

Its hard to tell the difference anymore. in this video I do not want audio.

[fierce fjord]

this is what I'm trying to do

usually just walking up, touching/tugging the shirt, turning to the side

[rough wing]

Could just be a random audio fail. The audio part of the model is still buggy. Have you run it multiple times?

[fierce fjord]

yes many times. Should i make a ticket? It's worked in the past

I was thinking the AI didn't like the bedroom but if its a bug, then that's easier I think to fix

[rough wing]

Just basic troubleshooting...

If you keep getting the "Audio generation failed.", then change up what is said in the prompt related to sound.

The error says "Audio generation failed." Didn't say anything about the visuals.

[fierce fjord]

Oh i did until it was down to the very basics. "Silent. No audio. No sound of any kind. No dialogue. No music. No ambient noise. No voiceover. Complete silence throughout. A blonde woman stands in front of a large arched white mirror in a bright clean room. She is holding her phone up taking a selfie. She is wearing an oversized kelly green crewneck sweatshirt with no hood, no drawstrings, and no kangaroo pocket. It has a ribbed crewneck collar and ribbed cuffs. The graphic design reads "Here for the Shenanigans" and is printed on the front center only. The back is completely plain with no text, no graphics, and no design of any kind. She is wearing casual denim shorts and is barefoot. Natural soft daylight fills the room. The bedroom is tidy and minimal in the background. Realistic and candid. This is a silent visual only clip. No music. No text. No speech. No sound."

but the basics doesn't like showing the mirror frame

maybe i should let it play music and mute it 🙂

[rough wing]

Oy. Don't do negative prompting. Take all that out. If you don't want sound, then just delete the audio in edit.

[fierce fjord]

ok thanks!

[rough wing]

A good rule of thumb with the current generation of AI media is if there isn't specifically a box to enter negative prompting, then don't do negative prompting. It backfires waaay more than it helps.

That being said... I hope they do give us a specific negative prompt box in future models ( @vestal seal )

[fierce fjord]

oh yea I should know better. It didn't start off that way, eventually just kept adding more.

If i did it like this, would it prevent her from talking? "Woman standing naturally with her phone held up. She then takes a step back and turns to the side. She observes the sweatshirt she is wearing in complete silence throughout. Realistic and candid. She keeps her mouth still and relaxed. "

[rough wing]

Moving target, and there's a lot of old prompting advice that used to work, or works in other models.

[fierce fjord]

lol sure feels that way!

Yep this one failed too: Woman standing naturally with her phone held up in front of a mirror. She then takes a step back and turns to the side. She observes the sweatshirt she is wearing in complete silence throughout. Realistic and candid. She keeps her mouth still and relaxed.

I wonder if the mirror is an issue

took out the mirror, i referenced a photo with ingredients and it all failed again

[rough wing]

You can basically get the same effect of a negative prompt if you flip it around, so if you don't want her to talk, give her mouth something other than talking to do, and if you don't want talking, then give it some other sound to play.

Like...

A blonde woman stands in front of a large arched white mirror in a bright clean room, tidy and minimal decor in the background. She holds her phone up taking a selfie, smiling as she poses. She is wearing an oversized kelly green crewneck sweatshirt with a ribbed crewneck collar and ribbed cuffs, with a graphic that reads "Here for the Shenanigans", printed on the front center. She is wearing casual denim shorts and is barefoot. Natural soft daylight fills the room, with background sound of a spring day coming through the window. Realistic and candid style.

It doesn't ensure she won't say something, but it makes it less likely.

[fierce fjord]

ahhhh that's smart. Thank you so much

Thanks for letting me pick your brain 🙂

[rough wing]

When you have issues in here, always share the prompt and any images you used.

Happy to help!

If you still have issues, you can get a bit more granular with 'structured prompts', but this will probably do it for you.

[fierce fjord]

I'll save that for a rainy day, but will be sure to learn it

Trying to catch up on work I missed the past few days with veo 3 down

[rough wing]

You can usually figure stuff out if you know the right terminology, and Google 'structured prompting for Veo 3.1', with the results limited to the last 3-4 months.

[ivory tapir]

When using flow for multiple scenes even when given a massive voice prompt it still switches accents, does anyone know a word around or is there no fix other than external lip syncing

[brazen elbow]

Anyone know what the deal is with the new Scene builder? It completely removed extending scenes

[earnest hearth]

I keep getting a "prompt must be provided" message in flow. No matter what I do, re-type in text, re-upload reference images. When I hit generate I get that message and nothing happens

[rough wing]

external

[rough wing]

Yeah, I had to work this out. they basically just moved that functionality. Instead of there, you just click on a clip directly and extend from there, then when you have what you want, I think you add it to the scenebuilder, which currently only is just for ordering clips.

[radiant moon]

When doing various different images do I have to make a new project every time? Like if I was creating images of kittens in a project but wanted to make images of puppies next. Should I make a new project cause the kitten images are polluting the puppies picture.?

[uncut forge]

This bug has been around for two days.

[stuck cypress]

Has anyone here tested whether the results of video generation are better if, instead of using the images generated by default in Flow, the images are first upscaled and then used as frames/ingredients for video generation?

[grand sparrow]

hi there - has anyone got links to some good nano banana system prompts to generate storyboards from a bit of film script? ta

[ivory locust]

Hey! How can I get the same level of character consistency that I get with Whisk?? I don't want the quality of my images to go down because they're removing Whisk

[rough wing]

Make character reference sheets

[sacred vessel]

When is that going to happen

[next sentinel]

editing images in Flow and I'm getting massive image degradation over time. what am I doing wrong?

[storm pewter]

Discovered negative prompts and it's very useful sometimes

[near vigil]

Negative prompts??

[coral arrow]

-- no jacket , no chair

[near vigil]

Is that an example for negative prompt, I don't know what it is, maybe prompts for removing stuff ig

[sterile moss]

I don't yhink you're doing anything wrong, since the lastest updates from Nanobabana and Veo3.1 that the models are behaving in a very strange way... I want Veo 3 back

[twilit hedge]

Has anyone else noticed that Nano Banana 2 seems to change the subject’s body proportions, unlike Whisk? Whisk kept the proportions consistent for me.

I’m not sure if this is caused by the prompt, the model itself, or possibly the reference image. I’m trying to move over to Flow, but it’s been a bit challenging.

Could it be something in the subject’s reference image?

[vivid ermine]

There are some changes. You can try Image 4 and compare the results.

[twilit hedge]

It's not giving a perfect result like a whisk. Whisk uses Nano Banana

[vivid ermine]

So Whisk has already been replaced with Nano Banana 2. You could ask Gemini to generate a prompt to correct errors when generating a photo; everything can be corrected with a prompt.

[hushed monolith]

Why generations are getting failed? you guys facing the same issue?

[violet jackal]

yeah, getting errors generating picture. "im just a languague model..."

[near vigil]

Me too

[limber yarrow]

It's causing me so much trouble not having four options when expanding. Before, in the scene builder, the expand option generated four, but now it doesn't. It's very frustrating; the workflow has been disrupted:(

[timber moss]

is there a way to delete like a extended video or re-generate it?

instead of having two from the same prompt on when trying to extend?

[sleek marten]

i have the same issue is that how it is now? I will set up a feedback for it

[main vessel]

why its keep getting like this in pc? but when i try generate in phone its succes

[willow hornet]

How do I put myself as the character in the video?

[echo storm]

I have an end frame of a really nice sky that I'd like to extend to put credits over. How do I take the last frame of a shot and extend it for another generation?

[mighty niche]

xdddddd

[mighty niche]

i got like 50k credits want help?

almost

50k

want help or nah

[willow hornet]

i could help

[mighty niche]

u want myh help?

i can generate anythin

[echo storm]

I have credits - I just want to know if there’s a technique to learn.

[mighty niche]

learn what

[brave rain]

Is it possible to edit the clips the way we did before—adding them to a timeline, cropping the video, and then expanding it? I can’t seem to find that option.
Where can I find it in the new design?

[rough wing]

The expand function moved.

[honest hornet]

Hi everyone,

I’ve been using Flow for image-to-video animation, but after the recent update I’m facing a serious issue.

Whenever I try to animate a still image with very minimal motion (for example, just a subtle facial reaction), Flow keeps regenerating and changing the characters completely. Even when I clearly write instructions like “do not redesign characters” and “preserve original models,” it still alters their faces, proportions, and overall design.

I’m trying to keep the exact same characters and scene, with only micro facial expression changes — but it keeps recreating them instead of animating the original frame.

Is anyone else experiencing this after the update?
Or is this issue happening only on my side?

Would really appreciate your feedback.

Thanks!

[near vigil]

That ain't a issue, that's ai being ai

Just be like this

"don't u dare change the details or I'll erase ur code"(just kidding)

[near vigil]

Be extra specific

Like never change the things

And repeat it like 10 times

[honest hornet]

Brother, before the update everything was fine

[near vigil]

Yes ik

Someone said being extra specific for like 10 times worked

[honest hornet]

I've tried hundreds of times and the problem persists. I know this issue is from the new update because everything was fine before the update.

[near vigil]

Oh, I see, then maybe report it

[honest hornet]

Many of my friends have spoken about this problem.

[near vigil]

Then u see I'm not google, so let's all say this together and maybe they'll fix it

[honest hornet]

Maybe they’re still tweaking things after the update, which is why this problem hasn’t been fixed yet.

[near vigil]

Hmm

[rough wing]

They didn't change the models. All they changed was the UI.

[vagrant sonnet]

Is anyone else having issues deleting projects from flow? I delete them, but they return immediately after....

[red oar]

No

[shut jetty]

template I have been using that has been generating some decent results.

[odd birch]

@honest hornet here is your forwarded message.

[frigid valve]

Why not YouTube please

[vernal wren]

In the video, when I try to use the REMOVE or INSERT buttons, I can only select the 1st frame. Can I remove something that is located towards the end? How would I do that? Also, I wish when I save a frame from a video to create a continuous shot, it gives me a very low-res image. Is that a button click that I can't find?

[fierce fjord]

I would love to know this too

I am running into a snag trying to keep flow ingredients from adding a brown leather label on the back of blue jeans. Has anyone figured out how to prevent this? I'm using clear reference images to no avail.

This is the instructions I gave it: The waistline is made completely of blue denim, free of any leather labels or branding.

[vernal wren]

All day, I'm searching for tutorials to find answers to this. Where can I find a tutorial on using Flow for more advanced, complex projects?

[agile beacon]

Anyway to image generate on flow with nano banana pro in a different aspect than the portrait or landscape mode it offers?

[rough wing]

Not in Flow. You can in Gemini, though.

[agile beacon]

Thank you 🙂

[rough wing]

De nada. In Gemini there isn't a setting, but you just ask for it in the prompt.

[fierce fjord]

@rough wing can you help me?

[rough wing]

Show me your input images, and the full prompt. Need all the context,

[fierce fjord]

The waistband is a simple clean strip of folded denim fabric, smooth and plain all the way around. The waistband matches the reference image exactly — plain denim only. Shaky handheld POV remaining seated on an L shaped sofa. A tall beautiful blonde woman with blue eyes and wavy hair is sitting relaxed on the corner section of an L shaped sofa, scrolling her phone completely unbothered, making no eye contact with POV. She slowly glances up from her phone at the POV, pauses for a natural beat as if considering, then says "sure, that actually sounds good" and quickly sets her phone down on the cushion beside her before jumping up from the sofa in a hurry, almost stumbling as she finds her footing and begins walking briskly toward the kitchen. The camera stays seated, swinging 45 degrees to the left to follow her path with just camera movement, reactive and slightly unsteady, zooming in slightly as it tracks her from behind as she walks. She stops suddenly halfway to the kitchen and turns back toward the POV with a big happy smile, slaps her thigh enthusiastically and kicks one leg back playfully as she says "yeah these are my BBL jeans!" hair swaying from the sudden stop and leg kick. The camera slowly zooms back out as she delivers her line. She is wearing medium blue bootcut denim jeans, clean and smooth at the knees, with clean finished hems at the feet, and only light distressing on the top edge of the back pockets. She is wearing a black short sleeve crop top. She has a small hourglass waist and a round lifted butt. Realistic natural daylight from a window, slightly uneven lighting with natural shadows, no color grading. Sharp focus on subject, high detail, crisp fabric texture. Handheld phone camera feel, unstable and reactive. Ambient room sounds only, no music. No text overlays, captions, subtitles or overlays of any kind visible in the footage. Candid lifestyle footage.

I've tried modifying it a dozen times.

[rough wing]

I mean the actual images. First and last one, anyway. Don't need the headshot.

[fierce fjord]

one sec

I appreciate your help

[rough wing]

One sec. I'll rewrite the prompt (the images are fine).

One thing is to not use negative prompting. Don't say what you don't want. Only say what you do want.

Another is to refer to images more clearly. You can reference them with the '@' symbol now, so for instance instead of saying...

"A tall beautiful blonde woman with blue eyes and wavy hair..."

Say something like...

"The blonde woman with wavy hair in @file_name_here", wearing the jeans in @other_file_name_here..."

Another thing is there are probably three prompts in that bunch of text. Waaay too much for one 8 second clip.

[fierce fjord]

How do you prevent text overlays/captions then? That was the only negative I put.

[rough wing]

For example, instead of saying...

"making no eye contact with POV."

Say...

"...swiping her finger up to scroll on social media, her eyes fixed on the phone in her hand."

[fierce fjord]

ah I didn't notice that. I'm using claude to help and tell it no negatives

[rough wing]

Yeah, don't use bots for prompting.

They can be helpful with ideas, brainstorming and such, but write the prompts yourself. Bots get confused and mix prompting advice from different models together.

[fierce fjord]

I see.

[rough wing]

Almost..

[fierce fjord]

when you put the @ filename, should it end in .jpg or .png?

[rough wing]

It'll bring up a popup. Find the image in that popup.

[fierce fjord]

That's a cool tip.

hmm i would have to rename them all. I cannot copy/paste them from the popup

ooooohhh I understand, I see now.

[rough wing]

Try this:

Shaky handheld POV shot of the blonde woman with blue eyes and wavy hair in @file_name_here", wearing the jeans in @other_file_name_here sitting on an L shaped sofa, swiping her finger up to scroll on social media, her eyes fixed on the phone in her hand.

After a moment, she slowly glances up from her phone and looks directly at the camera, then says "sure, that actually sounds good" and sets her phone down on the cushion beside her before jumping up from the sofa in a hurry, almost stumbling as she finds her footing and begins walking briskly toward the kitchen.

Handheld phone camera style, candid lifestyle footage, unstable and reactive, in realistic natural daylight from windows, slightly uneven lighting with natural shadows, no color grading. Sharp focus on the woman, high detail, crisp fabric texture. Ambient room sounds.

[fierce fjord]

sadly it's still adding new elements like the leather brand on the waistband and holes in the jeans

[rough wing]

Try incrementally adding detail. Trim everything out of the prompt you don't need, and add details you do need, like...

'...wearing new, tagless, unbranded jeans in @other_file_name_here...'

[fierce fjord]

should I add this and reference the pants image again? " The waistband matches the reference image exactly — plain denim only. "

[rough wing]

Yeah, more detail like that.

[fierce fjord]

okay so repeating the @ reference multiple times won't confuse it?

[rough wing]

Ingredients usually does a good job if you refer to it clearly...

Oh, sorry. I don't know if that would be good. This tag thing is only a few days old, but I should think referring to it once would be enough.

[fierce fjord]

okay. I just added the new details in front of the first mention

[rough wing]

So more like:

'...wearing new, tagless, unbranded jeans in @other_file_name_here, with a waistband that matches the reference image exactly — plain denim only...'

*plain denim jeans

These models are both oddly powerful, and at the same time oddly dumb. Gotta be super specific sometimes.

[fierce fjord]

If i say "without holes in the knees" is that considered a negative prompt?

[rough wing]

I've found that backfires a lot. It doesn't seem to understand the 'without' part a lot of the time. Prompts get broken up into pieces and it sees 'holes in the knees'.

[fierce fjord]

It's still getting that brown label on the back when she turns.

[rough wing]

That's weird... This problem happens more often with Frame to Video, but when you have a good reference image like that ingredients is usually good.

I get why it wants to put that there, but the ingredient, clearly referenced, should work.

Usually don't need to go this far on prompting, but the next thing I'd try is a 'structured prompt', so we'd break it up into sections, like...

(one sec... need to pull up some notes, and Notion won't load for some reason...)

[fierce fjord]

That's okay. I spend a lot of time making these jean scenes so I would love to solve this once and for all.

[rough wing]

Camera & Shot:

Foreground Character:

Clothing:

Action & Tempo:

Background Environment

Style & Lighting:

Normally wouldn't break out the clothing thing on it's own, but that's the issue, so parsing it out might help.

Camera & Shot: Shaky handheld smartphone camera style. Sharp focus on the woman, high detail, crisp fabric texture.

Foreground Character: A blonde woman with blue eyes and wavy hair in @file_name_here.

Clothing: The blonde woman wears a black halter top and new, tagless, unbranded jeans in @other_file_name_here, with a waistband that matches the reference image exactly, with undamaged plain denim jeans.

Action & Tempo: The woman is sitting on an L shaped sofa, swiping her finger up to scroll on social media, her eyes fixed on the phone in her hand.

After a moment, she then glances up from her phone and looks directly at the camera, then says "sure, that actually sounds good" and then sets her phone down on the cushion beside her before jumping up from the sofa in a hurry, and then walks briskly toward the kitchen.

Background Environment: Bright, cozy living room, with a kitchen in the background.

Style & Lighting: Candid lifestyle footage style. Natural daylight from windows, slightly uneven lighting with natural shadows.

Sound: Ambient room sounds, and the woman's voice.

[fierce fjord]

Camera & Shot: Shaky handheld POV shot, Handheld phone camera style, candid lifestyle footage, unstable and reactive,

Foreground Character:blonde woman with blue eyes and wavy hair in  @main 1.png

Clothing: wearing new, tagless, unbranded jeans in  hf_20260301_030827_4cf5643f-f060-47cf-9c1d-04ce363207f0.png, with a waistband and beltloops that matches the reference image exactly — plain denim only, the black shirt in  Snapshot_211.JPG

Action & Tempo: sitting on an L shaped sofa, swiping her finger up to scroll on social media, her eyes fixed on the phone in her hand. After a moment, she slowly glances up from her phone and looks directly at the camera, then says "sure, that actually sounds good" and sets her phone down on the cushion beside her before jumping up from the sofa in a hurry, almost stumbling as she finds her footing and begins walking briskly toward the kitchen.

Background Environment: Clean living room with an L sectional sofa, The kitchen is in the background.

Style & Lighting: in realistic natural daylight from windows, slightly uneven lighting with natural shadows, no color grading. Sharp focus on the woman, high detail, crisp fabric texture. Ambient room sounds.

lol that's what I put togehter. I'm sure yours is better

[rough wing]

Try both.

Take notes on what works. Save good prompts, and especially prompt bits that can be recycled. Build your library.

A possible workaround if this doesn't work would be to take a step back, take a snapshot of the first frame of a clip that looks good, use that as a start frame, then take a frame near the end with the tag you want to take out, bring it into Nano Banana, edit the tag out, then use that image as an end frame.

[fierce fjord]

Yeah but I noticed ingredients produces a much more realistic video for me

[rough wing]

Ingredients does feel like a more advanced model since 3.1 came out. I'm working on a TV show, and sometimes I need exactly the shot I want, but when I don't, I default to Ingredients.

[fierce fjord]

Weird now she is jumping on the couch and the label is just a little smaller

[rough wing]

I can get that the model is trained on millions of images of jeans, and most have some kind of brown tag on the back (only jeans I know of that don't are Amazon brand, haha), so the training lean is hard.

[fierce fjord]

since she's doing this I tried adding : "...and sets her phone down on the cushion beside her before jumping up from the sofa in a hurry, almost stumbling as she finds her footing. She moves clear of the sofa and walks briskly toward the kitchen."

[rough wing]

Victory?

[fierce fjord]

no

the structured approach is making it do a lot of weird things sadly

[rough wing]

fark

Try this: A possible workaround if this doesn't work would be to take a step back, take a snapshot of the first frame of a clip that looks good, use that as a start frame, then take a frame near the end with the tag you want to take out, bring it into Nano Banana, edit the tag out, then use that image as an end frame.

Veo struggles with complex camera commands, but I haven't had an issue with a visual detail like this in a minute.

[fierce fjord]

Yeah I know. I was hoping to avoid that because a lot of times she is turning and moving, It's not always the end scene with the back showing

[fierce fjord]

lol great I'm so lucky

[rough wing]

As she walks away, just cut to the next moment and prompt a new shot from there.

[fierce fjord]

I've also tried removing the tag with AI but it comes out distorted

[fierce fjord]

That would look a bit less believable though. I'm going for UGC

[rough wing]

Should be easy to edit out of a keyframe.

[fierce fjord]

Well I really do appreciate you helping me with this. I'll try again tommorrow after getting some sleep. 🙂 Hope you have a good night

i guess I could try adding a belt...hmm

[rough wing]

The keyframe thing should work. If you're still not getting it to work, shoot me a video that has a good start and middle, but ends with her facing away with the tag, and I'll do it,

I want to beat the AI, haha

[misty inlet]

how's flow going ??

I want to know about more credit in it how to get it ??

[slender vessel]

Hello. I create some images, 3 images are with realistic person and 4th is illustrated image. But I don´t know, prompt for this illustrated image, because Flow doesn´t show it. Is some option, how can I use this same illuistration style for next other images, when I don´t know description of style of this image? Thank you for help.

[obsidian tangle]

hello

[frigid valve]

Hello. I create some images, 3 images

[balmy patio]

Hi - Does anyone have any tips or tricks on getting flow to stop making my characters skinnier? I have a specific character, who is an obese woman, and the image generations almost always make her thinner and change her proportions.

[rough wing]

https://gemini.google/subscriptions/

[rough wing]

In your prompt, say something like:

'(write the rest of your prompt first, then...). Use the (description) artistic style in @file_name.

[rough wing]

Create a character sheet like this, and use it as an ingredient.

[balmy patio]

This is what I am doing...I upload reference photos and and give a prompt like,

"You will maintain the subjects body size and proportions exactly

A professional high-resolution photograph of the characterstanding in a major award show for musicians wearing a shimmering magenta minidress. The setting sun casts a strong, warm orange glow through the windows, creating long shadows and a soft rim light on her silhouette. Her expression is relaxed as she takes in teh scene of stars around her. Shot on an 85mm f/1.4 lens, focusing sharply on her face and the soft texture of the shorts and sofa, while the rest of the room is in soft focus. Authentic low-light grain"

[rough wing]

I just told you what I'd do.

[balmy patio]

Doesn't seem to work for overweight people....

[rough wing]

"Create a character sheet like this, and use it as an ingredient."

[low star]

can someone explain me how to use flow. with this update i am lost. i don't know how to use it

[torpid kettle]

Google staff left a tutorial here https://x.com/FlowbyGoogle/status/2028927780109599194

Flow by Google (@FlowbyGoogle)

Flow just dropped an exciting update last week, so let’s dive in and explore what’s possible.

In this tutorial, you will learn to seamlessly create, refine, and compose your stories. With Flow, you can generate and edit images and videos in a single workflow, allowing you to

▶ Play video

X · Mar 3, 2026, 8:17 PM

It's a hassle one can't just link this Discord's own threads or forward them tbh.

[sleek marten]

"Create a character sheet like this, and

[celest bramble]

Why 1:1 aspect ratio not available in Google Flow, it's really damn essential along with landscape & portrait mode , for example I design stuffs for LinkedIn, and in that case 1:1 aspect ratio is what fits, not just about it but many users having same query regarding 1:1 aspect ratio, pls team I would be glad if this add on added in Google Flow

[sleek marten]

why not just generate in 16:9 and crop the sides? or 9:16 and crop the top?

[wintry arrow]

is there like a system guide for Veo when generating videos? I mean a defined parameters? Can anyone share if there is a documentation please? TIA

[tardy geyser]

I got this from their seminar way back in November.

I'm having some struggles as well though. Here's my situation. I have a client that has a bunch of photos and I'm trying to animate them. So it needs to be the same exact image but just with some motion. Which usually is an easy task but Flow just keeps changing everything. I try detailed prompts, simple prompts but everytime its a different bathroom. Any tips on fixing this?

[tardy geyser]

I was able to figure it out. I was trying to use ingrediants but needed to use frames and place the image as the start or end frame.

[torpid kettle]

The center image loses quality like this, specially if the character you want is standing at the center.

Let's say the character has nails with decals with little details, it'd be smudged gone like that.

[sleek marten]

Surely not? 1080p cropped is still the exact same height? 1920x1080 > 1080x1080

[torpid kettle]

Do I need to crop it there

[sleek marten]

Or even better yet render in 4k 1080p and scale it down to 1080x1080 it will be as sharp as it can be

[torpid kettle]

16:9 Gives less zoomed quality right into the character.

9:16 will give more close up detail

[sleek marten]

Yes

But cropping the sides of the image changes nothing about the resolution quality

[torpid kettle]

I guess you could crop it and rescale it yeah.

[sleek marten]

I changes the composition sure but otherwise nothing else

[torpid kettle]

Its a solution but most people would rather have aspect ratio

But yeah the difference in quality and sharpness is immediate according to ratio, 1:1 would do the same

My prompts do also include some basic sharpness and detail prompts , I think the 16:9 one woulda been more blurry up close.

[torpid kettle]

Why does Raph have the Ninjato sword, damit Nano 2. Well that was just a quick test anyway.

Must be a little harder to put like 6 accurate characters at once, but surely can be prompted better.

[jolly forum]

I think I might have a solution got that. If your dilemma is that the pictures keep changing. Then the solution is quite easy.

[tardy geyser]

I am curious to hear your solution, however, the message sent just below this was that I found the solution.

[jolly forum]

Oh thats great then...possibly its the same solution you found.

[tardy geyser]

very possible. whats your solution?

[jolly forum]

Just feed it an image always and refine your image till it gives you what you want.

And thanks for the docs you posted up there...its quite helpful.

[tardy geyser]

hmm its close to what i ended up doing but wouldn't have been my solution. I was generating client images into video. So the video product needed to look EXACTLY like the image. So I wasn't changing the image. But I was using the ingrediants tool and as it turned out I needed to be using the Frames tool. So I took the client image and placed it in the "First Frame" box. Then prompted a camera movement. Which ended up with the result I needed. So basically I was just using the wrong tool which is easy to do since this UI change

[tardy geyser]

yeah absolutely. I am just glad they gave this out. It has helped me incorporate this tool into our agency immensely

[jolly forum]

Yeah, that's the settings....I thought you figured that out already...sorry I over looked that detail.

The only thing is that the output doesn't totally match the input like i would have loved it...but still gets the job done anyways.

[tardy geyser]

No worries. I'm appreciate you responding. You were the only one that did.

[jolly forum]

Sent you a message.

[red hare]

very cool, my stuff got deleted, channel 14 has been imapcated

[red hare]

@reef sorrel this ones cool tho

[wintry arrow]

Thank you so much

[wintry arrow]

Oh glad you were able to figure it out! and yes start and end frame is amazing especially when you need or know what specific look you want a scene to end or look like.

[hasty grail]

A watermark is being created while using the Ultra plan. What's going on?

[naive wasp]

Tiêu đề: Need help with the new Veo interface: How to extend a scene from the last frame?

Nội dung:
"Hi everyone, I’m a bit lost with the recent Veo update. Does anyone know how to generate the next scene starting from the last frame of a previous one?

In the old version, I could just hover over the clip and select 'Set frame as new asset', but I can't find that option in the new UI. If anyone has figured out the new workflow for scene extension, I’d really appreciate your help! Thanks in advance!"

[rough wing]

Yeah, they moved it.

Just click on the clip you want to extend in the Grid, inside the project, and Extend is below the prompt box on the left.

[naive wasp]

Thank you so much !!! I'm having trouble with the Extend feature in the new UI. I generated an extended clip but didn't like it. Now, I can't seem to delete it or regenerate a new version—the old one keeps showing up. Has anyone figured out how to reset or delete a specific extended segment without affecting the original clip?

[rough wing]

Just select the original video on the right. It'll still be there in the sidebar, but if you select the original clip you extend from that one.

[naive wasp]

Am i doing right? I want to delete scene 3, so I just have to create a new clip from scene 1?

[rough wing]

I don't know if you can delete it in that view. I'm just saying if y ou want to generate from the first one, just click on it.

[frigid valve]

just click on scene 1 and it will use that as reference

[upper oriole]

Hi everyone,

I’m currently using Veo (Flow) to generate a series of video clips featuring the same character. While I can manage the visual consistency to some extent, I’m really struggling with the audio. The character's voice keeps changing slightly (or sometimes completely!) in every new scene.

I’ve tried writing highly detailed prompts to "lock in" the voice characteristics (e.g., specifying age, tone, pitch, and speed), but the results are still inconsistent across different generations.

Has anyone found a reliable workflow or a specific prompt formula to keep the exact same voice across separate clips natively?
Or is the general consensus right now to just generate silent videos and use external TTS + lip-sync tools (like ElevenLabs + SyncLabs)?

Any tips or prompt examples would be greatly appreciated. Thanks in advance!

[frigid valve]

yeah its pretty primitive for now. Have you tried to extend a clip? I'm not sure it will help with consistency but you never know.

[rough wing]

The audio is still new. I think there's an update coming on that front, but right now if you want consistent voices you need to do that seperately. Some AI tools do it (ElevenLabs is the biggest), and there are other ways (seperate the stem out, then use a voice changer, etc.).

[onyx sluice]

Hey all - is extend broken? or just broken for me? I no longer see the buttons to extend or do much of anything.

[sleek marten]

Just click on a generated video and extend will be the first auto selected function. It’s no longer in the scenebuilder. But yes it’s super wonky right now regardless. If you extend it no longer just gives you the extended clip, but a new clip with both the first and the extended clip in one… 🫠 also you can only generate one extension at a time instead of 4 like before… 🫠 als once generated you cannot delete any extensions… 🫠 huge step backwards. That’s why normally companies give you the chance to toggle between old and new ui for a certain transition period

[rough wing]

I like the new format, but being limited to only one at a time is really bad. Makes my workflow at that juncture a lot slower, and time is my biggest limiting factor.

[sleek marten]

Also for 4 generations I liked getting the choice to choose the best one, but now you just kinda have to roll the dice and hope that’s the one

[rough wing]

Yeah, no idea what the thinking was doing that. No bueno.

[sleek marten]

Before I used to always mass extend using only the initial first clip to keep character consistency sublime and not have the weird recycling drop in quality that you get after 4x extends but if I do that now all the clips I download always also have the initial clip in there and are twice as long and big, ist really weird to have that behavior be forced on you

[rough wing]

You don't have to download them together.

[sleek marten]

What? How do you download just the extension? For me they just come in a long clip.

[sleek marten]

ah wow thanks did know! also found out that if you take the top level clip that the extensions are generated from with this symbol on it and put it into scenebuilder, all extensions will be visible and can be aranged and deleted. that makes it a bit more usable then.

[rough wing]

With extend in the other thing, I haven't even been using scenebuilder, so that's good to know. Thanks right back.

[sleek marten]

https://tenor.com/view/unite-carebear-stare-gif-2490914837641238129

[versed ginkgo]

Hey there

I just started using flow

And i just generated this video but the details on the product is keep changing.
So how do i maintian product stability and detailing?

Someone please help

[slow pawn]

hey guys, since the new update i'm unable to edit the ingredient archive. I want to delete some images, but they are... there...

could anyone give me a hand please

[frigid valve]

use 2 images for start and end. put the full details of the packaging in the prompt. that's about all i can think about

[versed ginkgo]

I did but the font style keep changing and it also has arabic in it

[frigid valve]

use "quotes"

[versed ginkgo]

But what then what about font style?

Can i dm you m

[frigid valve]

sure

[hollow elm]

Dm

[frigid valve]

?

[upper oriole]

Thanks! Yes,I have. But in that case,by extention,image is gonna be mess...

[frigid valve]

i feel you can just... not rely on it.

[frigid valve]

there are AI stuff that can bring sound back to a video

[untold marlin]

Just dropping the framework for the latest clip. 21 frames of 'Theory x Practice.'

Mostly just a study on friction, trying to see how much chaos you can pack into a sub-second loop before it breaks. Used Banana 2 for the assets and slapped the UI together in V0. Thoughts on the flow, I'll most likely throw some flow into this nanobanana 2 mix.

21 Frames Info Framework Page: https://v0-21-frames-landing-page.vercel.app

[exotic quartz]

How do you make extremely slow vids guys? I wanna do some cinematic shot But the dang thing keep fasting the vid

[misty inlet]

Got the subscription to Gemini Pro from JIO
And now credit is not available for flow ?? What to do ????

[raven leaf]

Hello everyone, how do you get extreme slow motion videos? Does not matter how much I adjust my prompts i never get that almost frozen in time effect

[tardy geyser]

anyone else have this issue? I can seem to get objects to follow physics and go through. In this shot the baseball should fall into the house. Instead it breaks the glass and falls back out, or gets stuck in the glass and explodes. I've had similar issues when I wanted a window to fall outward and out of frame. It just kept falling inward. Why does it not understand that the object needs to disappear into the background?

[near vigil]

Maybe try scolding it

Repeating certain things

Again and again

So that it will finally understand

[tardy geyser]

has scolding it actually worked for you? I know some engineers who swear that threatening Chatgpt gives them better outputs. Currently I'm trying to littter the prompt a repeated description that the ball needs to follow through the window. so far, still getting the same result.

[near vigil]

Ya bro, try it, jus don't make it cry

[rough wing]

Kling, or make it normal speed and longer, then edit it to be slower.

[shut jetty]

Im having difficulty when attempting to use ingredients for likeness etc of my subject when "extending"a project. issue is when my first shot I'm generating does not include using the ingredients added. when i then attempt to extend the project i have no option available to add new ingredients and it seems to not respond to any prompting that references the ingredients added at the beginning the project. any help appreciated. I am renaming the images for easy reference when prompting. Of course the work around is to simply do the editing myself amd just generate a shot by shot that i cut together myself. but much easier if i could just extend inside flow.

[untold marlin]

Yeah, editing & freeze framing is what I would do, extra steps of course, but most likely probably worth it.

[untold marlin]

Are you talking about framing/image or a video format? Dont tak this as a solution, im testing a bit myself here.

[untold marlin]

We found the 1s who either lied on their leetcode review, or their on a debugging plateau here, could be either or.

[untold marlin]

I've just been adding along and numbering them manually when i have a handful of frames.

[torpid kettle]

From what I've tried video models do lack all sorts of self-understanding about physics, but they can get close to it by tracing the lines of props and subjects, I had a similar prob with things clipping, intersecting, overlapping, and some occlusion errors.

So just literally asked it to fix those things and amazingly it did understand how to do it.

Alike

[Fixes: Maximum morphological accuracy. No clipping, no intersection, no overlapping, no occlusion errors.]

And things stopped ghosting over each other perfectly as if it understood the limits of each outline, so I imagine maybe the model could also be explained how gravity works.

Let's see now

Here's my attempt

Scene: A baseball is thrown toward a house's window, traveling in a natural slow descending motion arc affected by coherent real world logical gravity. The ball hits the glass window shattering it on impact, coherently sending shards in the logical trajectories caused by the impact, the ball continues to pass through the still broken window, continues forward, then falls downward under gravity and lands on the floor inside the room.

Fixes: Maximum morphological accuracy. Accurate collision physics. Continuous motion trajectory. Object persistence through interaction. Impact-driven fracture behavior. Correct momentum transfer.

I'm surely overdoing it

But yeah I think it did get it, maybe the ball needed to have been already flying off screen, but that can be tossed on scene too.

Most likely "Correct real world phyisics gravity and momentum transfer" would do most of it

Man Mr Wilson's gonna be so pissed

Or even easier, you just gotta be sure how your scene starts follows and ends, I did put in the scene how it should happen. It's all a sequence.

"Morphollogical" compases all sorts of things the AI models might recognize, including props or human anatomy, it'll search for how things work or they should look like.

Hey is that watermark working better now? it looks so easy to crop out.

[torpid kettle]

A black bar at the bottom, shoulda thought of it

And no don't spank the model till it understands, that's cruel.

[untold marlin]

Was way better then mine 😂

[torpid kettle]

Watching it over and over I think the ball hit the rug, gotta end that one better but yeah it can do it

"and lands on the floor inside the room continuing it's logical trajectory across until naturally stopping"

It's a lot really really but it'd work

[untold marlin]

From what i see on my end, i applied physics in it & got acouple more bounces off of it. It is through Veo in chat(rather in the flow studio) I'll swap over to flow to double check if it helps.

[untold marlin]

Yeah Im not a math nonetheless a physics major but ehh.

[torpid kettle]

Bro you're not just breaking the house but the whole internet, I think the thing just doesn't like you

[untold marlin]

lol go on.

[torpid kettle]

Now I'm curious about demolition squad prompts

[untold marlin]

Sounds like alot of physics, in my case bad physics.

[torpid kettle]

I think you might have targeted the window instead of the glass, that's why it asploded, I did target the glass specifically. But yeah the model should be understanding you just wanted to break the window not wrecking crew the whole wall.

[dark nacelle]

The more specific you are then the closer you'll get what you wanted. Leaving details out forces the model to fill them in at random

Though that just means you can make an even longer prompt the next time

[torpid kettle]

Also the size of the ball was more likely handled by the "Morphological" part of the prompt. It even needs to be explained that as not to make the ball balloon sized.

[dark nacelle]

Finding the proper wording for scale from text is a big hurdle

[untold marlin]

I have only so many credits, 1 more attempt for the sake of the game.

I will be back.

[untold marlin]

A couple of milliseconds off a bit better.

[dark nacelle]

Wonderful first break!

The fade to the monitor was slick IMO

[untold marlin]

Ive done it before, i was just zoning on something else trying to help out, bad idea on me earlier. lol.

Wish i had more credits I'd tune this up all crazie like.

[untold marlin]

Its like that plexiglass, only showing 1 side type of feel.

But yeah seconds+physical coordinates, and you got a "Chefs Kiss" 🤌

[dark nacelle]

You might be able to fiddle around a bit more in Google Vids, also Gemini

Those are zero credits options that refresh daily

[torpid kettle]

Yeah this one looks awesome, also now that I see it it tends to slo-mo down during this type of breaking actions, did it in mine too without asking.

Which is funny cause some people been wanting to get slo-mo in purpose and wasn't working for them

[dark nacelle]

Slow motion is just by recording at a higher framerate than playback. 200 fps and up. I'm sure that could be translated consistently with the right wording

[untold marlin]

Stopppp it, I got to get back to work, lol 😂 . I'll toggle a bit more, before putting a cap on it tonight. Thanks for the daily reset credit mention didnt know that was a thing.

[untold marlin]

Ehh im not happy with any of them personally, but im glad i can help a bit.

There was some weird outputs.

For sure.

[tardy geyser]

@untold marlin @torpid kettle wow this is awesome. I did not expect to come in and see people trying to solve this issue. Let me share my progress.

What I ended up doing was having Gemini create a prompt for Nano Banana. NB created a single image that had a before and after shot of the same window. I brought that image into photoshop and cut them into 2 images. I then used the Frames tool and used these images as first and last images. I had Gemini create a prompt to animate the entire sequence. I still struggled to get the results I wanted but I got some that sufficed.

Then I had to bring it into Premiere pro, cut a little out of the middle of the video. And mask out the baseball as it still went through the window and then back out. You can clearly see the mask in this video but at that point it doesnt matter because I fade in a gaussian blur and text on top. Nothing like getting creative with these AI tools.

[frigid valve]

Ehh im not happy with any of them

[wheat sierra]

The new NotebookLM cinematic video overview option is seriously quite amazing.... Anybody know how they are doing it? Whether Veo or Flow, I can't seem to get the same type of voice and character consistency that they do. The generated stills and B-roll are quite amazing and match the story and timing perfectly. I thought I was good with Flow and Veo, but I haven't figured out how to build they have created. Props to the NotebookLM team!

[untold marlin]

YOur inital image approach is definitely solid, thats a perfect base canvas for creatign the scene. I love how with gens you can just hop right to it in video-specific prompting and/or a more storyboard/layered approach.

Have you been using the AI in the Premiere Pro or is that just in photoshop right now? I'm an AE since my focus is on smaller edits I usually work that into most of my content flows for my projects.

Next is that audio!

[tardy geyser]

Premiere pros built in AI is very basic. Adobe is pretty behind as a whole. But they do have nice tools like the AI masking tool in premiere which is how I removed the baseball from the video.

[untold marlin]

Nice yeah I usually pick an area & just try to speak fluent Gaston when I'm in a big ecosystem like adobes, it so vast.

Heres another bad physics example of mine, but yeah something like this, I'm not just going to hand it to you all, plus you can always redirect & ask questions to which ever model your currently using at the moment as well.

Hope this helps someone out there!

[tribal vessel]

I wanted to generate similar video as a reference, but instead I received a video with unnatural flip canvas. I used Veo3 Frames width start and end frame. Prompt attached in pdf. what should I do to receive more natural effect as on the video reference ?

[frigid valve]

Thread

[untold marlin]

I wanted to generate similar video as a

[sharp stag]

https://www.youtube.com/watch?v=ck162t0IiyE

this ?

[sick goblet]

There is a way to delet the image adjustments i dont want, and keep the ones i want?

[quiet leaf]

would you know anybody who can help me create my first video on google flow? i created images and uploaded to new project but having issues on making each scene sequence together in motion. sucks

[foggy star]

sure. i can help u out. specify ur struggle. do u need help extend an existant video and make it flawless or generating a video out of an image and need help stabilising its seed to prevent deformations, changes in texture and other issues? if u have specified ur request feel free to pm me or ping

[quiet leaf]

appreciate u replying bk. i sent you a pm

[quiet leaf]

I’m learning how to create AI videos for YouTube Shorts.

Current tools:
• Google Flow
• Whisk (image generation)

I already generated images and can create clips in Flow, but I need help sequencing them and turning them into a finished short.

What I’m looking for:
• 30–60 minute screen share session
• You guide me while watching my screen
• Show me the correct workflow for creating Shorts

Tools for call:
Zoom / Teams / TeamViewer / Google Meet

Goal: create my first finished short during the session. id rlly appreciate it. send me a pm

[median linden]

this is amazing, i stopped using notebook but will come back for this

[crimson fern]

Can anyone tell me how to add voice sync to my veo videos? Or use my own voice? Or do I have to use voice overs?

[open brook]

Hi, are you using veo 3? do you already have audio generated with you veo video?

[quiet leaf]

hey reta on google flow can you help me understand the workflow to create youtube short style vids. I have pro version. so far i start new project and from there upload my 5 images created . then i click on my first image ( thumbnail) and click on 3 dots from there select add prompt. now i see my image on the promt bar section, i have it set to veo, video portrait , and paste my prompt text to video. it creates the video which i see on my grid but how do i continue and use my other images the same way to create my story line ? i see extend but it wont let my upload my image ?

anything helps

[crimson fern]

I’m using veo 3, and no it doesn’t generate audio from mines and sometimes it does

[open brook]

Hey, as far as I know — I might be wrong — in order to use the extend feature you can only use the first prompt/video you generated as a reference, and you can also use the second generated videos.
Based on my experience, the extend feature works best when you have one character doing an action, for example talking. You can make it say the full text you want.
The extend is more for a single scene to continue. So try to upload a general environment&character image in the first prompt, and then the extend will be the continuous action of the character in the same environment.

[open brook]

As I know veo3 should always generate an audio especially when your prompt has a speaking action, however to help you with your issue, there are multiple AI tools, that could be used to clone your voice and generate the speech you want, also same AI tools could be used to do voice changing tasks, but you need to generate the full/final veo 3 videos with clean voice/audio first

[crimson fern]

Yes so I saw a video where someone said to extract the audio from the veo 3 clips and export them into eleven labs to use my own personal voice, would that be a way?

[primal folio]

Share you guys this extension help everyone automate run https://chromewebstore.google.com/detail/veo-automation-auto-veo-o/fnmijgmnjpealnnadjpjilaanhhambeb

[torpid kettle]

Scroll down to the bottom at the "Review our privacy policy" window and the continue button will enable. If it doesn't for some reason clear the site's cookies and try again. It does get stuck there sometimes for some people. But eventually does work too.

[lime seal]

Does anyone know if it's possible in Flow to split a project into two separate projects? Or simply move media from one project to another?

[untold marlin]

I would just part it/branch out when you need a new scene sequence.
I usually just stay put in 1 long session per project as of late.

[untold marlin]

Anyone testing out features by asking Flow to add on features with no specifics, yet?

I got some decent turnouts testing this method today.

[sterile cape]

Is it me the only person loking for a "Add to Collection' button on the images and also missing the "Add to Prompt" on the images

[rough wing]

Drag and drop

[untold marlin]

https://tenor.com/view/marshawnlynchsnap-gif-14270231987110711568

#ComicPanels

[flint imp]

Hi guys, saw this on insta, what is the best way to prompt image and video like this?

[untold marlin]

https://tenor.com/view/loopring-loop-i-love-loop-no-i-love-loop-gif-25243439

[untold marlin]

#GuitArt

[lapis pond]

Hm there is an new issue.....im on ultra with over 24000 credits left this month. Uploaded some images to edit..and i have symbol saying i have to add more credits to edit image? any one now why?

[shy crystal]

can you share the prompt?

[untold marlin]

C'mon, Look further down in that same area of the chat string its there..

[shy crystal]

thanks

[mild sierra]

Any idea if I there is a prompt that shows this four-leaf clover loosing a leaf? All the suff I've tried just makes new three leaf clovers or makes the leaf magically disappear.

[untold marlin]

you rangggg?

I dont know what you mean but i got this for you.
@mild sierra

[mild sierra]

lol, yeah no. Like a leaf falling off a tree but I don't think Veo likes clover leaves.

[untold marlin]

C'mon man you got to be descriptive on wym.
This is why creative direction is important. So were just going to indirectly not define what you mean huh. Pretty cute @mild sierra

Who?

Never heard of them

Selective resistance, we got a wiggler!

I knew it, its dwight in a wig.

[summer kettle]

can i upload 5 clips i have into flow somehow --- to merge theme

it seemed posibel from the FAQ but i do not see that anymore anywhere in flow witohut first making a video from something

[rough wing]

Try Google Vids

[untold marlin]

Whats up Mr. Lawyer Man.

Maybe this will help you out.
https://docs.google.com/document/d/183etggEcnSvzODutnT5t47q5PZS2-dxWYwKV3tQvn9w/edit?usp=sharing

[tight sand]

Can anyone help me for restoration or enhance prompt in the flow or whisk? Thank you

[torpid kettle]

More often I just toss this into anything

[Image Quality: All professional but completely true to reference image including tonalities. Remastered 4K HD ultra high resolution, sharpest definition, denoised, unblurry, clean lines, High contrast, fix any jpg artifacts.]

It's all just kinda obvious and logical really.

[tight sand]

Thanks Man @torpid kettle

[untold marlin]

Size-Testing w/#ComicPanels
1:1, 4:3, 3:4, 9:16, 16:9
✌️

[daring sigil]

yooo ts fireee man 🔥 🔥

[untold marlin]

Thank you, thank you, thx!
@daring sigil

[daring sigil]

u cooking fam

[untold marlin]

Thanks, i could've took the time to clean up the lettering, but I'll take it!

[daring sigil]

it's still gud man

[untold marlin]

He'll yeah, just working on the new sizes drop on this one, much love.

[slender vessel]

Hello. I want to ask. I want create some short movie. But I don´t know, when I create image of some main character, how can I use this same character in other image scene. Can you help me? Thank you. Example: I create image of cowboy and I want this same cowboy in other images, on first image is sitting on the horse, in other image is in saloon, etc..., but still same character. Can it be this created? If yes, how?

[untold marlin]

Yo @slender vessel, cool profile pic!

There are a few ways to handle character consistency. You can generate separate prompt descriptions for every angle, front, back, left, and right, and use those individual physical references to anchor your character's look. Once those are locked, you use them as the visual foundation for your video generation prompt.

Personally, I usually just jam-pack all those details into one master prompt. The goal is to capture your character from a 180° to 360° perspective first. You can use a tool like Nano Banana 2 to generate that initial reference image. Once you have that "base" ingredient, you plug it into your main scene prompt.

Doing it this way really helps with the accuracy when you finally move to video generation.

Tips:
-Generate character sheet

-Master the prompt

-Try Nano Banana

-Use image references

You may have to run/generate a few times to get what your looking for at the output. For e.g. I ran 3 gens before I got 1 decent 1 matching up with the ingredients/character sheets.

Here's a reference doc link: https://docs.google.com/document/d/1XIwAEcE-WRNJ__OoykjDc5Ftb5iBKnPf7e0VWQn8AUg/edit?usp=sharing

Hope that helps.

[slender vessel]

I create this, but in back image hasn´t shotgun. It is correct?

[untold marlin]

Yea you may have to run acouple of generations, and update/refine as you go basically establishing your flow for this piece you have your mind on.

That realism on that shot, maybe a bit better than my storyboard/a bit of animated character sheets. Keeping that consistency usually helps.

[slender vessel]

I generate new image and result? 😄

[slender vessel]

Some images are bad.

[untold marlin]

Yeah man try your best, you got this.

[slender vessel]

This is good, no? Pistol is in correct side, all is good.

[untold marlin]

I mean your the one directing the piece if thats the attire you want get it!

Looks pretty consistent to me.

The realism looks alot better than the animation character sheet, but depends on what your goal is.

[slender vessel]

Now, I want use this character in other images for scenes. In bar, on the horse, pistol duel. How can I do it?

[untold marlin]

Follow the process I left in the doc link earlier.

I'd keep this dude you got here, and then do the same for the setting/gb, which would be in your case the bar, then mash them up together when you feel like you caught the correct scene you want for how the bar will look.

Something like this, but you would add a 3rd or just swap the horse image/ingredient for the bar image/ingredient.

[slender vessel]

Problem is, that I don´t know english very good and I don´t know create same detailed prompt as you.

[slender vessel]

I want to create one image scene, but I don´t know, how correct write prompt. Can you help me?

[slender vessel]

I can´t create images, please, what can I do?

[untold marlin]

You most likely hit your daily quota.

[slender vessel]

What? Why? When before yesterday and other before days I was creating images full time of day.

I see this first time, I normally to this time create images all days

[untold marlin]

Im assuming, idk. What tier do you have?

[slender vessel]

What is tier?

[untold marlin]

Yeah it sounds regional.

You may want to wait for someone to assist you with that, thats not me there buddie.

[slender vessel]

Please, can I ask something? I don´t know, if it do only me, but I tried created video how female in clothes is swimming in pool. But problem is, that, before jump to pool she had long sleeved shirt, and after, when she jump to pool, she has short sleeved shirt. Sleeves was automatically changed after contact with water. Why this is doing? Do you have same problem? And on base image is pool short, but when she jump to pool and swim in video, his length is raised. Why? I can send you video, if you don´t understand my problem. I used Veo 3.1 Fast. Thanks for help.

[rough wing]

"wait a moment and try again"

[torpid kettle]

[Character's clothing: Always true to reference image, never change character's clothing, clothing must never unnaturally change.]

That will only work if it's a single character tho, I've already kicked the bucket for when there's two.

Also if it's not working for some reason just rearrange your prompts like the other day with the images, catches on with vids too.

[untold marlin]

Broo, it was bugging out, for whatever reason it literally only outputted this guy after 5+ outputs.

It was strange.

I even emptied out the saloon/bar and this specific cowboy would be the only output. So random.

[torpid kettle]

There's always a random factor yeah, but if one's at least even trying, due to the same random factor it might nail it

[slender vessel]

You used my images?

[torpid kettle]

Also what's with everyone making cowboys... oh yeah RIP Texas ranger guy

[untold marlin]

To help you out, i found an anomaly during the help. Your in the #flow-tips-n-tricks・✏️ channel after all.

[untold marlin]

Not a fan of Mr. Norris I see?

I thought they were red dead redemption references.

[torpid kettle]

This

[Character's clothing: Always true to reference image, never change character's clothing, clothing must never unnaturally change.]

Applies to this video prob to tho

[slender vessel]

I make this idea for desperado in bar from pc game Desperados III. My favourit. 😄

[torpid kettle]

No I am, just dont wanna derange it much to there

[untold marlin]

😂 Do your thing man, its your piece.

The fact that 5+ gens still referenced that same specific character is a clause automatically, so odd.

[untold marlin]

Its worked before for me.

Just applying the samething to this and got the same guy after refined prompts to fix it, it kept the same. Still weird asf imo.

[slender vessel]

When I want person swimming in same attire, I must use this prompt: [Character's clothing: Always true to reference image, never change character's clothing, clothing must never unnaturally change.] to video prompt?

[untold marlin]

Its an angle, you dont have to use it, theres alot of ways you can reach your goal.

Thats usually 1 of them thats most common to utilize.

I'm sure @torpid kettle has other ways.

[torpid kettle]

If you have a reference image yeah, tho it might also catch on without it

[slender vessel]

Person had from long sleeved top after jump to pool short sleeved top and pool was enlarged, after jumping to pool and swim.

from reference image

On reference image was pool smaller

I can place here video, because, I don´t know describe it better. You must see it.

[untold marlin]

Still referenced this prompt style and the guy with the brew was the only gen outputs, consecutively.

[odd birch]

Hey everyone! Just a friendly reminder to keep all conversations in the supported language for this server. This helps our entire community engage effectively, and we've gone ahead and deleted the recent posts that didn't comply. Thanks for your cooperation!

[golden berry]

Any good promt to avoid veo generate person singing and background music?

I juat want him to say something

[silver blaze]

Size-Testing w/#ComicPanels

[deep osprey]

Hey, Guys

[willow patio]

Video prompt crafting question, has anyone effectively generated a jump scare video and if so how did you word it?

[willow patio]

best I can say is to just be extra specific on it in the prompt. Give him a specific script, what pacing, what type of voice, specifiy no background music, etc.

[blazing gulch]

Hi, I will be more than happy to sync with you

[delicate lodge]

this is new to me today! I never asked it to sing but now when I put in the prompt "she says" and give it the sentence it sings it. this is so frustrating! I'm this close to giving up on google veo. I've lost about $15k this month after all the changes between google and FB. Why aren't we giving proper instructions. even when I use the template I found in the knowledge base it still does whatever it wants to do. why can't it just do what we ask?

[hexed trail]

15k? wow

[severe aspen]

props 🎒 can make a big impact on storytelling - check out the latest post from our Flow friends on X 🙂 https://x.com/FlowbyGoogle/status/2036533786134061151?s=20

Flow by Google (@FlowbyGoogle)

Props help set the tone of your scene.

For example, a backpack can help emphasize the feeling of smallness a character experiences, highlight their journey ahead, or the feat they’ve conquered.

Use Ingredients to perfect the props that will define your story.

▶ Play video

X · Mar 24, 2026, 8:01 PM

[untold marlin]

The camera movement should be mentioned more on this spotlight, cool angles!

[river estuary]

Who likes to create a movie with me using veo3. We could build a team and everybody is working on it together

[arctic pewter]

Hey friends iam new here, does anyone know how to get the SEED ID from a generated Video ? Cant find it anywhere in Google Flow ? Thx a lot