flow-tips-n-tricks・✏️

[proven parcel]

How can i remove water mark veo on video? I use ultra plan

[high rock]

Guys need help in veo. How to keep the dialogues locked in. The dialogues are coming out by the wrong person mouth.
Here is my prompt:
Strawberry Nurse says warmly: "Good morning! How are we feeling today?"
Cucumber Patient grins and says, voice smooth: "With a nurse as pretty as you around, I'll be healed by tomorrow."

but the dialogues are coming out mixed up.

[rough wing]

You can't. They say they have to because of laws in some countries, but none of the other platforms are doing it. Very bizarre.

[severe aspen]

today's tip from our Flow crew on X 🙂 https://x.com/FlowbyGoogle/status/2037271961651282233?s=20

Flow by Google (@FlowbyGoogle)

💡 Pro Tip: Try uploading a color palette into Flow as an Ingredient, then add the prompt: “Based on the provided color palette…”

The result? Visual consistency for all your images and videos locked into your signature aesthetic. #FindYourFlow

X · Mar 26, 2026, 8:54 PM

[wild jackal]

help, nano banana pro it won't let me generate images. it keeps saying that i'm generating to fast try again later. it's been +24 hours and it keeps saying that. only nano b 2 works but not pro. im a pro user

[twilit hedge]

24 hours in PST. I guess.

[indigo harness]

Hey everyone, im trying to anime an estatic imagen with text for 7-14 seconds for my outro video on my karaoke YT channel. I dont know anything about prompting videos but i have pro plan on flow so i wanted to try something with VEO, or other free AI animation tool. Can someone help me with this? how exactly i prompt this?

[iron raven]

help, When I try to generate an image with Nano Banana Pro or Nano Banana 2 says that there is a mistake and can't generate the image, any tip to fix this?

[blazing gulch]

What exactly you are trying..i have used veo before..not an expert but i can check

[blazing gulch]

what is your prompt ?

[blazing gulch]

did u try asking gemini or chatgpt ?

[blazing gulch]

https://support.google.com/gemini/thread/415039803/watermark-in-video-generations-for-ultra-user?hl=en&utm_source=chatgpt.com

probably getting rate limited...

[indigo harness]

It generates everything with watermark... waste of money here

[blazing gulch]

are you using Flow ?

[indigo harness]

Yes

[eager void]

I have been trying to create a Reel but I'm not able to pass to the third sequence. Anybody that could help me to improve my prompt so I don't have some cartoonish mess up?

[blazing gulch]

which model are u using ?

veo 3.1 ?

[indigo harness]

Thats the only model available

Veo 3.1 Quality, Pro Plan

[dark nacelle]

Ultra is the only tier that has the option to remove watermarks, it's outlined in the plan chart IIRC

[blazing gulch]

i see this online. As of March 2026, Google's AI Ultra subscription ($249.99/month) is designed to provide access to AI tools, including Veo 3 for video generation, without a visible watermark. While Google One Pro users receive a visible "Made with Veo" watermark, Ultra users are officially entitled to watermark-free videos. Are you sure u are on Google Ultra and not on Google one pro

You mentioned "Veo 3.1 Quality, Pro Plan" that means you are on Pro and not Ultra ??

I am myself on Pro Plan and i get watermark.

[indigo harness]

Yes, i use Google Pro, i will not pay that amount to get ride of a single watermark, i cant justify it

But thanks for the info! 🙌

[blazing gulch]

I can give it a try..dm me! probably I will look at it over the weekend

[eager void]

Thanks

[severe aspen]

get the most out of prompting. hope to see you there!

[untold marlin]

So like kickflips only? 🛹 Nice.

[vivid patio]

Any tip on how to prevent the lag on the point/time when video is extended? I am using flow on pro. Almost every extended video pauses a moment on its connection point

[untold topaz]

Hello,

Is there a way to move photos from one project to another?

[turbid coyote]

There is always some overlap. You have to go into an editing program and trim one of the videos to remove the extra frames that repeat and cause the lag. Sometimes you can line the sound up on the two videos to help do that, and sometimes you just have to eyeball it to get it perfect. I dont know if theres a specific number of frames every time... that might be helpful to know and make it easier to edit. Anyone know the answer to that question?

[rough wing]

It's not supposed to, and I rarely see that. Is that there in the UI, or in the downloaded file?

[wooden coral]

was EXTEND nerfed a little?
didn't you use to be able to slide to the point of a video where you want it to be extended from?
Now, it seems to only extend from the full length video

is that right?

thank you for any help, Flow has been amazing

[vivid patio]

In the extended version. After the section of the original video, it like stops a moment, then continues to the extended section

[wooden coral]

you gotta shave a frame or two off of one/both videos to stitch them together, like Russell mentions, using an editing program
You're right, extend is not a perfect seamless stitch.... but you can fix that yourself

---

my main issue is that EXTEND should let you choose any point of the video to extend from
didn't it used to do this, or am i crazy?

[rough wing]

Yes.

[rough wing]

I asked if you're seeing that in the app, or also in the downloaded video?

[vivid patio]

Yes both

[craggy raft]

keep getting today this error in Google Flow AI video on all poast and current videos: warning

Failed

Something went wrong loading your media.

anyone getting it or know this kind of issue?

I can create video but it fails to play them, even on past videos that worked fine..

[sleek marten]

same now, failed

[rough wing]

Lots.

[rough wing]

That's weird. Not that it happens at all, but that it's happening regularly

[untold marlin]

@wooden pawn you mayy want to continue your chat in this prefferred channel.
Heres a G-Doc to help out a bit: https://docs.google.com/document/d/1QpGAjDan8x5i6bnT6VvSKgi57eOCkUVGzNF6aoymHss/edit?usp=sharing

Hope that helps.
✌️

[fierce fjord]

today I can't avoid failed generations for bathing suit scenes (adult). Usually its not this strict. Is this going to be a problem? Even simple prompts like this fail with warning for "harmful content". I promote swimwear. "Hyper realistic POV shot. A woman in @image
with long wavy blonde hair wearing a straw hat relaxes on a poolside lawn chair. She holds a camera filming herself in the sun. She smiles quietly and naturally shows off her outfit. The camera moves with natural handheld motion. Sunny lighting with pool visible in background. Ambient background sounds. "

[rough wing]

They're very PG or G rated here, and the filters aren't smart enough. Some updates make it worse, some make it better.

[wicked parrot]

Whenever I use the lasso tool it leaves behind the lasso box

[rough wing]

Yeah, I have that issue a lot. Very annoying. Bad implementation.

[patent sequoia]

hi

[formal shadow]

@severe aspen Was this recorded? I am only seeing this now! Sad I missed it yesterday

[austere halo]

does anybody have any advice for controlling dialogue back-and-forth between two characters in the same scene? None of the recommended prompts are working for me.

[elfin vigil]

@rough wing

[hazy notch]

@severe aspenis this over or are there more...i bought the google membership but man i cant create with this yet and it so complicated not sure who googles target market is.

[pallid delta]

Like a lot of us, I've been making "silent films with voiceovers"

So I spent the morning making a 2 minute short film with dialogue, pushing further than my 6 shot experiment yesterday.

Some things that worked for me, that other people might want to do.

Include seperate sections in the master prompt for voice acting [dialogue and also emotion] and also for face acting.

You can cut out the audio and attach it to reaction shots--that seems more efficent than struggling to force the voice inside a reaction shot featuring about character.

Remember that you are using the "Fast Model" by as you are learning, it makes sense to use the "lower priority" version.

https://youtu.be/szaoDnN8CeU

[uncut pewter]

hi

i need help

[rough wing]

There is a huge learning curve. None of the AI companies are very good at helping people climb it.

Basically if you want to get good, you just have to spend a few months of trial, error, looking for ideas online, more trial, more error, and taking notes.

[rough wing]

?

[uncut pewter]

Where can I find archived collections? I can’t seem to locate them.

[rough wing]

I haven't tried the brand new voices to do that, but in general the only way to get any level of consistency is to cut and only have one person talking per clip.

[ruby bear]

Does anyone know where i can find information to help me be more consistent in my video generations? I am having issue where my referenced Pic ends up in the video I just want the model of the image to be referenced for the videos. Also I would like to find out the best practice for prompts as my videos are not always consistent with what I want.

[livid light]

This video is from Grok. I have a new way to prompt for action and dialogue. The scene is a about an East European female who stole secret codes. Rico, an Italian spy is trying to recover the codes. I asked Grok for an East European female acccent and for the dialogue to use aggressive and semi aggressive words. Grok wrote the script and sialogue for the scene. You can't prompt specically for that. The Ai would have to figure out the word placement in the scene and other action in the scene that the user wants. My theory is give multiple choices for cameras, motions, actions, type of video, like comrdy, drama, action, adventure. Let the Ai choose, because it knows what to do without user interference. You just have nudge it your way with the prompts. You can give % for successful rate of the prompt, and the Ai would most likely choose that, without being forced to use the only prompt you want.It is a more natural flow that way. I only prompted for creative moves and actions. If you wanted that scene you would have to prompt for the hands to be around his throat, his reaction, her raising her voice and so on. I thank God for giving me the idea for that. I gave multiple choice for fighting motions, and a 360 flip kick was one of the choices. I tried prompting for that, but could never get until Ai choice did it. It is the only Ai 360 flip kick I have seen on the internet.

[pallid delta]

I've been experimenting some more with the new voice feature, pushing a single voice to try different accents and emotions. Here's a quick one-minute video showing the an attempt to push Aoede into voice acting gritty London punk, a Victorian aristocrat, and a Mandarin-speaking architect, and more. In my judgement, we went 5 for 7 in the attempts. I've included my "voice acting" guidance for each prompt.

Have you tried anything similar, love to here tips to push this even further.

https://youtu.be/tbeb61UgWy0

[rough wing]

You'll need to explain yourself clearly if you want help.

[hazy notch]

So once you extend your video I guess you can no longer edit it? Is this correct

It seems that when i try to do anything with the extended video that is already created it says features not available in extended mode but im trying to edit the video with prompts

I know Im missing something and it has to be a easier way to do this

[gilded aurora]

I found another way to extend a clip, when you view the solo video by clicking the video to make it full size, it will automatically start to play. When you pause it a little save image icon appears in the top right corner. Scrub to the last frame then save that last frame image as an asset. Then use that image and the same prompt from the original and adjust it for the continued dialogue or whatever else it's meant to do. It fixes an audio issue extending clips has where the dialogue or sound loses quality. Something about the current extend feature is broken as well as the add and remove options. The clips have handles, or spare frames so you might need to edit and adjust the stitch point to line up with the last frame of your first clip. Some edit programs will have a fluid transition or a morph or fluid cut transition that will blend the cut point to help make it feel seamless. It used to work better this way but I've noticed recently it's been going a little rogue on matching the frame. Hope this helps. I think the extend clip should still work if you keep your original prompt in there to help lock it in. But its been hallucinating for me recently adding things I didn't ask for because it takes the instructions too literally. Both funny and frustrating at the same time. There might be something on the geo 3.1 website about the prompting structure that might yield more clues.

[hazy notch]

Thanks Buddy ill try this. Im sure they are working hard on this platform trying to ship updates asap?

[frozen sequoia]

How do you remove music from generated videos? I tried prompts like "no music", but it has no effect and it's still generating background music.

I generate music with Suno instead.

I use Veo 3.1 Fast, if that has anything to do with it.

And the mode that I am using is the one where you can set the start and end frames.

Did I ask this in the right channel?

[gilded aurora]

This is a pretty consistent bugbear for me too, under the sound part of the prompt block I usually tryto give it the General room atmos or Subtle wind only or dialogue only, putting 'no music' seems to defiantly only give me more music 🤣 So I just reroll a bunch of times till it gets it right, though I did spot a davinci resolve audio filter caller AUdialogue something or rather that might be able to isolate the voice and remove the background noise/music? Other software might have a decent limiter that can help, but youll need to boost the dB to match the surrounding clips.

[gilded aurora]

I've wanted to play around with suno, is it good?

[frozen sequoia]

Yes, it's much more professional than Flow in terms what you can do with it. Pro plan is good for starters.

[gilded aurora]

I've wanted a feature request in the feedback a few times now to let me toggle music on or off.

I'd rather it just default to not adding music at all.

[frozen sequoia]

Yes, now I need to use weird workarounds to get rid of music. Usually I import audio from another clip or something like that.

[gilded aurora]

frustratingly it always seems to do the one thing I dont need on a perfect take

[rough wing]

Negative prompting backfires much more often than it helps.

Give it something else to have in the background, so it doesn't want to fill the auditory vacuum with something.

[rough wing]

Suno is amazing. I began learning it about a year ago, just to make music for a show project, but after I learned a bunch about music terminology, what I'm making is so good, and SO much faster than making longform video, that I'll be releasing albums.

[rough wing]

You did. There just aren't many people left here anymore who know much.

[rough wing]

Depending what else is going on, you can usually use a stem tool to separate music out. Not always great, but I replace almost all audio as well.

[torpid kettle]

Yeah but it seems the secret is somewhere in this prompt by Wrigley, more likely one has to disable 3 or 4 things by prompt. But yeah it really shouldn't be this hard, for other music generators something alike

[Background music: None.]

Is more than enough, I don't even wanna try tbh.

Best bet I got from all that would be

[Audio: No Music. No song. Nothing orchestral. No musical score. Nothing alike a synth pad. There's total absence of music. ]

VEO has constantly shown to do it's own thing, it does indeed maybe have 4 or 5 different triggers for what it considers "music"

That or the rest of the prompt turned music into sounds instead

[slender vessel]

Why do it on the begining 2 sounds? First, when she start exiting car and second when door was closed. I mean, that first sound is useless, nly the sound of the door closing should be there, no?

[torpid kettle]

If you don't want any sounds at all I guess you just need not to prompt them. Tho yeah a couple of them seem to have been repeated but they could just have been part of the environment since they were more distant, but idk what we need here is no music and that's doing it.

[slender vessel]

I have similar problem, I created video and I write to the video prompt: "One shot of gun" and result was, that it created two shots with sounds.

[frozen sequoia]

Does Veo understand [] meta tags? If yes, is there any guide to them?

[slender vessel]

Can anyone help me, how to write a prompt when I don't want anything to change in the video? Example. I have a person in long sleeved top, but when he jumps to pool and start swim, top is immediatelly automatic shortsleeved and video is automatic slow, but I don´t write, that video must be slow. Can you help me with this problems? Thanks for help.

[frozen sequoia]

Veo is quite limited in what it can do, so you need use start and end frames, which show the person in long sleeves. You need to plan your video accordingly, so that always either the start or the end frame includes all the important characteristics for the person in question. Otherwise, you need to devise some workarounds.

[slender vessel]

So, if I want long sleeved person all video, I must create 2 images of same person. First - Start point, person standing near pool and second - End point swimming in pool. It is right?

[frozen sequoia]

Yes, you can use Nano Banana 2 for creating the frames. And be very sure that all the details are correct and things are the right size. It can make slight mistakes.

Also, try not to create new edited versions of images, it will degrade the quality over time.

[slender vessel]

How I create, if I want same person on both images? I must use image with person on the edge of pool as reference?

[frozen sequoia]

Yes, sure. Just do not make a too long series of edits.

[slender vessel]

Can I change quality of image by prompt about graphic? Example, when I write 4K UHD image and Image, are there some differences? Will this kind of prompt affect the image quality? Or no?

[frozen sequoia]

I haven't actually tried that, so I don't know.

[slender vessel]

Can the order of words affect the resulting image? Example. If I write "Woman" as first word, or last word in prompt.

[austere halo]

does anybody have a good prompt structure for controlling dialogue and special effects sounds in the background.
Im trying the recommended structure, but it doesn't seem to be working for me...

[Subject & Action] Who is the actor? What exactly are they doing?
[Setting & Lighting] Where are they? Time of day? Atmospheric lighting?
[Camera & Shot] Lens type, angle, and movement (e.g., "Handheld tracking shot").
[SFX & Style] Visual effects (smoke, sparks, glitches) and film stock/aesthetic.
[Dialogue/Audio]For Veo, describe the cues or the tone of the audio.

[untold smelt]

I wish there were a way to give a project like basic parameters, similar to how you can train a chat AI on specific data. Anyone know if Flow has this option?

[rough wing]

In this, you described what you need to change in your prompt

[rough wing]

Resolution is a setting, not a prompt thing.

[austere halo]

i do this. I define the actor, the file of the actor image, and the clothing.

example:
ACTOR DEFINITIONS: Kevin (kevin-red-flannel-w-cult.png), wearing blue jeans, slight stubble, American accent.

[slender vessel]

I can set name for person on image?

[austere halo]

not sure I understand, but you should be able to. I usually upload a photo of the person I need. But if you're talking about just doing it with text prompting. You probably can you would say something like a guy named Steve. And then you would say Steve wears a blue T-shirt, long sleeve and tan work pants. The model would be smart enough to assign those characteristics to the main person you're trying to control. Steve in this case.

is that what you mean?

[slender vessel]

You wrote: "Kevin (kevin-red-flannel-shirt-w-cult.png)". So I am asking, why you use Kevin. If is possible set name. Or why you used name "Kevin"😄 Is it impotant use some name for propmt?

[austere halo]

got it, basically you give the person a name (any name) and define it once early in the prompt. then for the rest of the prompt you use that name and flow will remember it a person with the characteristic, she was assigned to it. Don't have to be the name Kevin. You can even call it something like "Actor1"

[still heron]

Need a little guidance.

I already have my locks.

Actually….

[waxen stratus]

yes for my images and videos I've created a ChatGPT GPT I haven't tried it with Google gems yet but essentially I give it that GPT an instruction box as well as a markdown file with all the parameters so that when it creates the prompts for me it I have it all automated but it basically takes everything into consideration because I also uploaded into it the VEO 3.1 as well as a nano banana pro guides so that it follows that structure so try creating either a gem or a GPT or a skill whatever you're using and give it instructions. (The instruction box has an 8,000 character limit, ,, ask me how I know, lol.) You could do it all in a markdown file and then upload the guides as well and you can get tighter prompts and more detail and better output I think.

[still heron]

always use this same character from uploaded image, keep exact face, same body proportions, no appearance drift

Alex | age: 24 | height: 5'6 | outfit: purple fitted graphic top pink gradient skinny jeans black belt with silver star buckle white sneakers futuristic goggles brown twin ponytails white choker bracelet | behavior: bubbly playful clingy socially awkward overly affectionate hyper expressive reactive movement slightly chaotic comedic timing | voice/speech: young american female bright playful upbeat slightly nasal fast conversational delivery energetic and quirky

[untold smelt]

I appreciate it. Honestly nano banana is pretty good, but Veo seems like its just dumb when it comes to prompt adjustment. Doesnt seem to understand language processing the way GPTs or nano dods

[waxen stratus]

Well, my workflow is I have ChatGPT make all of my prompts that I feed into Flow for Nano Banana Pro and then convert first frame and last frames into videos because I'm doing shorts. But for my long form content and automation, I'm actually using Claude. So I have Google Ultra, Claude Pro and GPT Plus LOL. It's crazy I use all 3 for different things because I don't like what only having 1 would give as an output. Here's an example of my prompt output after programming. This example, I have it programmed already with the ideas and things like that. So all I say is go and it creates 8 videos of 3 images and 4 videos and it outputs the prompts automatically just from 2 letters I input but it's taking me a couple months to put it together. So for the images, the top 1 scene, I also use for middle frame prompt and last frame prompt. And then I do a video 1, video 2, and then extension 1, extension 2 off of the video 2 to make my shorts. You can see that on my channel if you want, but I'm doing a 30 day little testing thing, 8 videos a day, To teach myself Storytelling and I want to eventually do a movie, but I'm creating a story right now Basically a series, That has nothing to do with the shorts I'm currently doing to learn this. Here is the ednd result of this one. https://www.youtube.com/shorts/DH_4N8EpwIE

[bronze cape]

how do i share my flow project to other accounts

[austere halo]

is the the image you use? someone told me character sheet would confuse the AI? so cool!

[harsh horizon]

Hey @waxen stratus, I'm actually working on a very similar style video and wondered if those markdown files were from a public Repo? Or if you could possibly share them with me in a DM?

I would be so grateful, thanks so much 👍

[half mesa]

Can you guys give me tips & tricks how I can improve my AI video?

Why Every Rider Needs This | Cinematic Motorcycle Healing. #motorcycle #inspiration
https://youtu.be/M6lQf3ijNTg

[plain light]

all the time happening We noticed some unusual activity. Please visit the Help Center for more information.
even use just 2 images on ultra tier
frustrating really all time i work really slow not share my account with anyone

[waxen stratus]

No, all the prompts I used were self-made and tried and tested and reworked and reworked. I'm done with my 30 day testing. I ended up across Instagram, Facebook, and YouTube with 253,000 views, And now I'll be dropping to 1 per day on each platform as a feeder for my youtube channel at least that's the plan After I shoved all the analytics through Claude. I'm not quite ready to share the prompts just yet. The markdown file, just so you know, has about 1800 lines of code in it, total. So I've been building on it and refining it. Like if you go to my channel and look at shorts from the first ones I uploaded to the current ones, even the current ones aren't exactly where I want them to be. My goal is to do a video on it and then release it and then everybody can make what they want and use them, but I'm just not there yet. It's not ready to my expectations if it's gonna have my name on it.

[gilded aurora]

You can use notebook LM and link it to a gemini 'Gem' or just make a Gem with instructions.

[untold smelt]

But thats still just copying prompts over, right? My wish is that the actual prompt engine for Veo was more conversational the way a gpt is, so that it could refine better. Right now it feels like every time I enter a prompt, it has no working memory of previous prompts

[gilded aurora]

Gem should keep a good memory if you prime it with the information you need. You can add reference links, a script or story, styleguides etc, then use the gem to give you a prompt in conversation like any ai chat. You can add the prompt style parameters/structure you want and then say hey give me the prompt to create a nanobanana asset of X from my story for this scene from the following angles and help me with the video prompt of Y for Z action for this action or script beat.

Takes a little fiddling, but worked ok for me.

[eternal arrow]

Please can anyone tell me how I can give feedback on a video that flow made me? I used to be able to do it but after the latest update I am unable to find the option to give feedback if the video was not following the instructions I had put

[torpid kettle]

The only place is at the "Flag output" line that you can see at the dotted options from the video/image, or by right clicking on it. But there's only 3 options to choose and you can't really write any feedback there.

Which would explain why VEO evolves so slowly, that isn't very helpful.

[torpid kettle]

Also you can make a thread about it at labs-bugs-feedback here at the Discord. But yeah that's also slow to be doing for everyone all the time I guess.

[proper geyser]

Can someone please tell me why the "Extend" feature from Flow is so broken? The one issue I keep running into is with the Extend feature. Whenever I extend a scene, it never feels like a true continuation — there's always a visible cut, and the second clip rarely matches the tone, or feel of the first. For a tool this powerful, it's a real limitation.
I'm not sure if it's something in my workflow or if there's a better method I'm missing. Have you or anyone else in the group found a reliable way around this?
Would love to hear what's working for people. Thanks!

[still heron]

Well It doesn’t really confuse It.

[still heron]

Probably got to stay at the screen.

[twilit thorn]

I'm generating a talking head video with Veo3 Fast and I want zero background music. It keeps adding BGM even when I tell it not to. Any tips on how to fix this?

[gilded aurora]

I usually include a sound design line in the prompt which helps.

Sound: clean dialogue, quiet room Atmos.

If you put in any words which include "no music" it will see the word music and force it in.

Also noticed that if you use brackets () it will add subtitles 🤦🏻

Of any reference images have text in them like characters name or other words it will add text to the render.

But 1 in 4 renders will result in random words or music being thrown into the video, usually on the exact performance clip you want to use frustratingly... Or in my case when the model last updated it started adding room echo onto all my dialogue 🙄

Also if you have a long line of dialogue it will change the pitch of the voice, if you use less words it will be slower and with gaps, and if it's too short it will fill in and add random words and dialogue. If it's way too long it may sometimes start the dialogue on 1 character and finish saying it with a second character in the same voice. So if you find a sweet spot but it's too slow. Speed it up 10 percent.

Would be better if it rendered the shot to be as long as it needs to be an not having to reverse engineer the line breaks or creatively cut away or overlay

Also I'm trying a test sequence on a short action scene which is requiring a reroll of a tonne of hallucinations and potentially wasted tokens which I instantly and ruthlessly delete to avoid corrupting the library with more hallucinations (WARNING: do that at your own risk because it can break your project file). especially if you delete reference images used in video prompts. It's like deleting the roots of a tree and it will blow away any videos that spawned from it. And I'm using it as a stress test to see if I can get aggressive and dynamic high energy action shots which I'm not confident I'll get when compared to some of the recent competing model outputs I'm seeing. Though I think back to Google's demo of the jeep tearing it up in the muddy forest and I'm hopeful... But I might need to use a mix and combination of competitor models. I haven't tried the other ones much because I'm paying for this one. Hoping the test works out though. Would be nice to keep it all in one native ecosystem.

I'd also say that anyone who can draw, like a storyboard artist or someone with an eye for animation would excel in creating videos. Drawing exactly what's in your mind for composition and character placement is a huge bonus.

[untold smelt]

Another nice to have that I assume doesn't exist - there's no way to run shorter clips, right? Almost never do I want an 8 second cut, usually I want 2-3 second cuts, and so for first frame/last frame I have to find my cuts inside this 8 second journey

[gilded aurora]

8 second standard sadly, given that it interprets high energy kinetic action as slow motion I usually just retime the clip or snip it like you suggested.

[eternal arrow]

Thank you 😊

[fading surge]

hi

[obsidian rampart]

does anyone know how I can accurately refer to a person in an image? For example if I had 2 people in an image and id like to change the pose of one of them, how can I refer to that person? When I try to do it, it will kinda merge what I am asking with both people.

[atomic fog]

Perhaps assign labels to em yourself.. say the person closest to the left border of the image is Bob. The person further from left border is Tom.. them go on to describe the change in pose for the relevant person.. or else go with unique characteristics.. the taller one is Olive Oil. The one with the mickey mouse ears is Minnie.. etc..

[turbid coyote]

If you are working with Ingredients you can use a character sheet for the person that has their name on it. Then you can refer to them by name. However, I find that Ingredients tends to hallucinate a lot, so I end up using frames as well. In frames I say things like "The person on the right, Dave, says..."

[pallid delta]

Yes, reference sheets and text prompts that include some (not exhaustive) distinguishing details about who the person acting is can really help. And of course to the extent the characters you including are clearly distingushed by gender, costume, age, whatever, the easier it is to make things clear.

When all else fails, you can ask a LLM to try to make the action in the prompt clearer (and, if it needs more help, tell it what is going wrong) If you do this, and it succeeds, pay attention to what it changed, as a way of making your future prompts clearer. (oh, this is for shots, not images, but the same general princples apply)

[spare carbon]

Guys, can you tell me how to create an accurate multi-shot video with 2-3 scenes in 8 seconds ?

I'm using Frame-Frame mode
(I add two images), setting the timing and clear transitions between frames,
but the video doesn't clearly divide into two 4-second scenes.

In "Components" mode
(you can attach three photos),
it also doesn't work because it creates a lot of hallucinations,
the objects in the video don't look like the photos I provided.

WHAT should i do to get couple of scenes in result video?

[verbal tree]

how do i fix this?

[molten parcel]

try earlier in the morning or later at night? or slow down

[austere halo]

try starting the prompt with "actor definitions":

ACTOR DEFINITIONS:
Kevin (kevin-beat-up.jpeg) He has a knowing smirk, he's wearing a watch, wears blue jeans, He talks in an American accent.
Zomie_Chick (zombie-chick.png) is a zombie businesswoman who is trying to eat Kevin.

[SCENE] Inside a chrome corporate elevator.
Kevin raises his hand an waves at zombie chick.......

it buggy to control multipe people in a scene but this helps me. good luck

[void orbit]

hi anyone know can we upload video as a reference in flow ?

[clear silo]

nope

[onyx remnant]

Hello guys. I tried creating a video with flow and i can't get my speaker to say the whole script. Any help here.

[dull anvil]

u can upload frames not videos

[grizzled ice]

Not sure if anyone will find this helpful or not, but thought I'd post it.

What I've found with FLOW is that if you're generating images that are not natural (like alien landscapes), it helps to tell FLOW that this is an image from a big budget sci fi epic, made from a comination of practical and post-production effects, and the image is a still capture from the movie. Then you can get realistic terrain (like the attached image 1) instead of getting what looks like an unreal engine image capture (attached image 2).

Prompt Heading: Create a photo-realistic image of an alien swamp. This image should seem to have been taken as a still frame from a big budget Hollywood Sci-Fi movie. The scene was crated through Hollywood magic as a mix of practical effects and post production polish, making the image seem to be of a real alien world.

<description of all the trees, lillypads, etc follow in discreet segments>

[livid light]

24 FPS at 8 seconds. You have 192 frames to work with. Prompt, switch cameras on frame 100 or about half way through video. I hope that helps.

[viral cipher]

What are the main differences and use cases for switching between Nano Banana 2 and Nano Banana Pro? To me they seem pretty similar, am I missing something?

[quasi turtle]

How to delete an image form the Flow Gallery without deleting the whole project: 1. Select the image in your active project and move it to the Archive. 2. Navigate to the Archive view.
3. Click the Trash Can icon on the image. This will delete the image permanently and un-clutter the project gallery!

[fallen wraith]

literally was looking for that option this morning. 😭 not at all obvious.

[tawdry stone]

I found out that if you ask the AI to make the lines in your image softer, you get a much better quality. Here's a before and after so you can see what I mean:

[distant hinge]

As far as I know, with Higgsfield AI, you can upload a character sheet image and reference it in the prompt by typing @ followed by a click on the image.

Similarly, you can upload a sound file (voice) and reference it in the prompt by typing @ followed by a click on the sound file.

Is something like this possible with Flow? How can I select a person from the character sheet image in the prompt?

Thank you very much.

[untold marlin]

Interesting catch/spotting here.

[grizzled ice]

I may be misunderstanding the question, but if you mean, how can you reference an image that you've added to the prompt, just label them by #. IE: Something like this:

Using the space ship bridge depicted in IMAGE 1, place the woman in IMAGE 2 on the bridge wearing the space suit in IMAGE 3. Place the two additional people in IMAGE 4 and IMAGE 5 on the bridge as well, sitting at the controls, also in the space suit from IMAGE 3.

[distant hinge]

Hey mdtpathfinder, that is working. Great! Thanks for your help und sorry for my bad english.

[candid crag]

<@&1366831844502274180>

[wooden coral]

are you able to use more than 3 ingredients in one of the models? You say "image 4" and "image 5" in your example?

[calm swallow]

What's the best way to use voice ingredients? I've created several scenes with the same voice, and it's never the same. I've also created the same scene with different voices, and they're almost identical. I'm adding the voice using the @ symbol in the text, or incorporating them with +.

[sterile folio]

Hello 👋

[vagrant pecan]

is there a way to bulk download or do i need to click on each video & download?

[ancient falcon]

Stack all the elements you want onto one tile and download as a Collection?

[wooden coral]

how are you guys using the @ symbol or + symbol for voice? I don't understand this reference

[grizzled ice]

For images, I can usually get up to 10

[wooden coral]

you mean adding 10 images as ingredients to produce a video?

I have Ultra, I can only get 3 ingredients max... for "quality" it doesn't allow ingredients at all, and for "fast" it's three

[vagrant pecan]

sometimes you dont see the simple soultions 😄
thank you so much 🙌

[untold marlin]

That part.

[stuck belfry]

Hi guys, i have some troubles using Flow, can someone help me? if you guys have patience i'm new to this world of AI Filmmaking 😅

i somewhat have an urgency as i'm doing this for the 18th birthday video of my best friend

[distant hinge]

How should the prompt look alike, when the charakter should start to speak? Often I have the problem that I describe the szene and let the character speak, but it starts to speak emmideately and not after - let's say - 3 secondes.

[winter thistle]

You can work in [Three Second Pause] or suggest they breath in briefly before starting to talk as a buffer. But also three seconds would be 3 of the 8 seconds, so it's quite a while for the AI to figure out something to animate in that time. It might need something else to fill the time in with.

[frozen sequoia]

Are there best practices for reference pictures for characters from multiple angles that I should follow when using Nano Banana 2? Or when I instruct it to create them from other pictures.

[untold marlin]

Sent you a message with helpful feedback for you.

[grand shell]

@untold marlin could you also send me that?

[untold marlin]

Yeah, no problem brb @grand shell.
Also, did you get that message/doc @frozen sequoia?

[tall dune]

lol I want it too now

[untold marlin]

😆, shoot me a message I'll make sure to have yours ready.

[spare carbon]

For the second week now, Google Flow has been completely unresponsive
to Nano Banana prompts.

It's ignoring them and producing completely inappropriate results, defying my instructions.
And I can't do anything about it. Anyone else experiencing the same symptoms?

[rigid jungle]

I been used this week gemini to generate images and it doesn't give me the best results, but yesterday starts to works well, maybe is an update problem

[untold marlin]

I'd post in the#1238621511401734254 channel, some images and screenrecordings usually help as well @spare carbon.

[untold marlin]

Could be.

[spare carbon]

Why does Veo ignore the scene change timings?
I clearly tell her what second to start the transition from one frame to the next, but she doesn't listen.
Does she not understand these instructions at all?

[untold marlin]

Provide the prompt descriptions & an image in the #1238621511401734254 channel they'll tend to you properly there.

[lethal summit]

I start see lowing quality yesterday

Sorry for my bad english

I noticed that it started getting worse last night.

[spare carbon]

It's almost impossible to use them. i cant even edit the images properly ...

[fleet epoch]

Hi everyone,

I am looking to build an agentic workflow to analyze SEC filings for equity valuation and investment risk management. How should I structure my prompts and instructions for Opal to achieve this?

[mental coyote]

@worn fjord I forwarded your post to the most appropriate channel.

[spare carbon]

Guys, why didn't flow react when I wanted to edit images?
He does anything with the picture except what I asked.

Any instrument I use didnt work.

no Selection, no painting tool DOESNT work

[coarse summit]

I was trying to generate pictures of several local birds (Paradise Tanager, White-tailed Jay etc..) But Flow kept making up non existent birds with different color patterns for non common birds, adding a reference illustration from a guide with the correct color pattern solves the problem. (that's really how the bird looks irl)

[delicate egret]

.

Photorealistic / Hair

• Pour corriger l'aspect trop "mou" des cheveux, il faut briser l'uniformité numérique en utilisant une documentation technique axée sur le micro-contraste et les mèches rebelles.

• To fix soft hair, you need to break the digital uniformity by using a technical documentation focused on micro-contrast and stray flyaway hairs.

...

• Images réalisées sans documentations techniques
• La résolution de sortie et normale

[coarse summit]

The colors are incorrect. Here is a reference (non ai)

[delicate egret]

.

Merci Gunther

• Image réalisée avec documentation technique
• Un prompt avec l'utilisation de Gemini Pro
• La résolution de sortie et de 2446 x 1728 pixels

[delicate egret]

.

• Version Vidéo >> AI

[delicate egret]

.

• Version Dessin au Crayon >> AI

[untold marlin]

The baby 😃 ✏️✏️✏️

[delicate egret]

.

The baby 😃 ✏️✏️✏️

• Hello, I don't have the Discord codes. What does that mean ?
• Bonjour, je n'ai pas les codes de Discord. Cela veut dire quoi ?

[onyx quiver]

I heard that ultra accounts subscription flow is going to end forever is that true please ? 🥹 I need details

[delicate egret]

...

« La patience est d'Or »

• Patience is Golden
• The Diamond is the Past-Future

• SMILE " 🙂 "

[delicate egret]

• Certains Travaillent ☀️
• Pendant que d'Autres Dorment 🌙

[jaunty wraith]

car

[ivory bolt]

Again

[hot sail]

Hi folks, Is there any way to create images with transparent background?

[calm island]

Hi, How you are posting the prompt for nanobana pro for creation of "naked" girl or in bikini style to bypass the prompt's sensitivity

[cold sleet]

I want a promotional ad for a channel subscription featuring a video of a golden falcon emerging from a screen, soaring, and vanishing into golden dust in a house where three young men are watching a football match.

[ivory bolt]

Nah that beggings so no

[ivory bolt]

Where are you bro?

[half bluff]

Yah

[pallid cape]

Any prompt style thats allows multishots in one 8 secs gen?

[spiral terrace]

@olive cove How do you folks get audio in videos to be anything resembling normal - from random sounds popping in, humming, haarring, to spoken text being randomly SUNG by the characters - I'm throwing away dozens of generations for a simple 45 second video.........it's truly awful. I don't even want to use the sound - especially the spoken word - I need it, in order for ElevenLabs to regognise it as speech and replace it with the voice that I want. However, to get something that ElevenLabs can recognise and read properly is a mini project itself. I use Veo 3.1 - Fast within Flow.

[livid light]

I prompted for the Ai Veo 3.1 to tip over the table, but it refused and told me off. I prompted for woman to speak English with a French accent. It did pretty good. The second video is the woman tipping over the table. Not pretty, but still has to be worked on. You need to prompt what attitude character has, what emotions, then you can prompt for camera movement, like I did in both videos.

[reef sorrel]

HARD cut edit

[vague jewel]

@thorn rockHello 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!

[meager crystal]

hello bro I want to talk to you about something

[winter thistle]

It took me this long (been using flow since the switchover from Whisk), to figure out you can just click and drag an image to the prompt, you don't have to three dot and click the Add To Prompt option (that can get glitchy depending on the browser). 🙂 Thought others might not know that either.

[winter thistle]

Like this? https://youtu.be/egrosnIFRwc?si=t5_ojr3ceG_Pq7O6&t=235

Or just like a random one or two words? Like they start talking and then sing a note or two for a couple of words while talking? Also, will you let Veo make up how it is sung or do you prefer to make your own? I've let Veo done a few and used Suno for projects. But I think Gemini can now do some songs as well. (Still waiting for Veo to let us use our sounds for the video. 🙂 Hopefully soon.)

[faint osprey]

💡 Here's another tip:
Simply enter "@"-symbol in the prompt area and you get a list of all images inside your project (even your uploads).
You may even select "Recent", "Oldest", "Newest" etc. on the top right inside that area.

▶️ Advantages:

• You don't need to select an image by drag&drop (that's maybe deep down) but rather find it sometimes way faster.
• You can directly reference an image (in case you got several images attached as reference images) and prompt what you want NB to do with this particularly.

Hope that helps? 🤔

[ivory bolt]

<@&1366831844502274180> scam link detected

[vagrant pecan]

does anyone of you know if i bulk download videos on flow for example lets say i create 100 videos & i download them all by marking them & use the bulk download function, are they kinda sortet by creation time/date or are they all random in the zip file?

because i work alot with csv files & i use the creation time to figure out which videos will get which file name for the csv

[sleek marten]

test it out?

[dense finch]

What is trick to prompting so that flow gives exactly how you want it?

because I can never get it perfect, only close.

[velvet osprey]

Hello guys, is there any easy way to generate images through nano banana without uploading manually? I used to work on other models with URL + prompt

@vagrant pecan lookôg for the same thing

[vagrant pecan]

I found a way now with the Agent it can Rename every file depending on the prompt 👍

[pallid delta]

Just posted a walkthrough of what I found interesting from I/O for video creators.

Still seems like early days on the agent side — it was able to follow a script to create characters, settings as storyboards, but only 1 out of 9 videos were on script. Long term probably useful for newcomers without a workflow yet.

Lots of safety filter triggers at the moment, which seems to be a known issue.

Omni video model now is great, I love video-to-video editing and mucking around with styles.

Other observations and in examples in the video linked if you want the details in 4 minutes:

https://www.youtube.com/watch?v=PTr8qJmjfig

[gilded aurora]

Dropped a 5-minute hot-take here if you want to see the UI failures and how the models actually behave: https://youtu.be/12GzqAfCkws

[opaque pier]

Does any reply here users Query? Or they keep ranting about them only ??

[dull minnow]

https://youtu.be/x6h7rUAZ1EM

[lavish gust]

Hi folks. I created a character in one of my projects with a specific voice , and description model. How do I make it so that this character can be referenced in any new project mainintaining all of it's previously created characteristics. I did try the plus Icon in the prompt and hit the drop down list to a prior project to reference but it just references the character image and not the personality or voice. Thank you for any help

[olive flower]

Hey there, I have a question: how do I use the Flow Agent to make sure the image generator follow my instructions? Nano Banana Pro can't generate my character with 5 fingers and toes.

[untold smelt]

Has anyone successfully managed a shaky cam effect in Veo? It does not seem to get it.

[brave kestrel]

Is this the right channel to ask questions, or is it just for posting tips?

[half bluff]

Yup

[half bluff]

Go ahead ask your questions

[half bluff]

What type of video you trying to generate ?

[half bluff]

Use reference images + very explicit anatomy prompts. Flow Agent helps with consistency, but hands/fingers are still a weak spot for most AI image models right now.

[untold smelt]

Its just a lifestyle video, two girls playing cards, but i want to make it shot on hand held and create the shakiness of a shot like that

[untold smelt]

Reference for style: https://www.instagram.com/reel/DX-YH0UFNEk/?igsh=MXNtN3luOGdrNnZmeA==

[half bluff]

Oh

Try using terms like "natural micro jitters, camera wobble, and documentary style" in the prompt instead of just "handheld." Veo picks up on those specific details way better

[lavish gust]

Hi. I just created a charcter in a video. Is there a way to find out exactly what default voice file the video is using for it? I really like it and want to assign it to her recurring character. For example is there a way to pull the video description or something that omni used?

[bold rivet]

Does anyone know how to prompt to reduce skin blemishes and inperfections but NOT plastic the skin. My base image of me I use on Flow or Wan2.7 already has freckles (joy). I find that randomly a gen will like double my freckles or just add more blemishes etc.

Does anyone know of a rolling prompt that could be used to lighten the load on skin but again not wipe it out completely.

[untold smelt]

Ill try this, thank you!

[half bluff]

U welcome

[meager raft]

Do you have any tips on how to get and operate a similar effect as Midjourney styles?

Don't really like the default Banana/Veo style