flow-tips-n-tricks・✏️

[dawn turret]

ask again I can't help you

[elfin flax]

Thank you. It seems like it would be difficult to train veo3, so I'll try taking screenshots. I didn't want to add more steps, so I wanted to avoid creating a video from image generation as much as possible. Anyway, thank you, bro! I'm Japanese, and I'm translating as I write here, so I apologize if the sentences are strange!

[odd birch]

Hello. As stated in #start-here・📍 , “Currently, our team can support engagement in English; the discussion/content in this server will largely remain in English for now.” Thank you.

[onyx sluice]

no worries - i understand. yeah, it's a bit tedious to add extra steps. Hopefully Google is developing some kind of built-in character consistency function that will make this process easier!

[vagrant pecan]

Does anyone of you have any tips to name videos correctly?

Because i create around 100+ videos a day & need them to have an exact filename.

Currently i make it all manual, so after i download i copy the name i need & change the filename. Would be great if there would be something where i can create a filename in flow directly before downloading.

[sturdy flower]

I am not sire about anh flow filename settings. But if there isn't, you could download a batch rename utility, sort your video directory by time/date ascending so it will start with the first one yo made then rest will be listed chronologically. Then its just a matter of picking your files with the rename utility, choosing a filename and how many numbers you want in the filename. If you do over 100 day 3 should suffice. Just Google something like "open source bulk rename utility" and you will find something.

[shell pewter]

does anyone know how to get more dynamic camera movements?

[reef sorrel]

You might wanna check this out
https://medium.com/google-cloud/veo-3-a-detailed-prompting-guide-867985b46018

[soft shore]

Guys, I am facing some difficulties with flow.. every trial is being denied due a "sexual violation rules" but there is NOTHING wrong with it.. Is there any expert that could be help with / or at least to understand what I need to change in order to make it work?

[reef sorrel]

Maybe show the prompt here

[hallow lance]

How to generate characters in green screen ?

[bright creek]

I dont think there's a current good reliable method for it?

Try one of ChatGPT''s Veo 3 tuned prompts as they are pretty good getting ideas across to the ai unless your already very good at prompting in veo 3

[austere scroll]

What for? To re-use them? You can always get a no background picture of your character, then place it using whisk or nano banana in any scene you want to create for a first frame

[fresh valve]

Now I take my character to nano Banana and prompt it for a green screen. It's been working great so far

[frigid stream]

I have a prompt like this to generate image to video, but it keeps getting error for violated prompts. Could someone explain how to write better?
"The female character with vivid violet dragon braid, yellow jacket, white top, denim shorts, and necklace, holds out a bouquet of roses toward the male character. He, with silver-violet hair covering his eyes, purple vest over yellow shirt, jeans, and watch, takes the bouquet and looks at it briefly. After a moment, the bouquet slips from his hand and falls softly to the ground. He turns to the right and walks out of the frame. The female character remains still, eyes blinking once, lips parting slightly, posture quiet."

[reef sorrel]

Let me see the image 👁️

[hallow lance]

@austere scroll @fresh valve Thanks you for the ideas. I will try them.

[fresh valve]

It's the 'lips parting' that might be kicking it back

[light stirrup]

I use JSON for images and videos. May i ask, how do you generate your jsons? Maybe its in the structure?

Does anyone know how to download generated images created in flow?

[rotund jolt]

I’d love to know this as well!

[limpid summit]

I think it's too long for one scene shot. Divide into two parts and I think it will work.

[echo storm]

Subject
A young woman with a vivid violet dragon braid, a yellow jacket over a white top, denim shorts, and a necklace, and a man with silver-violet hair partly covering his eyes, wearing a purple vest over a yellow shirt, jeans, and a watch.

Action
The woman extends a bouquet of roses toward the man. He accepts the bouquet, glances at it briefly, then lets it slip from his hand, falling softly to the ground. After a pause, he turns to the right and exits the frame, leaving her still. Her eyes blink once, lips part slightly, and her posture remains quiet.

Scene / Context
A softly lit urban evening setting that conveys intimacy and subdued drama.

Camera Angles
Mid-range framing with a slight upward angle on both characters.

Camera Movements
The camera begins still, gently tracks his movement as he exits, then holds on her as she remains in frame.

Lens & Optical Effects
Shallow depth of field, with background blur isolating the characters’ expressions.

Visual Style & Aesthetics
Muted color tones, subdued drama, naturalistic textures, emphasizing emotion through subtle detail.

Temporal Elements
Real-time pacing with no acceleration or slow motion.

Audio
Faint ambient city sounds underscored by a low, restrained piano note.

Negative Prompts
No text overlays, no logos, no exaggerated visual effects.

[frigid stream]

thank you

[reef sorrel]

is this kpop demon hunter character lmao

[reef sorrel]

A dramatic, cinematic 4K shot, slightly low-angle, begins with a slow zoom into the two characters standing on a rain-slicked Asian street at twilight, with neon signs glowing in the background. The female character, rendered in a 3D animation style with vivid violet dragon braid, a bright yellow jacket over a white top, denim shorts, and a necklace, extends a bouquet of red and white roses toward the male character. He, featuring silver-violet hair partially obscuring his eyes, wearing a purple vest over a yellow shirt, light blue jeans, and a watch, reaches out and deliberately takes the bouquet. He holds the roses for a beat, his gaze briefly resting on them with a stoic expression, before intentionally opening his fingers, allowing the bouquet to dramatically tumble and scatter petals on the wet asphalt with a soft, melancholic splash. With a decisive, sweeping motion, he turns sharply to the right and walks quickly away, his figure receding into the bustling, blurred street scene, leaving the female character visibly devastated, her pose frozen in rejection as the camera slowly pulls back slightly to emphasize the fallen roses.

sorry if i misinterpret the character but the descriptrion you wrote does sounds like kpop demon hunter characters 🤣

[frigid stream]

yes, you make the right character 😄

[cunning eagle]

Here's my version

A dramatic, cinematic 4K shot, slightly low-angle, begins with a slow zoom into the two characters standing on a rain-slicked Asian street at twilight, with neon signs glowing in the background. The female character, rendered in a 3D animation style with vivid violet dragon braid, a bright yellow jacket over a white top, denim shorts, and a necklace, extends a bouquet of red and white roses toward the male character. He, featuring silver-violet hair partially obscuring his eyes, wearing a purple vest over a yellow shirt, light blue jeans, and a watch, reaches out and deliberately takes the bouquet. He holds the roses for a beat, his gaze briefly resting on them with a stoic expression, before intentionally opening his fingers, allowing the bouquet to dramatically tumble and scatter petals on the wet asphalt with a soft, melancholic splash. With a decisive, sweeping motion, he turns sharply to the right and walks quickly away, his figure receding into the bustling, blurred street scene, leaving the female character visibly devastated, her pose frozen in rejection as the camera slowly pulls back slightly to emphasize the fallen roses.

[light stirrup]

https://tenor.com/view/spongebob-still-still-a-mystery-mystery-flying-dutchman-gif-13664330

[frozen lotus]

there is any trick to unlimited access

[topaz bough]

x2

[cunning island]

Hi, I'm wondering if VEO-3 can generate videos in Vietnamese? If so, can someone share an effective prompt?

[untold smelt]

Does anyone feel like this channel would be way better as a forum so you can sort by topic

[rotund kettle]

x2 meaning?

[austere scroll]

you do know where you are asking this, right? lmao

[austere scroll]

You can, basically you can write whatever as basic as this.
The man goes to the toilet and says in Vietnamese, "Oh no, the toilet paper is gone"
I think now you can replace the english words in vietnamese as well but I would always write before the part SAYS IN VIETNAMESE

[manic stirrup]

Hi everyone,
I’m looking for an official prompt guide for Google Veo 3 or any recommended documentation from experienced users. Could anyone share where I can find a reliable resource or perhaps suggest a guide that you personally found helpful?

Thank you so much for your help! 😊

[reef sorrel]

You might wanna check this out
https://medium.com/google-cloud/veo-3-a-detailed-prompting-guide-867985b46018

There’s also an official simplified veo 3 prompting guide by Google deepmind https://deepmind.google/models/veo/prompt-guide/

And another official veo prompt guide by Google cloud Vertex https://cloud.google.com/vertex-ai/generative-ai/docs/video/video-gen-prompt-guide

[reef sorrel]

Personally I find this one very helpful and useful

[cold quiver]

I'm new to flow The problem I'm having is. there.. are gliches when I create I've followed the prompts but I can't fix things like my character is on a phone call they'lll finish the call and the phone will just be akwardly in it's hand or i had a character light a cigar and the flame never touched the cigar and it was in his hand then ihe put his hand to his face with another cigar with the cigar alreday in his mouth how do you fix these gliches?

[manic stirrup]

Thank you so much for your help！！！
This information is really useful and exactly what I was looking for. I appreciate it! 😊

[copper raptor]

just keep generating

[dawn totem]

Hello friends. Good day. I'm new here. I need help, I want to make a video of myself meditating, and I want to make a light effect, but I'm not there yet. Yesterday I did a test in Flow, I added the ingredients, a photo of me already characterized meditating in the lotus position, and another image of the scenery, I placed the JSON prompt, and I get an error that says: "This feature is not compatible with Portrait mode", at the beginning the prompt mentioned that it was in (9:16), I corrected it to (16:9), in Flow I also changed the option to horizontal and cropped the images in such a way that they were also in (16:9) without having to crop the image when adding the ingredients, and the error still appears. I had to join the two images with the help of Gemini, both of the character and the setting, and even then the video with the ingredients could not be created. The only way I could create the video was with the "frame to video" option. If there is someone here who can help me create the video with the ingredients, or explain to me why this error occurred so I can correct it, I would greatly appreciate it.

[topaz bough]

Me too

[clear dagger]

top notch!

[grave pewter]

Good afternoon!

I'm Brazilian and I'm creating viral videos on Flow. The videos are already good, but I want to add a female voice to the video. How do I do that?

Can I send the prompt here?

[main swan]

old man facing the sea, receiving the sea breeze, ship in the background waving its sails

[mental coyote]

Hello, friendly reminder images or videos cannot be generated in the server! Please do not post your prompts in the channels as it will not generate, unless you are asking for advice or tips. Thank you!

[wary verge]

https://youtu.be/txn7EOp8uVY?si=-U_j9riCXC-KNzZ7

im trying to create videos like this with Veo but using anime characters like Future Trunks from Dragon Ball Z for example. Anybody got advice/tips on how to prompt it to do that? I’ve been mostly using Frames to Video since I don’t believe I’ll be able to achieve what I’m trying to do with Text to Video

I’ve been using this pic as my reference picture

[wary verge]

I’m also not worried about the audio generation, I plan on inputting my own audio

[spice glade]

I am mobile right now, but what I would do is take the photo poured it into Gemini and Gemini to help you construct a prompt for veo in json format.

It is difficult to provide suggestions if you do not provide the prompt that you used.

But from what I’ve tested on current limitations, you might need to use multiple ingredients and not frames, but you will have to deal with the fact that you cannot get from the ingredients or recipe

But it depends on your goal. Are you trying to convert the people to look in that style or that style to look more realistic. For that you need to blend a new image if you’re going to use veo three to start from a frame

[wary verge]

{ "input_image": "TrunksRef4Video3.jpg", "description": "Cinematic close-up of Trunks rapping in a neon alley.", "motion_drivers": { "seedance_face": { "intensity": "High", "synchronization": "Audio_Driven", "expression": "Intense, aggressive delivery, lip_sync priority." }, "seedance_body": { "intensity": "Medium-High", "rhythm": "Sync_to_Beat", "primary_motion": "Sharp, rhythmic head bobbing (forward/back).", "secondary_motion": "Subtle shoulder bounce." } }, "camera_control": { "style": "Handheld", "shake_intensity": "Low", "zoom": { "type": "Rhythmic_Pulse_In", // Subtle zoom synchronized to the beat "intensity": "Medium" } }, "effects": { "atmosphere": "Enhance foreground smoke drift, maintain neon glow." } }

So I got pretty close using the JSON method, if you see the screen recording I posted earlier I wanted more of that aesthetic but this still came out pretty clean. Def want more of a “hood” vibe tho but maybe I need to generate an image with more of that specific detail

0:00 — Stable shot. Trunks stares intensely at the camera. Subtle breathing movement. Camera begins a very slow push-in (zoom). 0:01 — The beat drops. Aggressive rapping begins (lip-sync). Sharp, rhythmic head bobbing starts. 0:04 — Intensity increases. Facial expression becomes fierce, eyebrows furrowed. Camera adds a subtle handheld shake. 0:07 — Peak performance. Trunks leans slightly closer to the microphone for emphasis. High energy delivery.

That was made using the timeline method & the starting image is the thumbnail essentially so I figure if I generate an image with more of the specific details I’m looking for (characters in the background, wider camera angle, hood/ghetto setting, etc…) then I should be able to get the video I want

[sterile berry]

I am nameing my videos in Flow, but when I want to find them I can't. Can anyone tell me the best way to name your video's to keep yourself organized?

[solemn igloo]

Hi everyone. I'm having trouble getting camera movement in Flow. I've been using the VEO 2 model because the VEO 3 model does not support video extension and I have had zero luck. The camera will just not move no matter what command I enter (dolly in, tracking shot etc.) Has Anyone had any experience with this? Have any tips?

[arctic condor]

It's been a month since my first flow generation on September 6th, but the free monthly credits haven't refilled yet. Am I doing something wrong?

[lapis fjord]

Veo 3 does support video extension unless it is a portrait. Try scenebuilder.

[solemn igloo]

Yes, I am using scenebuilder but it reverts every video extension to veo 2 fast

[grand sparrow]

Hello - Im having a persistent issue creating a shot in Flow using ingredients to video - i have a wide shot with a character in it and want another character to enter the shot. Im giving veo the still image of the wide and then separately the full length image of the secondary character. the issue im having is it keeps changing the wardrobe of the secondary character - is there a way to pin the prompt to the ingredient image ive given it? thanks

[keen brook]

Try describing what thr chsrater is wearing in detail

[cunning eagle]

any workflow tips on generating a full end-to-end ai video

like what tools for voiceover, images, character ideation, creation, consistency, action and stuff like that

most of these are all possible with veo 3 in some capacity, but what if we need more?

[trail canopy]

I use Nanobanana or Seadream to create characters and settings. You can also use Nanobanana in Flow but it's kinda janky and you can't save your edits off site. I normalize all of the dialogue using Elevenlabs V2V so it's consistent between clips.

[feral dew]

Animate this reference image in a cinematic, mythic tone. A divine forest drenched in fog and golden sunlight. Rama stands firm, drawing his glowing arrow — Kabandha’s elongated, root-like arms stretch toward him through the mist. At the first second, the camera pushes in slowly; air vibrates with divine tension. As Rama releases the arrow, a burst of golden light streaks through fog, hitting one of Kabandha’s reaching arms — no gore, only light and energy dispersing. Second arrow follows instantly, crossing the frame like lightning. Shockwaves of divine wind ripple through the trees; leaves and particles scatter in all directions. Kabandha’s arms recoil in blinding light — his chest glow dims slightly, showing the impact. Camera trembles with echoing thunder, then the light fades. By the final second, the forest stills — only drifting dust and silence remain.

[upbeat wharf]

Probably been answered a million times, but if I want character consistency throughout my scenes, how do I go about generating the correct prompts. I want to create a 30-40 second video, but I want the characters to stitch seamlessly from scene to scene.

[vagrant pecan]

Does anyone know why some of my AI-generated videos have black bars on the sides or top and bottom, instead of using the full frame? It happens randomly. Any tips or fixes would be really appreciated!

[upbeat wharf]

Why doesn't Ingredients to video support Veo3 - This is a real pain. When trying to create something that is in line with my brand and using reference images to make sure its accurate, I then cant use veo3 for audio. 😭 😭

[trail canopy]

What is your prompt?

[vagrant pecan]

It doesn’t seem to be directly related to the prompt itself, since I can use the exact same prompt multiple times — for example, 10 times in a row — or even use hundreds of different prompts, and the black border appears randomly.

[edgy moth]

@vagrant pecan Skip terms like "cinematic" in your text prompt, as they increase the chance of bars appearing.

@vagrant pecan Also start your prompt with: "Avoid letterboxing" repeat "avoid letterboxing" 2–3 times throughout the prompt, and end with it. Specify the desired ratio clearly "Full 16:9 widescreen video with no black bars, edge-to-edge framing."

[golden sluice]

I have what I believe is a simple Frames to Video request but keep getting weird output. What am I doing wrong? 3 frames are just overlays building off the first where the 2nd just adds 6 dotted boxes, last is the reveal. Any suggestions? I can add the prompt and even the output video. Thanks.

[oblique hemlock]

For the timestamp trick, does it work less well if -- before I set the first timestamp -- set up the main factors of the scene that will stay the same throughout?

for example:

Static camera: Tress and grass blowing with the wind, an eerie sound droning in the background

timestamp: "00:00-00:03"
blablabla

timestamp: "00:03-00:08"
kakakaka

[worldly iris]

Gotta use whisk, or Veo3 on gemini web chat

[lime gate]

Whew ok I put this initially in Flow General 😂 😭 Any tips or tricks for words guys for Flow? 👀

[warm stone]

guys how can i make the camera fix in veo 3 fast by prompting??

[merry jackal]

Hi everyone, can anyone advise how to set up a static camera for Veo 3? So that the camera doesn't move, there is no zoom...

[hoary dock]

locked camera distance, locked camera movement, locked camera zoom, try these

[stark walrus]

Hi everyone, wondering if Ingredients to video is currently down on flow? I upload the reference photo, type the prompt and click the arrow and it isn't proceeding

[nocturne light]

Static Shot also

[cunning eagle]

the best way to ensure character consistency is to have that character in the start frame of every scene

[cunning eagle]

it will be supported soon

[onyx sluice]

you could try it but that might confuse it. it's generally better to avoid describing any actions or activities outside of the timestamps when using that method. You can add detailed descriptions of settings, props, character descriptions, etc. though, and that works.

[cunning eagle]

ok flow 3.1 is not even following basic commands like not to play music in the background, what is this?

[sullen spindle]

its having issues atm

[cunning eagle]

yep and its a pain it had to come up now

i'm generating videos for an ad idea and its treating my instructions as suggestions lol

[sullen spindle]

same, im in the middle of a music video which i need to get done for friday! lol

[chrome grotto]

it added random comments from the prompt, I don't know why now they read the prompt out loud in the video

[stark sphinx]

What we can and can't put on flow ingredients?

[oblique hemlock]

"ingredients to video" does not allow adding audio ingredients?

[random plinth]

Generations stuck at 99%. Anyone else dealing with this?

[onyx gull]

not anyone, but everyone 😉

[random plinth]

O i see. so its a known issue

[mental coyote]

Fwd! @haughty atlas

[candid furnace]

anyone have some tips & tricks to make the video less jumpy/hyperactive

[frail torrent]

Which channel should I use the bot in to convert an image into a video?

Please im new

[severe aspen]

hey there- no content will generate within this Discord server. you'll need to head to our website, labs.google and pick a tool to start generating with. we encourage folks to then share their outputs across this server in dedicated channels.

for generating images, try Whisk or ImageFX! LMK if you have any questions.

[normal star]

Does anyone have a good starter prompt? Like the one to exact the scene, dialogue and more so the AI doesn't just do random stuff?

[thick flume]

Appreciate users experiences creating realistic collision physics. IE. A billiard ball hitting other balls. A sample prompt would be great for a ball hitting another ball at a specified angle then deflecting at the correct angle to hit a third ball

[smoky cosmos]

What should I add to my prompt to make the scene feel more classic? I’m trying to get that old-movie vibe but not sure what elements to focus on

[devout pine]

Any idea how to keep voice consistent. I use the same prompt for voice, yet get a different one everytime.

[viral spade]

Just had this pop up in my feed. Maybe it will help? Haven't had a chance to read through it yet. https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-veo-3-1

[stark sphinx]

we cant use human faces in ingredients?

[candid furnace]

i feel like that's a pretty reasonable safeguard

[stark sphinx]

But doesn't make a lot of sense since we can use realistic faces on img2vid

[candid furnace]

lol yea the inconsistency is pretty dumb

[odd birch]

Hello, we wanted to remind you of our server guidelines, stated in #start-here・📍 : “Currently, our team can support engagement in English; the discussion in this server will largely remain in English for now.” Please keep in mind that non-English discussions and content may be removed. Thank you.

[thorny pecan]

Hey guys i would like to know if some of you could guide me for something. I'm making some cartoon with ChatGpt and VEO3. I run it with Json and i have to say that sometimes it's not good at all for keep some details like caracter's look. I'm really not pro so probably my Json are broken but.. i don't if somebody have same problem, i'll be glad to discuss about it or receive some help. Thx 🙂

[keen lichen]

Hi everyone. I'm trying Flow for the first time. I'm not a filmmaker or anything like that. I tried to recreate a scene from an animated movie in live action version and I realize I didn't know how to sync lips speed (I used an external audio track). Here's the clip I made. I don't know how to set the light I want either. I tried to edit the first part of the scene, but I didn't get any changes. Any tips?

[rich hound]

Hello... Does anyone know what prompt to add to my text-to-video so that VEO3.1 generates correct faces in crowds?

[soft ice]

You have to mention this specifically in every prompt I guess

Have to put certain instructions to not put irregular faces

[soft ice]

Scene looks solid tbh

[azure condor]

I used to be able to easily generate consistent character voice with flow in my generations but I am not able to generate consistent voice anymore. The generations just randomly generate different accents idk why

I even try to describe the audio but that doesn’t work. Even with the expand feature I am not getting consistent audio. Can anyone please help. I know audio is beta but if there is any possible solution. I’d really appreciate the help

[sleek marten]

I feel like this should go in the bugs/feedback channel?

[abstract dagger]

how can we make extended video as they annouced in veo 3.1

[burnt star]

I am trying to make a video, but the main issue is how to keep the characters consistently . It is hard to keep the same face from the whole video.

[sleek marten]

I am having the same issue with consistency and have posted it in the bugs and feedback section

[sleek marten]

you move a scene into the scene builder, click on the plus icon and select extend.

[severe aspen]

fwd!

[rough adder]

Try using JSon script in the prompts, what better way for AI to understand but a programming language. json
{
"prompt": "A blue-haired man riding a horse through a ghost town in 1880s New Mexico.",
"subject": {
"man": {
"hair_color": "blue",
"action": "riding a horse"
},
"horse": {
"color": "striped yellow and blue",
"action": "galloping"
}
},
"setting": {
"location": "ghost town",
"time_period": "1880s",
"geography": "New Mexico"
},
"art_style": "photorealistic",
"lighting": "harsh desert sunlight",
"composition": "dynamic action shot, medium distance"
}

[compact wren]

How do I prompt for orbit shots in Veo? I prompt the camera to orbit around a car or orbit around a fight scene, but im getting a stationary camera movement or the only camera movement I get is dolly in or dolly out.

Whenever I want the camera to start with an over-the-shoulder shot and then orbit around to reveal a subject's face, the subject either morphs or turns around to face the camera.

[iron ice]

Try "clockwise arc shot"

[hasty oriole]

I found that music or annoying sounds went away when I was more concise in my prompt and verbose, to the point where i'm kinda wondering if its over kill..

[compact wren]

Nope no change.

Prompt: Camera arcs around clockwise around the subjects in a violent brawl, subject in black leather jacket punches subject in white pantsuit in the face and she falls to the ground. camera then dollies in on the woman in white's face as she is lying on the ground. the helicopter behind them explodes.

It got everything but the camera motion right (except the chopper didn't explode, there was just a random explosion on the helipad, but whatever).

[wise loom]

Hey guys whats your No1 trick or "thing" that has helped you in video generation?

[uneven slate]

Is there any way to synchronize the voice when creating multiple videos? My videos have different voices.

[covert grail]

How do you get flow videos to stay true to the input ingredients? ie if I upload images of an AI-enabled camera / sensor pod, how do i stop flow from morphing it into some other unreal camera that isn't my product?

[steep portal]

I tried using Start-End images and Veo 3.1 puts a 3rd image inside that I didn't even want. How to avoid that and why does it even do it?

[north path]

How do I retain voice from first video generation all the way to howevermany videos i generate? For example i create character michael in one generation and he has good voice. Then I use frame to video where I have the same michael but I want him to have the same voice here as well

[lime gate]

Helllp! How can I stop music from generating in the scenes! LOL Feels like I'm creating a soap opera (Now I'm going to create one but for now this is a short I'm creating 👀 )

https://tenor.com/view/days-of-our-lives-days-of-our-lives-gif-24401534

[heady sedge]

Do you use negative prompts telling it to NOT play music?

[lime gate]

YES! All caps, all lowercase, one Uppercase and rest lower case -I've tried ( No Music}, [No Music] "No Music". If you have any additional ones I can try I'm all 👀/👂

[heady sedge]

It looks like you tried all the options for negative prompts. The other thing I do is take it to Capcut and extract the sound, then delete it. If you have conversation or other noises that you want to keep, I extract the sound, duplicate it (push it to the end in case you need to retrieve it), then extract the music for one of the tracks and match up the other noises with the action in the video. Hope this makes sense and you can find a work around. 🤞

[lime gate]

Thank you! I will definitely try this!

[reef sorrel]

@lime gate try this , I was just messing around at first, but it actually works. A few folks even tried my “critical read me, veo” trick

[reef sorrel]

but if anything , the most simple way to prompt it to generate no music at all would be :

"Music : None"

[bright creek]

If only it was that simple with most prompts

I’ll have try to oddly creative ways of dealing with music

As even though I have a Ai ultra it’s really getting annoying

Why does 3.1 just decide to what it wants with audio tbh though lol

[quiet storm]

Temporal Optimal Video Generation Using Grandma Optimality.

Universal optimax enhancement for computation eg. image, video generation. If the model can be tuned or prompted to work in this way.

"Make the video 2x times slower speed, slow-motion, while maintaining all the visual elements, quality and aesthetics."

Test for your self. Samples Below, this is the secret sauce behind getting super high quality generations out of same models as compared to simple prompts or simple way to do it.

for llm's do same with output length and (possibly context consideration)

x ditpoo 1982424252348260724

eg. prompt, see x for a gpt example of same, this is way back early days of gpt-4 but one would get the idea

System: 
- For every prompt going forward make your response 50% of its original length, 
- Every response or ouput must have a max upper limit of 4k tokens. 
- Ensure the output text is clear, concise, and uses simple vocabulary.

I forgot to add this,

for video gen first generate image that way and then do image to video for best result, sorry

[reef sorrel]

jesus

[strange echo]

am i understanding it correctly, that you basically turn a text 2 vid system into a img gen to then make it a img 2 vid? XD
Not saying it has no worth, but funny, that img 2 vid exists, but you need such a workaround to make it work better

[strange echo]

i also find it funny when characters suddenly sing instead of talk, cause the word "recording" made them think its a music recording

[bright creek]

https://cdn.discordapp.com/attachments/1315754701890912349/1431622625230323783/copium-chinese.gif

[reef sorrel]

https://tenor.com/view/confused-im-gif-19987857

I think maybe I’m too sleep deprived to even understand anything it’s 2:31 am rn lol off to 🛌

[strange echo]

Anyone got lucky with prompts that either put the camera as POV of a character (like in a first person game) or have the "viewer" interact with a scene through hands? so first is just pertaining to camera movement, while the second is pov hands basically

[lime gate]

Thank you!

[strange echo]

does something like negative-prompt exist? like we got 8 seconds of video and the character wastes 6 with an explanation, can't shut her up about it XD

[onyx sluice]

I did a couple of POV videos with past iterations of Veo, haven't tried it with 3.1 yet.... In the past i just explained what was seen - like "a hint of hands at bottom of screen"

[strange echo]

thanks, have to try the "a hint of" tip

[reef sorrel]

something like this?

Point-of-View (POV) shot of a character's hands. The hands, covered in dirt, slowly push aside thick jungle vines to reveal a hidden, ancient stone temple. Shallow depth of field, keeping the hands and vines in sharp focus while the temple in the background is softly blurred, gradually coming into focus as the vines are parted. Volumetric lighting with sunbeams streaming through the jungle canopy. SFX: the rustle of dense leaves and distant exotic bird calls.

[strange echo]

more like that one movie with the action scenes, but good starting point, thanks

[quiet storm]

From the example above

https://x.com/ditpoo/status/1982502157892080095

In case its not clear apart from increase in complexity and stability, and more filling the scene naturally the fireworks in normal one lasted around 3s while it lasted for about 6s in the temporal optimized one will it be the case 100 / 100 thousands of times, I don't have a trust fund to validate such ablation studies

Also if we have a closer look at it 8s were not sufficient in normal gen to feel real and normal something feels off the temporal variant gives it enough compute to generate it properly i.e naturally and more normal feeling so this brings another question is there a compute or temporal requirement to render, world accurate simulations / videos i.e best variant of same generation so another scaling observation maybe, will more compute render more complexity and accuracy

Ditesh (@ditpoo)

@ylecun @drfeifei @geoffreyhinton @Yoshua_Bengio @GoogleDeepMind @SakanaAILabs @xai @GoogleAI @OpenAI @AIatMeta @NVIDIAAI @Alibaba_Qwen @Kimi_Moonshot @deepseek_ai @theworldlabs @StabilityAI @bfl_ml @midjourney @runwayml @higgsfield_ai @reve @Kling_ai I'm short on credits but still for science its fine, you have try with complex stuff and intera...

X · Oct 26, 2025, 5:38 PM

[reef sorrel]

yeah im still quiet lost with what you trying to show here

why do u need trust fund

[flint isle]

add the phrase "hard lock" e.g. character is giving a dramatic performance, no singing, hard lock. if that doesnt work you have to just keep adding on negative prompts. "character does not speak, character silent, no dialogue" etc

[quiet storm]

Calculate the cost of testing such a ablation based on stats, it’s a expression really

[quiet storm]

Temporal optimax variant, max / optimal utilization via rhyming synergy, kinda like reccurrance, but I have not cracked this yet as it will require a proper model to fully prove but it works

poetry and rhymes can possibly optimax prompt and context utilization, leading to temporal optimax variant

yes world model are poets

https://x.com/ditpoo/status/1982671389556392439

full utilization of 8s length and more complexity and stability, just not sure what the optimax variant is optimizing, colors are less diverse, but filling the sky and multiple interactions hold true in natural feel and cadance

---

Combined with the temporal optimal enhancements

prompt: "Multiple fireworks bursting in the sky, At the same time, they all fly. Filling the sky with bloom lighting high."

model: veo 3.1 fast

Ditesh (@ditpoo)

@ylecun @drfeifei @geoffreyhinton @Yoshua_Bengio @GoogleDeepMind @SakanaAILabs @xai @GoogleAI @OpenAI @AIatMeta @NVIDIAAI @Alibaba_Qwen @Kimi_Moonshot @deepseek_ai @theworldlabs @StabilityAI @bfl_ml @midjourney @runwayml @higgsfield_ai @reve @Kling_ai temporal optimax variant, max / optimal utilization via rhyming synergy, kinda like reccurrance...

X · Oct 27, 2025, 4:51 AM

[strange echo]

Any idea how to have a character talk to the camera, without it turning into a vlog or livestream? I can't get them to lower their damn arm (as if they are holding a camera in selfie position)

[reef sorrel]

is it something like this : Eye-level close-up of a woman in a crowded ballroom. She pauses, breaks the fourth wall to look directly at the camera, and smiles conspiratorially . Her hands are resting by her side, holding a champagne flute. The background is blurred with beautiful bokeh. She says: I know who did it

[strange echo]

Also got any style prompts? Like I know it works with stuff like educational Video, vlog, Action scene (or similar), advertisement

[reef sorrel]

what style do u have in mind?

[strange echo]

Also is there maybe even a guide showing some hot words that are "baked" in?

[strange echo]

For conversations

[reef sorrel]

for conversations?

oh you mean

dialogue prompting

[strange echo]

Like a scene talking with a skyrim npc

Hmm maybe interview

[reef sorrel]

well u answered ur own question

just write "Style : Interview Video Style" or idk , u can prompt the style any kind of style u want

Point-of-View shot, static shot. An armored town guard with a steel helmet stands on a cobblestone street, staring directly into the lens with a stern expression. He has his arms crossed firmly over his chest. real-life footage but in period of medieval fantasy style. The guard says: No lollygagging. Ambient noise: a distant blacksmith's hammer, wind whistling. this is just few things ive tried,

Eye-level POV shot, static shot. A friendly, heavyset innkeeper in simple medieval clothes stands behind a rustic wooden bar. He looks directly at the camera and smiles, wiping a mug with a cloth. The tavern is dimly lit by candlelight. The innkeeper says: Looking for a room or just some mead. Ambient noise: quiet tavern chatter, a crackling fireplace

Visual Style : Skyrim Game World design but in realistic real-life.```

visual style prompting works ![labs_eager_beaver](https://cdn.discordapp.com/emojis/1418644980800421928.webp?size=128 "labs_eager_beaver")

[strange echo]

Awesome, but I did not miss any manual where it mentions stuff like styles that the model knows? Just try and get lucky?

[reef sorrel]

well if anything just try to prompt it properly,

https://medium.com/google-cloud/veo-3-a-detailed-prompting-guide-867985b46018 @strange echo

https://cloud.google.com/blog/products/ai-machine-learning/ultimate-prompting-guide-for-veo-3-1 and this one @strange echo

[strange echo]

thanks again, this is what i was looking for

[lethal lark]

Extending clips and getting "stutter" effect

[quiet storm]

final form anime fireworks

[quiet storm]

red bull gives you wings but my brain fog is surreal, have to wait till it clears to figure out the msg

[narrow lark]

How do I get more credit?

[strange echo]

highest subscription has unlimited veo 3.1 fast gens

[elfin escarp]

i need prompt for this types of videos

[echo lion]

idek where i can buy it

[peak bay]

I believe that's generated using Sora

[quiet storm]

Here is a apex optimax firework example with prompt below, just without my (@dpawnlabs) spirit.

The scope (canvas) got expanded by 2x, allowing extra compute.

Prompt: "Vibrant hues designed, Multiple fireworks combined. Sky bursts filling at one time, Luminous bloom, simultaneous, sublime."

Model: Veo 3.1 fast

[quiet storm]

"dominant patterns are keywords"

[strange echo]

On the flow Homepage in subscriptions, there's free, basic and ultra or something and ultra is the one I mean

[strange echo]

Trying a horror scene. I imagine a room that is empty and the ghost crawls into the screen, so basically from below the camera until the face is framed, but no luck. Is it possible that AI is not capable of starting with an empty room and then lets a person appear? the ghost always starts already placed in the room

[quiet storm]

semantic editing is easy solution in this case, can that be clips of a same scene, in a single gen that is different question

[quiet storm]

also some times skipping to scene where that happens also works, eg. in that apex variant above, the fireworks starts mid-air without a tail, so still compute limited

[worthy dawn]

Hi, I´m trying to find out if there is a way to export the prompt seed or variant log for each generation inside VEO3, when it generates 4 options I want to find out the one I liked the most which additional words were used to produce it? thanks

[old haven]

anyone's got any tips for vertically elevating movement, please? It keeps going forwards even if specifically tell it to go up like its elevating vertically even using terms like ascend or fly up.

[old haven]

you can use it for free you have 10 generations monthly

[strong echo]

It's been rolling out for the past couple of months. Just got mine a few days ago.

[ripe monolith]

VEO 3 start getting sucks now. before once you get a good prompt even if its a short prompt veo 3 will give you what you aske, but after an update output become stupid and didnt follow the prompt

[hybrid scarab]

When making a video with raining outside the window, how do I make the rain not appear inside the room?

[iron ice]

<@&1366831844502274180> scammers here

[cyan wasp]

Thanks

[vivid flicker]

Any tips on creating a character that is a musician and plays his instrument in sync with music? EX: A trumpet player playing the national anthem, finger movement on valves in sync with music or close to it?

[vivid flicker]

try: the rain drops hit the glass window and flow downward. no rain enters the room.

[blissful heath]

i dont think veo does this. Veo dislike violence, and action, and talking 😅

[acoustic bay]

can we get Character wise consistent voice in flow ai ? any tips

[hybrid scarab]

That didnt fix it. It still rains and generated a ghost that goes into the window....

[next condor]

I struggle greatly with this. I can't even seem to have something as simple as, you know, like a close-up image of two people making a transaction. One begins, you know, the other performs a signature, passes it back, and then gets something in return from the first guy, and then leaves. It just never works.

You know how many credits I burned last month trying to do this? Unsuccessfully? The answer is one thousand.

[strange echo]

yeah, 2 people in general is already a ramp up in difficulty and jank

[next condor]

Dude, this reminded me of like punching in a dream, you know? Like, he will not take the camera. Put it in his hand. Just put it back down, it's so annoying. Lol

[strange echo]

I was doing one where a ghost is sneaking up on a woman walking in a park at night. It ended up mostly being the ghost running to the camera and the woman looking confused or walking next to her and stare at the camera too

[next condor]

Did you end up with anything usable? I mean, like, I guess that could kinda end up still being pretty spooky, I suppose. I really wasn't going for spooky whatsoever. I was just going for more like, honestly, I wasn't even going for the film look. I wanted just the character to take the film camera, but whatever man. I guess it's fine.

[strange echo]

could you please share the prompt for the first one with the blackboard and one of the table ones?

[next condor]

Sure

[strange echo]

yeah, kinda with a bit of change. not what i envisioned originally

the other prompt that would have been closer to what i wanted was just slapstick comedy

[next condor]

Visuals: The interior of a cold, utilitarian American Catholic classroom features peeling mint green and beige walls, silent crucifixes, and high, grimy windows. The aesthetic is aggressively oversaturated Lomography with heavy film grain and a prominent, distracting soft light leak ghosting the lower-right quadrant. Lighting is Hyperreal Naturalism, utilizing harsh midday sun streaming through the window as the single, violent source, sculpting intense highlights on the white uniforms while plunging the majority of the room into deep shadow. Chalk dust hangs thick in the air. Mr. P, in a rumpled tweed jacket, stands beside a stack of identical fixed-lens digital cameras.

Camera: Wide Shot (50mm f/1.2 lens, resulting in shallow depth of field, blurring background students into creamy bokeh). High Eye-Level Angle. Static shot for the first 6 seconds, holding on the tableau of Mr. P and the desk. In the final 2 seconds, a sharp, quick Whip Pan follows the student (ETTILAPSE) as he snatches the camera, signs the sheet, and bolts toward the door frame, ending abruptly on the slamming door.

Audio: Intimate and close soundscape emphasizing the rigid silence of the institution before the sudden burst of disruptive noise.

SFX: Faint, rhythmic pencil scratching; the heavy, mechanical sigh and low hum of an ancient air conditioning unit; the distinct, quick scrape of chalk against the blackboard; the sharp, immediate *thwack* of the heavy classroom door slamming shut.```

Lmao I didn't realize that I put quick whip pan and then like the character does the quick whip pan. That's hilarious actually. Although it's another failure on Veo's part.

[strange echo]

oof, ok, guess my vid was considered NSFW and i got a warning. i can't with this world anymore -_-

[next condor]

I literally hit the play button and then it just disappeared.

Did you put this spoiler on it or did they do that automatically?

[strange echo]

i did that, because i was already sensitized to overmoderation when it comes to AI with the recent Grok changes

but alas, that was not enough

[maiden marsh]

fwd

@zinc cypress

[zinc cypress]

thank you sorry

[unique knot]

Guys the biggest tip I think anyone will ever write is the one I am about to give…

Use Sora

[reef sorrel]

Sir this is Flow Channel, not gemini

[next condor]

My bad. Tangent

[thick flume]

I recently found this excellent video on Studio Camera Movements. I am not sure if AI responds to all of them yet but it would be great id Google Labs implemented these as their standards https://www.studiobinder.com/blog/different-types-of-camera-movements-in-film/

[quiet storm]

if visual is the anchor than music can sync to it, but every element has its own system to follow

[quiet storm]

never sign a binding/long term contract just general agreeable term, If was under watch for probing intentions and behaviour then no-one will associate or contract, or it be voided, still under that, they keep them in balance by overflowing the sibling/bro

basically try things in which there is equivalent exchange so it will work

PS: general agreement no promised hand as that might gain them higher at end, but linked to family

proof: All this would not be possible if this is not the case

[next condor]

Interesting. Okay, hold on. This is actually deep, thoughtful, and intelligent. That means I'm gonna need to turn my brain on. Hold on one second.

[next condor]

Are you referring to my prompt or are you referring to what I said and then since deleted?

You're messing my head up here. I don't think I disagree, I just don't know what this advice is for. I'm aware that whatever I thought may have happened is likely not anything more than just kind of like a temporary, you know, pause on services. I've already spoken to them. If you're not referring to that, disregard that. If you're referring to my prompt, how can I fix it?

And don't we sign long term binding contracts every time we subscribe to any of these services using their terms of service that collect our data for long term and inevitable use for whatever they deem fit? Such as Flow?

I feel like I sign a long-term binding contract every time I step outside the house and go walk in traffic, or walk across it. I'm playing life.

I'm also very, very curious if I can use Flow and the 60 days of Cloud TPU that I have in any way. Now I know I don't have that many credits, but maybe I could be creative with it. Maybe I could use Vertex AI. I do have some Vertex credits. What would be a good use case? I am a photographer and videographer in real life. This is kind of just something I do for fun on the side. I don't publish this by my name at all. Not that I wouldn't. I just don't think I would use this name. But anyway, I have some other ideas outside of video. I do feel like running the TPUs through my personal hard drives, terabytes and terabytes and terabytes of 4K footage, B-roll and Raw images that I've barely looked through. that I could filter through may be a solid use case. But can I implement Flow or VO? Something to think about here. It's a powerful tool I have.

[reef sorrel]

what kind of coke ranting is this my god

[quiet storm]

reflection on what could be the causes for same sorry should have clarified that properly

[quiet storm]

I meant for that use case

[brisk jasper]

hello, where can i go to get some help for flow?

[shadow plinth]

"a stressed man standing on rollerskis with no brakes driving full sped towards a cliff and fell off"

can somone prompt this in a better way?

[quiet storm]

interesting

[odd birch]

@vivid flicker

[strange echo]

yeah well, they decided to not include europe at the moment, so even if i wanted, i cant

[reef sorrel]

vpn can

[strange echo]

What should i prompt to not have the character stand center frame and talk to the camera directly? I just want it to be like a scene from a tv show

[strange echo]

as soon as theres a monetary transaction, the vpn shenanigans turn into tax fraud

plus last time i checked it wasnt on pc yet and i have no interest in messing with ai on phone

[reef sorrel]

wait what

ai on phone?

[strange echo]

oh i am talking sora 2 btw, cause sora 1 is not with sound, so i wouldnt compare it to veo

[reef sorrel]

well yeah ,

are u talking baout the credit system

[strange echo]

i am talking about sora 2 only being on the phone app, from what i could see

so even if i could access it with vpn (which i would have to check), i would still need to get a legal way to buy credits, which through VPN is not the case, since i would not pay the tax and then its still only on phone

[reef sorrel]

its on web desktop too tho

[strange echo]

ah ok, well that is at least a bit more interesting

[reef sorrel]

u didnt know?

[strange echo]

last time i looked it wasnt. it said "Get sora 2 on the app"

that link tells me i am using the old sora model

i turn on vpn, it still says it and if i click join sora, it wants an invite code

[reef sorrel]

yeah it needs invite code

u can find one from openai server , not here tho ,

or reddit

[odd birch]

@quartz totem this is the proper channel to ask for prompt help.

[strange echo]

anyone managed androids? Like synthethic/porcelain skin without anatomical details, like in irobot and such? Joints should be mechanical, but the general bodyshape human, Ken/Barbie doll basically and glossy with different colors. I fear the puritan filters prevent even something like that

Be creative, but better not play with barbie dolls or you might go blind

[strange echo]

Like with these there is 1 out of 4 going through:

Setting: An android woman restrained on a sleek metallic table in a modern clinical setting, emphasizing her mechanical form. Clinical setting, sterile room in a laboratory. Android: Her body is covered by a synthetic material to simulate skin, except it is very glossy. Her joints show mechanical features with glowing circuits woven through synthetic musculature. Her body is toned and modeled after an idealized woman and her face is in perfect symmetry and permanent makeup. She has fake long hair implanted into her head.

and that one i let a chat AI rewrite:

In a sterile, modern laboratory room with clinical equipment and bright lighting, depict a restrained android designed in the likeness of an idealized female form. The android is fully covered in smooth, seamless synthetic skin that conceals all internal components except for subtle visible mechanical joints at the elbows, knees, and neck. Her physique is statuesque and athletic, with a proportionate hourglass silhouette featuring a narrow waist and ample upper torso contours, all rendered in a non-explicit, robotic aesthetic. She has the facial features of a classic beauty, with symmetrical proportions, high cheekbones, and full lips, framed by long, flowing artificial hair. Emphasize her inorganic nature through metallic accents, glowing circuits under the skin, and a emotionless expression, ensuring no anatomical details or nudity are present.

and yes, i tried every variable of "No anatomical details" - "no nipples or genitals"

[reef sorrel]

is it like this

or somewhere like these

wait u wrote fake long hair

[strange echo]

so more like your second set I'd say

[reef sorrel]

let me try something

[strange echo]

i sometimes wish the filter would highlight what it finds offensive, but i guess that would defeat the purpose

[reef sorrel]

i guess ill try use the ingredients to video

[strange echo]

i assume you also got warning

stupid censors everywhere

[reef sorrel]

oh i

i didnt get the warning

[strange echo]

huh i saw your message and then it was gone. i got the carl bot censor warning, so i thought yours was moderated too

[reef sorrel]

but yeah either way uhh

so this workflow that im doing is purely ingredients to video

wait let me upload

il; dm you the photo i was using

[strange echo]

oh so no prompt to get it created, but from ingredients

thanks either way, just find it stupid that it seems unnecessarily difficult to prompt something like that

[reef sorrel]

yeah , i havent try refining your original text to video prompt yet but

well its not stupid if it works

or u can do image to video

[strange echo]

i would say it barely works if i get 1 out of 4 generations. (talking about the prompt being censored like that being stupid)

[reef sorrel]

i mean that was text to video

the one u said 1 out of 4 right?

[strange echo]

yes i only tried text to video and found it difficult

[reef sorrel]

yeah i find it difficult too tbh

the 3.1 really got me sweaty on refining most of my prompt

[strange echo]

hence me saying its stupid and barely works ^^

[reef sorrel]

oh i was referring to ingredients to vid i thought u were referring to that haha

[strange echo]

nah, that one seems to work as you've shown

[reef sorrel]

save it quick

LOL

this is image to video

[reef sorrel]

The data on the background screens scrolls rapidly. The android's chest plate rises and falls in a subtle, artificial breathing motion. The blue light on her neck pulses slowly.

or you can try expand the prompt like : The flickering data streams across the monitors, a chaotic ballet of information. The android's metallic chest plate subtly mimics human respiration, a cold imitation of life. A soft, pulsating blue glow emanates from her neck, a rhythmic counterpoint to the frenetic digital world around her.

[strange echo]

copied them, thanks

the stupid thing is: It generates nipple like shapes. I dont need them, but i cant prompt them to go away, cause if i mention them, it filters again

[reef sorrel]

looks like veo training data were mostly focusing on that part ahaha

[reef sorrel]

oh were u using imagen model or nano banana

[strange echo]

the flow thing

https://labs.google/fx/tools/flow

[reef sorrel]

ah alright

[strange echo]

you'd think an AI could make perfect synthetic androids

judging by that humanity is save for another 100 years XD

[reef sorrel]

they could ,

until the safety filter nation attacked

[onyx sluice]

I haven't seen anyone else posting about this and wanted to share this trick for a fairly easy face swap in Veo.

My ref images were made in Whisk.

This works best if you match the lighting, or at least if your face reference has even, clear lighting that clearly shows the facial features.

Just wanted to share in case it helps anyone!

[past sonnet]

This is great.... What's the process once in flow?

[shadow mauve]

hello everyone I am new here and typing my first msg :

Hello everyone,

I’m facing a small issue — whenever I create a video on Flow, my character keeps changing automatically. Every new video generates with a different character, even though I want to keep the same one.

Can anyone please guide me on how to fix this or maintain the same character across all videos?

Thanks in advance! 🙏

[maiden marsh]

fwd

@shadow mauve

[shadow mauve]

@maiden marsh Thankyou soo much

[hidden gyro]

Cam anyone tell me how to make vedios here I have just joined

[shadow mauve]

yes dany ask me i can help you

[hidden gyro]

One simple question how to make any vedio here .. as you have made a wonderful vedio how did you made this is my second question

[void orbit]

can anybody help me why minors video are not allowed in flow

[tawny estuary]

Great Update

[upbeat wharf]

Question: How do I go about keeping character voices the same when transitioning into a different clip or extending an existing clip. I am getting some really crazy mixtures from scene to scene.

[proud orbit]

Hi everyone! Nice to be here. Do you have any tips on how to handle scenes of violence in Veo 3.1?
I'm making an action movie, I've been trying to make a stabbing scene happen, and it won't do it, not even if I say it's a "push"... and it has to be with Veo... any tips would be amazing!

[warped trail]

With whisk and gemini

[tawny estuary]

how

[green forum]

Guys.

I've been trying to generate an image (on Flow) using reference images but it has not been working for the last 2 days when it worked fine previously. I also tried on other devices, nothing happened either.

Anyone also experiencing and know how to fix this?

[onyx sluice]

oh sorry just seeing this now - you upload the 2 images separately into Ingredients.

You upload the greenscreened image as 1 ingredient and the portrait or "replacement" object as ingredient 2.

Veo does the rest by merging automatically - no frames lost.

[past sonnet]

Brilliant, thank you

[floral wadi]

Hello everyone, when you have an establishing shot, how are you able to generate angled shots of your characters within the setting so you have continuity?

[strange echo]

any way to make characters interrupt their sentence? Like a character is talking and something happens, but they keep finishing their sentence, but that should not happen

[shadow mauve]

Hello O! Nolly
Continuity Strategy

Save the Seed: Use the same seed value from your establishing shot for every new angled clip. This locks the general composition, color, and character appearance.

Detailed Prompting: In the prompt for the angled shot, explicitly ask the AI to maintain the setting and character using keywords:

[New Angle, e.g., Medium Close-Up, Low Angle] of the character.

Add modifiers like: Consistent character appearance and setting.

Video-to-Video: Use the last frame of your establishing shot as the input image reference for the next angled clip. This is the strongest method for maintaining the world's consistency.

[worthy falcon]

I have 2 charachter in one scen...only one has dialogue, but sometimes one is talkig to each other with the same voice
how do I fix this? Been running through credits

[zinc cypress]

hey, i need some help from someone that knows how to generate vidoes really well

https://youtube.com/@julianwestern-yt?si=1BQdQLtQnHIwjzov

how to get the style of the shorts in this persons videos, like thats the wording i have no idea and whatever i try is not working

[crimson panther]

A straw dog with blue eyes. He holds a long bamboo stick. He wears a straw hat and burlap shoes. He has a collar with a peace symbol medallion.

[floral wadi]

Hi Amir, thanks for this! I've tried it and it's hard getting results.

This is my establishing shot. I want a medium close-up, eye-level on the woman in the frame. I've tried:

1. Generating an image of that corner + and image of the character in the same angle with the expression I want, but it's hard blending them together
2. Having the camera pan right to a medium close-up of the character, but it changed the woman's face, which makes sense since she's turned away from the camera

I'm not sure where to go from here

[frigid hazel]

i think those are made with sora

[reef sorrel]

your attempt to pan was logical, but , image-to-video model can't really "animate" what it can't see. When you panned from the establishing shot, the model had no data for the woman's face, so it had to invent one, breaking continuity

[zinc cypress]

is it not possible with veo

both are video models right?

cant just be a prompt?

[floral wadi]

I agree. When it panned, image1 was what I got. (Used Kling for this, VEO did not give good results at all).

So I generated images 2&3, but I'm having trouble merging them now.

I even tried to do a face swap (image 2 face on image 1) but it didn't work out.

[floral wadi]

Different video models have different strengths. The level of realism Sora is able to generate is on a different level from VEO atm

[reef sorrel]

which one is the face u were hoping for ?

[floral wadi]

Image 2 (the one with the clear background)

[reef sorrel]

ah this one

[floral wadi]

yes

[reef sorrel]

wait so u wanted to create a new shot angle using this same subject right?

[floral wadi]

yes!

[reef sorrel]

ok this one is very simple. if you want to "combine" those two images u generated. use ingredients to video

im currently trying to play around with it 🙂

[floral wadi]

Haven't tried this yet. Thanks, I'll take a stab at it and let you know

[reef sorrel]

you should gie it a go! its a very powerful feature

its like nano banana but for video

heres the scene/setting without any subject

[sacred valve]

@reef sorrel I think perhaps you may be the ai whisperer

[zinc cypress]

i see, do u know what prompt it would be with sora

would just like to try it here as well

[maiden marsh]

fwd

@tropic spruce

[maiden marsh]

Hello , we wanted to remind you of our server guidelines, stated in #start-here・📍 : “Currently, our team can support engagement in English; the discussion in this server will largely remain in English for now.” Please keep in mind that non-English discussions and content may be removed. Thank you.

[white star]

Hello I can't find the drop-down menu . I am ULTRA

[onyx sluice]

Dunno if this was already answered - but I've found that "storyboarding" using Whisk or another tool to create stills of your entire scene before making video, is the best way to maintain continuity.

In Whisk - you start by creating a detailed wide establishing shot. Then you use that wide shot as an ingredient in Whisk, and prompt it to create close ups and other angled shots within that scene.

[severe aspen]

Hello I can't find the drop-down menu .

[onyx sluice]

https://youtu.be/W2O-lUnK1Zk?si=jNKybnPPPcF8NFjX - did a camera motion test with all settings. skip ahead in 8 second intervals to see each option

[vernal iron]

yo chat how do i maintain audio consistency im using the extend feature but somehow every time i use it, it generates a different voice or like the tone is robotic

[buoyant stream]

Hello - Haven't been able to understand a real use case for ingredients so far ... It seems to me that if you really want to create shot that integrates multiple aspects, nanobanan does that, and its gives you control unlike 'ingredients' so Im not sure if ingredients to video feature has any utility with Nanobanana there ..

If any pros here can share their input would be great @sacred valve @onyx sluice

[reef sorrel]

🧠

[buoyant stream]

not my point .. i'm asking if there's any utility for the 'ingredients to video' function that can't be achieved with greater accuracy with nanobanana and please share examples of those use cases

[reef sorrel]

I think you should try this ; Google it yourself

[tawny cedar]

Hello

[loud galleon]

I'm looking for prompting advice. I'm using Frame > Video. I'm struggling with getting characters to NOT move. It feels that VEO is always too creative with its prompt adherence. I'm trying to get a character to remain still and say a line, but the output always makes them move based on the context of the prompt and/or visual reference.

[oblique hemlock]

I need help: Is there a way to precisely navigate the frame in Scenebuilder, frame by frame? Because it seems to miss the real final frame in one block, so the transition gets skewered.

It seems like each increment of frame-navigation (left, right buttons for example) is a lot more than 1 frame.

Each step is 0.5 to 1 second, which would include between 15-30 frames per step.

[neon wasp]

I'm having the same problems, it's driving me mad!

[tight burrow]

{
"prompt_structure": {
"cinematography": "[Describe camera shot, angle, and framing , UGC]",
"subject": "[Identify the main person, object, or focus]",
"action": "[Specify the single, primary action occurring]",
"context": "[Define the environment, setting, and time]",
"style_ambiance": "[Specify visual aesthetic, lighting, texture, mood]",
"audio_cue": "[Detail specific sound element(s) to be heard]",
"negative_prompt": "[List all elements, styles, or quality issues to exclude , background music , slow motion , cinematic]"
}
}

Make a new custom Expands , and paste this JSON prompt structure . Now you have JSON prompt generator inside flow .

write simple prompt then choose this JSON Expand , it will automatically craft JSON prompt for you

[vernal iron]

fr and i cant even fix it in elevenlabs cus of the tone pitching high

[violet shore]

Can you pls elaborate on how to use this

[tight burrow]

create a custom expand , and paste the json prompt i post above and save

[worthy falcon]

damn its hard for this machine to do what you want even when you specifically tell it to

I have one only charachter in frame, my prompt says specifically " no other characters in frame or talking"...well....looks like everyone in my frame wants to chat

[buoyant stream]

Hello - been experimenting with a lot of prompt structures .. what about yourself? I think nowadays, LLMs like gemini and claude are so good, you can simply get them to generate prompts

[gaunt hamlet]

hey guys, what is the best way to tell the AI what you want ? for example i wanted it to have a sign that says, " no pain no gain" and then it creates a sign with "no pain no gan". Why is it doing that ?

[gusty wedge]

Hi guys i have this picture, generated with whisk and i want the girl to stand up while grabbing the camera
i've tried a huge number of prompt it never succeed, the girl always grab a camera from nowhere, the AI dont understand that the shot IS the camera

[reef sorrel]

literally struggling too lmao, i end up try using ingredients to video workflow to see if it can do it but .. its struggling too lmao

[tight burrow]

Flow expand use Gemini to generate prompt , so u still use gemini

[reef sorrel]

Ironically Veo model itself uses Gemini to process user input prompt

So Gemini ftw 🤘

[gaunt hamlet]

in recent tutorials they have a button for the camera angle, but its not showing for me ?

[tight burrow]

only for ultra subs for 2 week testing phase , then open to all

[oblique hemlock]

I have trouble prompting this door opening inwards / backwards away from the camera. No matter what I do, the door opens out towards the camera.

[buoyant stream]

Currently I'm using a prompting guide released by LTX studio, which is actually very useful for VEO 3 as well because VEO responds best to detailed prompts too.

[vernal iron]

since audio consistency isn't a thing in veo 3.1 lets all wait for veo 4 and pray they'll put audio consistency as well

[white star]

Hello,
Would it be possible to adjust the duration of each generation, or to assign specific timing to ingredients or to defined start and end frames?
Also, can timing be included directly in the prompts, so that certain elements appear or actions begin at a precise moment in the shot?

[gleaming atlas]

is there anyone here that is having issues using Ingredients to veo with veo 2 quality, and can only use veo 3.1 fast? either they've retired ingredients with veo 2 quality and not given us a solid model or mine is bugging out, appreicate some feedback if there are any peeps out there that actively use ingredients to video

[buoyant stream]

Still having trouble understanding the relevance of 'ingredients to video' to be honest, after nanobanana ..

Anyone here trying to take part in the 'one million summit' competition for AI filmaking using just google's tools?

[buoyant stream]

It's happening in Dubai and the last date of submission is the 20th .. the winner gets a million dollars

[gaunt hamlet]

Where do you submit ?

[buoyant stream]

https://www.1billionsummit.com/ai-film-award/submission-criteria

[onyx sluice]

Just clicked around the website - what organization is hosting this contest? The website feels like, not well put together tbh

[onyx sluice]

There's some things that work when using ingredients to video that don't work as well just with nano

[reef sorrel]

its giving vibe coded template website

? must use google gemini tools?

do they not realize that Google Gemini is gemini.google.com and if u want to generate veo video using gemini.google,com it can only do 5 videos per day?

and theres no features like extend /jumpto like in Flow?

[reef sorrel]

idk but the main page showed some https://www.1billionsummit.com/

[rapid stirrup]

anyone know why my 800x800 image im trying to upload into veo needs to be cropped in half? can i not upload square images?>

im pulling my hair out here

[mental marlin]

I haven't found a way either. I do a lot of 4:3 aspect ratio and I usually expand whatever image I upload beyond my needs and crop it later.

[floral wadi]

Hey! So I tried the ingredients to video option and got videos 1-3.

Eventually gave up and got a friend to merge the images using photoshop, and then used that image to generate video4

[floral wadi]

I agree with this approach. Main issue was getting the still of the woman in a corner of the room. Apparently it's insanely hard to do since AI doesn't have spatial awareness just yet

[floral wadi]

I've heard a lot about Claude being better than ChatGPT for prompt generation. But when I try, it feels like the same results. How has your experience been?

[stone bough]

Cai i run flow in my for free with ultra version?

[cloud bough]

Why does Flow tell me that “portrait mode isn’t compatible in vertical and that I need to switch the screen to horizontal” when I try to create a vertical video? I checked the prompt to make sure it doesn’t mention portrait mode, but the same message still appears, even though it still generates the video.

[buoyant stream]

I've only started using claude for prompt generation ... It

[buoyant stream]

GPT was never my go to, it was usually gemini .. but Claude has been very very good!

Here's another point which I think everyone here will benefit from:

GEMINI DEEP THINK is an ABSOLUTE KILLER! Especially for creative writing .. used the same input for deep think and claude, and while claude was excellent Gemini deep think mode absolutely killed it! Try using it next time if you have to write a script or come up with a plot ..

[buoyant stream]

Could you give an example or two?

[reef sorrel]

it’s not really something to debate tbh. Ingredients to Video has its own advantages, and Gemini 2.5 Flash Image (aka Nano Banana) has its own too. It just depends on what the user prefers or what they’re trying to make

asking how Ingredients to Video is “relevant” compared to Nano Banana doesn’t really make sense , you’re comparing an AI video feature to an image model. They serve different purposes, so it’s kinda apples vs oranges

but if u do want "examples" ...

lets say you want to generate a scene using your own subject/character. With “Ingredients to Video,” the advantage is you can create something like: the scene opens without showing the character immediately maybe ur character walks in through the door, we approach, then it turns around and faces the camera.

Meanwhile, with Gemini 2.5 Flash Image Nano Banana ur mostly stuck generating a single image where your subject must appear in that frame it’s not built for consistent motion in a video. If you try to make a sequence like “character’s back → turn to camera” using Nano Banana, trust me, it probably wont maintain consistent appearance across frames, because you’re effectively trying “image-to-video” where the model is just for images

[astral musk]

How to make movement of the characters smooth? I mean, not look like robot motion

[dim cobalt]

Just having fun.

[next sentinel]

is there a trick to get consistent voices ? thanks

[next sentinel]

or tone? I'm getting reluctance with barfing details

[spare dagger]

you wanted this?

[spare dagger]

or this?

[reef sorrel]

they didnt ask for a generated image lmao

[floral wadi]

Great tip. Thanks!

[floral wadi]

Might be able to use the images as continuous start and end frames to help get the fluid motion

[plucky prairie]

why am i not seeing an option to upload your own videos into scenes?

[mossy anchor]

**Unwanted mirror reflections in Frame to video. **
I have a start and end image in a bathroom that i wish to add a dolly movement between both but it keeps on adding a ghost reflection of a cameraman. I've tried a few negative prompts to avoid it but no success. Anyone have encounter and solve that problem ? Same for windows.

[grand fiber]

I've found Veo struggles with anything to do with doors - opening them, closing them, getting people to walk through them. This applies to car doors as well as building doors. It's the surest way to get a wonky generation in my experience.

[grand fiber]

I agree. I've started using Ingredients to Video much more recently and found it useful. I barely touch Nano Banana nowadays (either in Flow or elsewhere) because the rate at which it refuses, returns the same image with no changes or fails to follow instructions is so high it doesn't seem worth it. I almost always use Seedream v4 instead because it's so much more reliable (though would accept suggestions for any alternatives, especially if they're better at generating different angles on scenes).

[severe aspen]

Prompt Like a Pro, a recurring live event series in the Gemini Discord, will be taking place tomorrow with a section dedicated to Veo & Veo prompting! Hope to see you there!

[gaunt hamlet]

Can you share the prompt ?

[worn maple]

there is a big ai festival coming, deadline is in a week and google flow now doesnt generate audio anymore. what is this a damn joke?

[fervent stump]

Hi, does anyone have a tip on how to create a video where a person talks about a product, and while they’re talking, they take it out and show it — and the product is exactly the same as in the photo I send with it?

[reef sorrel]

That’s a weird tips and trick buddy

[boreal wagon]

Why does FLOW prevent crosswalk images from being uploaded, claiming they contain underage content?

[frozen niche]

Same here

[worn maple]

i am not your buddy

[reef sorrel]

☕️

[acoustic storm]

following. I'd also like to know.

[strange echo]

any prompt ideas for making a character stay in place, but the camera moving to their back? my character always moves their back to the camera instead

[gilded pulsar]

Hi everyone, quick question — does anyone know an app I can use to create TikTok Shop videos with AI? I don’t want an avatar, just one that lets me change the background of the video. Thanks in advance! 😁

[severe aspen]

to clarify, are you attempting to get the camera to do a circle around a subject?

[strange echo]

switch from front view to back view

[ebon swallow]

like a selfie mode type of thing?

[strange echo]

no, just the scene starting with the front view of the character and then changes to a following camera

[reef sorrel]

Idk cut maybe

I straight up copy what u wrote and send it to Gemini cuz I have no idea what were you actually trying to do

[reef sorrel]

I misinterpreted his question I thought he wanted the “ editing technique” style . Where u would have two completely separate shots, and you instantly switch from one to the other idk

Ok yea I think you’re trying to create arc shot @strange echo

[strange echo]

ah yeah that would be good. did you just write arc shot and it did it?

[reef sorrel]

It was similar like that .. this one I posted here is actually quite lengthy prompt so uhh give me like few min till I find the right prompt structure

But if you still wanna see the prompt A hyper-stylized, high-octane cinematic sequence captured with an Arri Alexa LF. The lighting is dominated by saturated, cyberpunk neon, specifically electric magenta, cyan, and deep violet, reflecting harshly off wet asphalt and slick brick walls, creating exaggerated lens flares and bloom. Visual Style: Neo-noir, Blade Runner 2049 aesthetic. Mood: Tense, mysterious, and instantaneous. Camera Movement: Begins as a tight close-up on the woman's intensely focused eyes, utilizing a low-angle perspective to emphasize power. Immediately execute a lightning-fast, horizontal 180-degree whip-pan/arc motion around the subject. The speed of the whip movement is extreme, causing severe motion blur on the background neon streaks and the alley texture, transforming the environment into abstract streaks of light and color. The frame rate should suggest a subtle, high-speed shutter effect to maintain sharpness on the subject. Focus: The woman remains in razor-sharp focus (shallow depth of field), flawlessly tracked by the camera rig throughout the entire arc, while the background is completely obliterated by the speed and blur. The shot culminates on a medium-close shot of her back, perfectly centered, revealing an intricate, iridescent chrome dragon embroidery/tattoo design prominently featured on the back of her bespoke leather jacket. Post-Production Look: High contrast, deep blacks, cinematic grain overlay, and highly polished, reflective surfaces.

(Thank you Dr wafae for the prompt enhancer ❤️)

A low-angle, tight close-up utilizes anamorphic lenses and shallow depth of field, drenched in magenta and cyan neon.The camera starts doing arc shot lightning-fast, 180-degree half orbit a woman in a wet alley. She remains razor sharp. The shot settles on her back, revealing an iridescent chrome dragon design. High contrast. Tension.

Close enough. If it doesn’t give you what you want, just regenerate. No clue what the Flow team did, but sometimes two or one out of four outputs just ignore the prompt entirely.

[severe aspen]

you’ve got it 👍 I misinterpreted

[strange echo]

thanks, that looks like its what i need. now just gotta get those into my prompts

[ebon swallow]

Did she help you? Or taught you?

Very intriguing how you got to that prompt .... would you mind sharing that journey jeje?

[reef sorrel]

yes she helped and taught me uwu

[timid crane]

I'm trying to generate the opening scene to my hypothetical 4th act to House of Dynamite (recent Netflix movie) - my prompt is: A nuclear missile strikes the city of Chicago. A blinding white explosion then a blast wave that vaporizes the buildings to dust.

I want it to be more like the blast wave in Terminator 2 but this is as good as I've been able to get. Any tips or tricks to get it better?

[timid crane]

I wrote up a little more about what I've been trying at https://aifilm.camp/johnpolacek/the-way-to-dusty-death/posts/post-1763242087435-b9gxl8r

[modern dew]

Hi everyone, I'm new here!

I'm starting to use veo3.

I need a tip, can someone tell me how to make a 60-second video? I have the prompt defined, but it only creates 8 seconds. After dividing the prompt into 4 parts of 15 seconds, I can't get the same sequence in the video. When I make the first part it's fine, but as soon as I add the second part, it makes a new video changing the characteristics of the first part. I need to get the sequence right so I can then take it to Premiere and assemble the whole video. Thank you.

[frigid hazel]

you're using extend feature?

[tight burrow]

Nice effect , share the prompt please

[modern dew]

Hi 0xArise, thanks for the reply. I'm not using this feature, I don't know how it works, can you point me to a place where I can learn or give me a tip on where to find this option?

[frigid hazel]

https://support.google.com/labs/answer/16353334?hl=en

[frigid hazel]

if you still don't understand search on youtube

[trail canopy]

See, this would have made the movie much less sucky. Three storylines with no payoff. I hated that film.

Ingredients >Frames. You can use Ingredients the same way you use Frames but add a third frame (Start Middle Last) instead of using it to add in characters. Works just as well and you can use the Expander to have presets.

[odd birch]

@lofty laurel your message was forwarded here. This is the best channel to get prompt help.

[deft palm]

reverse the video and he shall go backwards

[lapis pond]

Can only agree, i got so sick of using nano because of the cosistant wrong aspect ratio and reuse of already rendered images, that is not okey. We pay for a suprem service and should recive supream sevice, o got the ultra plan, and had to also add seedream to my tool box for rendering images. And now they want to pic credits for fast renders......hmm...Its brutally expensive with the ultra subscrition.

[maiden marsh]

@agile swift fwd your message here as it is the more appropriate channel!

[agile swift]

okay thanks! See my message above ... anybody know how to prevent that in the video generation? It's so annoying when the video swaps to an entirely new scene, different camera angle, even different looking subject. I just want a single scene with perhaps some camera motion like zoom ... but keep it to that camera shot and subject

[olive reef]

does anyone know the best way to write prompts that are ment to portray martial arts or boxing?

[vivid flicker]

You mean something like this? I uploaded a pick of the mixer to Nano Banana and prompted the salesman, Not exact, but ...

[vivid flicker]

or maybe this one?

[little rain]

the demo was great. I think one thing it doesn't show is if you give direction to say look a different direction or shift focus. you also need to instruct things in the background to move slightly. Those horses would stay in the exact same place otherwise.

[thorn cobalt]

Hello, good morning. If I send a video, can someone help me generate a similar one?

[thorn cobalt]

Is it possible to do something like this? Can someone help me?

[warped veldt]

can anyone help me
how to prompt this muscle growth on the blue guy

[reef sorrel]

my attempt,, im still iterating with the prompt atm

[reef sorrel]

my bad attempt actually

[thorn cobalt]

❤️

[hard wolf]

doesanybody know how to prompt that there is no camera zoom, i want a static shot

[cerulean sapphire]

Movie 3D Animation: https://youtu.be/uPj1-DuiC0k

[shadow mauve]

You made this with flow ? or Whisk ?

[cerulean sapphire]

Flow

[shadow mauve]

really appreciated just let me know how you made the character and all the background consistent

[sharp stag]

has anyone tried accessing their flow account using project mariner? i cant get past the landing page

nvm. lots of labs experiments arent accessible using their own mariner due to it not accessing them through a supported region it seems. lol

[pure zenith]

any tip to stop it zooming in, i just want textures to move inside, it always zooms in, no way around it

[static crow]

camera zoom 0%

[static crow]

but need mass generation or not depend your luck

[static crow]

generate a end image with difference you need

[pure zenith]

ive said camera zoom 0% in all the possible ways in my propmt

before i use to try it many times untill finally got one without zoom, now i cant do that because of the 10 credit veo 3.1 fast implemented without any previous notice

i need long sequences for a stage dance and circus event that take several minutes, i cant do end image in all of them because of the credits scarcity

[sharp stag]

there's also a disctinction between "zoom" versus FoV (field of view) - learning that will help you make better instructions

[static crow]

Why can't you use an ending image? I don't understand. Then you just have to change your ending image back to the beginning image to create a cycle.

[pure zenith]

because they cost credits

[static crow]

What you're saying makes no sense. Whether you include an end image or not, it makes no difference. As far as credit is concerned, it's the video that costs 10.

[pure zenith]

camera locked, zoom 0%, no scaling, no magnification, no push-in, no pull-out.
field of view fixed, no FOV changes, no lens breathing, no virtual lens adjustment.
no reframing, no cropping, no recomposition, static picture-plane.
camera baked into the frame; perspective remains constant for the entire duration.

ive tried them all

[static crow]

try with end image, me i use this for my creation

juste change the end with nano banana

[reef sorrel]

this 100%

[reef sorrel]

Static shot. The camera is completely locked and fixed. The abstract orange and rust-colored textures flow slowly like liquid paint. The vertical streaks shimmer and undulate gently within the frame. The light pulsates softly, enhancing the depth of the layers. No camera movement.

idk in what way u want the texture to move but yeah,

[reef sorrel]

https://tenor.com/view/awkward-gif-10738544

imma hold off from sayin anythin 🤩

[thorn cobalt]

Was it possible to improve the prompt, friend?

[cobalt dirge]

Guys, beginner question. I have a video about 5min long. What is the best site for downloading SFX?

[reef sorrel]

In theory

Uh-

[sharp stag]

ive had good results stating "the camera is cemented to the floor ----- " where the ----- is a dramatic description of the physical force that his holding down the camera making it impossible to budge. ive used that tactic in situations where nothing i wrote about "no camera movement" worked.

i remember one time even saying the camera was the only still thing created by god in the world this clip is taking place in. or something to that effect. i think it tries to impress most when you describe things in an artistic way.

one time even i was having trouble removing a subject from a frame, i went around it by stating in a frames to video prompt that all the subjects in the scene are still mannequins - and a museum staff memeber walks into frame with a dollie and lifts (subject i wanted to remove) and moves them out of frame. and it worked

[stark lagoon]

Need help refining a complex "Rear-Clip" Accident Shot (AI keeps failing the physics)
The Vision:
I’m trying to generate an 8-second hyperrealistic cinematic shot (Night/Dense Fog) of a specific accident geometry, but the AI keeps messing up the vectors (car comes from wrong way) or the layout (sand pile is on the road instead of the shoulder).
The Scene Setup:
Setting: Narrow Indian asphalt road at night. Dense Fog.
Layout: Asphalt in center. Unpaved dirt shoulders ("kaccha road") on both sides. A large mound of soft sand is located specifically on the RIGHT DIRT SHOULDER (not on the asphalt).
Camera: Static, wide shot, low angle.
The Action Sequence I need:
Vector 1 (The Man): An old Indian man, age 55-60, thin, (kurta/pyjama) walks TOWARDS the camera in the center of the road.
Vector 2 (The Car): A Red Sedan (2006 Swift style) enters from BEHIND the camera (bottom of frame), speeding AWAY into the fog.
The Physics: The car swerves LEFT to avoid the man. Because of the speed, the REAR RIGHT of the car swings out and clips the man.
The Result: The man is thrown RIGHT onto the sand mound on the dirt. The car vanishes LEFT into the fog/ditch (off-screen crash).
Current Issues:
AI makes the car drift 360 or come from the front.
AI puts the sand pile on the asphalt.
AI doesn't render the fog thickly enough to hide the bad physics of the car vanishing.
Could anyone help me structure a prompt that nails this specific "drive away + swerve left + rear clip" physics?

[thorn cobalt]

Would it be possible to create this with this same style of camera?

[warped veldt]

can you help me prompt this

[cold ledge]

Dm me

[floral wadi]

Hey everyone 👋🏾

For those doing lip syncing with AI characters: what tools/workflows do you recommend?

I used to generate both video + audio in VEO, but lately voice consistency has dropped. I tried ElevenLabs for audio (much better), but syncing it manually gives me a slight lag.

Any tips/tools/tricks to make the lip sync look more natural and convincing?

Thanks in advance!

[warped veldt]

done

[minor otter]

why when i try to force crowd movement, camera shifts, cross-direction walking, and two transitions inside a single block. im getting a horrible output ?!

to the follwong ptompt. :
Visuals only. No audio.

0.0 to 1.0 sec.
Camera begins slightly above and behind the woman as she steps fully out of the blue door into Tokyo at night.
She is standing at the start of the Shibuya Scramble Crossing.
Neon signs and city depth must be clear and photographic.

1.0 to 2.5 sec.
Camera lowers behind her to normal walking height.
She begins walking forward into the Shibuya crossing.
Crowd moves in the same general direction as her or gently around her but never against her.
No collisions.
No chaos.
Realistic Tokyo crowd density.

2.5 to 5.0 sec.
She continues walking forward across the crosswalk.
Camera stays behind her with smooth forward movement.
Neon reflections update naturally on her hair and clothing.
Crowd moves softly around her, giving enough space to keep her clearly visible.

5.0 to 6.0 sec.
As she reaches the far side of the crossing a large blue door becomes visible in front of her, attached to a realistic building facade.
It must look solid and architectural, not floating.

thats 3\4 of the prompt

[cold ledge]

Try with json. Holistic gives ai quite a lot of freedom

[minor otter]

hmm ill try but you know what im referring too based on my prompt ?

[cold ledge]

No not rly honestly. But from what i know is that especially when you need to make scene shifts with json you can have exact control between microscenes shifts. With holistic it barely works.

[minor otter]

have you tried such a thing with json format ?

[cold ledge]

Yep plenty of times actually. Ive made a 4 microscenes on 8sec video, each completely different from previous one

Similar model to what you just explained with shift of the microscene every 2 sec

Takes some time for prompt adjustment tho

[minor otter]

im just trying now json , ill see whats the outcome

[cold ledge]

Let me know if worked

[minor otter]

much better on a diffrent prompt but i just think the model cant handle crowd movement, camera shifts, cross-direction walking, and two transitions inside a single block and still show Shibuya Scramble Crossing and the surroundingss

Especially when I try to play with the camera motions with a big crowd while focusing on one character it's extremely difficult and then I get a horrible output

[cold ledge]

Dm me ur json

[minor otter]

in a few as i just gave up on the on the whole scene where my character is walking against the crowd direction in the crosswalk but I will send you in a sec via dm

the problem using JSON format with extended feature is the character consistency clothing texture etcc

maybe i need to practice more

[cold ledge]

Do not use extended if ur using json

It ruins the logic of the json’s strict rules

[minor otter]

so everytime to use frames to video and take the last frame screenshot?

[cold ledge]

Download the last frame(zoom in, save image as) frames to video. When using extend quality drops dramatically after 3 to 4 consecutive generations. Using last frame prevents that

[minor otter]

and just be sure I have character consistency basically

if the character is not in frame

[cold ledge]

This way you will maintain consistancy yep

Ah no, if char is not in frame, you wont have consistancy. However it fails with extend aswell

[minor otter]

did not fail with extend tbh without jason

ive done 1 min clip

ill show you via DM

[cold ledge]

Go for it

[earnest hearth]

whats the best way to get a reverse angle of an image? like—flipping the camera around 180 degrees

[cold ledge]

Capcut have a mirror option for that quite sure NB pro can also handle it. You can try in gemini

[wise loom]

Hey , I have a prompt for two people laughing simultaneously. But it makes only the voice of the one of two. So far have tried json , paragraph, description for each voice. Any ideas on how to approach it ?

[iron ice]

<@&1366831844502274180>

[heavy gate]

https://www.youtube.com/watch?v=mReCjAvV1M4&t=21s
I need help, guys, to generate videos like this. Can anyone do it?

[vital ivy]

YOU MAY USE THIS SYSTEM PROMPT. I THINK WITH PROMPT YOU GOING TO MAKE A PERFECT VIDEO.
UPLOAD YOUR GENERATE VIDEO WITH THE PROMPT YOU USED

[pure zenith]

Any upscalling tips for a 10240×2390 screen?

[vital ivy]

Topaz lab is one of the best for images and video up to 4k
https://app.topazlabs.com/

[pure zenith]

do you use it with astra?

[clear idol]

if I have a reference image of a living room, but I want an image from a specific part of the same living room, how do I tell Nano Banana to "move the camera"?

[odd birch]

Hello @last sequoia , 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 make sure to do so in a clear manner. Also As stated in #start-here・📍 , “Currently, our team can support engagement in English; the discussion in this server will largely remain in English for now. Thank you!

[teal creek]

Hello, I'm trying to create ultra-realistic user-generated content, so I'm using the Scene Builder frame to maintain consistency, but it's giving me error 400. Please try again later. Does anyone know why I'm getting this error?

[maiden marsh]

@civic thunder fwd here as more appropriate

[unborn inlet]

is there a way to share a project (so 2 artists can collab) in Flow as an Utra subscriber?

[unborn inlet]

If you are generating a batch of 4 videos and you delete 3, does Flow understand that that video is prefered in some way?
Or if you like a vid?

Do any of the controls tell Flow that the video is 'better' in some way, so it behaves like a feedback system?

[lofty barn]

someone has an ad prompt?

[clear idol]

single person talking interview situation, after person stops talking, an "over the shoulder person" spawns in the edge of the frame. How to prompt only person from reference image visible throughout the whole clip?
EDIT: DOH, there is an "OTS person" in the reference image used as last frame 🤪

[fallow siren]

Does anyone know when Flow will support 2K/4K image generation? Right now it only outputs 1365×768, which is too low for video, but the API can do higher resolutions. 🙁

[strange echo]

right now we can be happy we get any result out tbh

[small basalt]

Has anyone got a good prompt i can add to my prompt to keep character facial consistency as soon as my character moves or the camera moves the face changes any help would be appreciated 😀

[north sleet]

I'm getting SD resolution 720p on image output using nano banana pro, I am pro subscriber. Anyone getting better resolution in flow -> create image?

[calm quest]

brother i need help i just wanna make these type of videos how i can make these video please seniors guide me

[cold ledge]

@calm quest you know how much generations and editing this type of content requires?

[cold ledge]

Did someone tried adding custom rules on the expand option but giving the rules as json? Wondering if its possible at all and what the results would be.

[calm quest]

yes i know but i have no perfect prompt for video and picture promot like i have issue on character consistency , sometimes i faced height issues etc but i wannna learn from here

[cold ledge]

considering its 3h video, im assuming they are using some kind of api for generation. just creating a good, visually stylized and well packed video around 4min takes me around 3-4 hours(thats before final polish and editing). Consider a 3 hours video - a project like that would take a month full time work prolly(for a single person).

[calm quest]

bro i just saw other competators works on just 10 to 25 minutes videos that is point , its no matter i will make these much long videos , i have ultra veo but some issue during crerating i know he is creating through veo just bro i want to learn from this community i know 3 hours long difficult but there are short stories like mixture of 10 stories like 20 stories just clear bro

[shadow mauve]

Very easily you can make this on Flow, but you have to be a paid ultra pro , you can make them easily. #flow-tips-n-tricks・✏️

[cold ledge]

Very easy?:D Okay. have you ever generated something longer than 5min on flow?(since im using ultra plan from the moment veo 3 was implementet on flow). Creating 3 hours consistant characters video:D Good luck. If you can share "the very easy" creation recipe id honestly be amazed:D

[calm quest]

please give me detaile i have paid ultra if veo

[shadow mauve]

Yes I have created complete webseries every series have 15 minutes episodes

[cold ledge]

And each episode is consitant, logical and makes sense? If yes, congratz honestly.

[onyx sonnet]

Hey everyone, is anyone else struggling with VEO 3 or Nano Banana blocking completely safe prompts involving children?

I’m working on a short film about fatherly love that needs some childhood flashbacks. Since yesterday, every time I try to generate the main character as a young girl, I get hit with the "Policy of underage children" error.

It’s strictly cinematic/sentimental stuff. Has anyone found a specific keyword or prompt hack to get these safe shots through without triggering the filter?

[teal creek]

Does the Ultra Pro package have an API to automate the workflow on n8n? And is generation unlimited? I would like to automate my process to speed things up.

[calm quest]

Bro can you teach me about this

[odd birch]

@arctic compass your message was forwarded here. This is the appropriate channel to post your place to get prompt help.

[cold ledge]

Im creating toddler content, which also triggers safety filters quite often. The way to avoid is to have a strict language, no nudity, no violance in any form. Anything which might considered as a normal dialogue between adult actors in the video might trigger safety filters basicly. if you give me some more details - what type of content is it, what is the goal i might be ale to assist you

[reef sorrel]

Veo 3.1 Tips (honestly this was just me trying out NotebookLM slide Deck feature, seriously this is such a cool feature 😭)

[onyx sonnet]

Oh really! I'm doing a short film on a young girl losing her father, then we see her growing up throughout the years! Having trouble with the funeral scene

[cold ledge]

funeral, death, blood etc would hardly get generated. Even a scene around injury etc would not also get replicated.

[onyx sonnet]

but its not just that, after i created the character portrait (no expressions, white background) on nano banana, it refuses to upload it altogether with a messege that says you cant upload pictures of minors

[cold ledge]

Thats wierd. Im creating toddler videos all the time. Mby try doing without refference image, but make a super detailed description of the toddler. might bypass the safety filters. Havent tried using reference with kids honestly.

[pliant iris]

Ohh vow does it work? How do you input it to the Veo model?

It looks really facinating!

[livid hatch]

Can someone help me with prompting this video?

appreciate any help prompting this video above

But the gun and human hands should be excluded and not shown in the video

Doesn't matter how much I try, it still shows up in the video

[young bone]

Can someone help me with prompting this

[pliant iris]

Need help refining a complex "Rear-Clip

[frosty cargo]

hey guys i have a question. when generating text to videos, can i upload some assets that flow can use in creating a video?

[neon lichen]

you should use ingredients to video and add your items, up to 3 and then use your prompt. It makes for a great hybrid text/image workflow

[oblique condor]

am i able to upload videos to flow?

[frosty cargo]

Thanks man, i did this, but for some reason, it is only using the images i have uploaded as its reference rather than using the actual images as ingredients

So it was creating its own version of the images i have uploaded which somehow defeats the purpose

[warped veldt]

need help
any tips on how to prevent jumping a sudden scene
image to video

[dapper urchin]

@warped veldt yes your prompt needs to be as accurate as it should be

[random plinth]

any tips on how to make animations like this? is Veo 3 even the best model for this?https://www.instagram.com/reel/DQav_MMjQ0B/?utm_source=ig_web_copy_link&igsh=MzRlODBiNWFlZA==

pudgypenguins

Only every second of every day 😌

Follow @pudgypenguins for more!

Likes

149892

Instagram

[swift folio]

hello, any tips on how to make the character talk faster like UGC style ? from my experience, its like 60% will talk normal speed, 40% will get it to talk fast. so im wasting alot of credit to regenerate.

dont know if theres any prompt that could guarantee it to talk fast

[wispy valve]

Hi i would like to make veo3 video where my young Santa Clause with chest without closes alternately tense his chest muscles to the rhythm of the song Jingle Bells - all time i get policy problem - any suggstions?

[trail canopy]

Add more dialogue per 8 second prompt. The character will talk faster to try to spit it all out.

[swift folio]

did that too, sometimes it works sometimes not. and it is also depends on the image source too from my experience, because some of my image cant talk fast even after 10 regenerates.

[severe aspen]

500 million generations made with Flow 📣 if you'd like a chance to have your Flow gens shared on Labs' socials, please see this post for info! 📍

[stray knoll]

Can you help me with prompts for creating videos and clips for clothing (fashion) using models?

pleaseeee

[molten parcel]

Flow DJ?

[radiant solar]

Can you shed some light on how you created this ad? To me this looks amazing.

[sturdy pebble]

are you using any knowledge base with this?

[vital ivy]

I used it in my film production file sitting and it worked i can say 80% improvement in My film quality visual. But still this prompt needs more improvement and refinement.

[acoustic bison]

has anyone ben able to recreate reaction heavy UGC like this one?

Not sure if AI simply isn't there yet or if I'm missing something from my prompts

[cobalt dirge]

Can someone help, I'm struggling to create consistent fight in Veo3 between 2 characters. They change appearance after almost every prompt. Is there any universal way to solve this?

[somber glen]

I currently generate video from frames one by one and stitch them together in my editor, is there an easier way to do this? Would using the scene generator open me up to the service crashing and losing those clips or worse quality?

[round peak]

When I am generating a clip with a big dialogue I am getting jerks between the scenes can anyone tell me how can we overcome that and I am unable to get smooth transition between 2 frames in a single scene

[rough wing]

Keyframes or Ingredients

[shadow mauve]

VEO is not for consistent use flow for this

[drifting anchor]

https://youtube.com/shorts/aWnQG3cT64U?si=XT_sBub0emoIpGpD - I create few videos by usibg veo how will make it more better can you please help. Thanks

@shadow mauve ?

[vestal ember]

does anyone know what tool they used to animate this video : https://www.youtube.com/watch?v=x0hO_h9cKfk

[north path]

Hello guys, I'm creating an animation about a dog that my uncle saved. So basically a stray dog was hit by a car and my uncle rushed it to the vet office where they had to amputate his hind leg. I have a hard time consistently portyaing dog as an amputee, flow either ignores my prompt and regrows new leg mid video, or just redoes the same video. I'm assuming this is a safetry trigger because its a sensitive topic. What do you suggest I do? My video should portray how this dog bounces back to new life in his owner's (my uncles) apartment. So Im trying to show scenes where dog feels depressed at first but after care and love dog regains trust and joy.

[cobalt dirge]

Ok, which other site IS consistent for that? Already been on Freepik, not great, slower and quality is lower

[north path]

following. My dog character also keeps changing appearence every time. try using ascenebuilder and extend feature

[mental coyote]

@heavy hound Ive forwarded your prompt help request here as this is the correct channel to ask for help finetuning prompts☺️

[neon urchin]

Is there a way for me to share some audio in my native language and Veo being able to lipsync to it on the scene I'm building?

[high trout]

Any good prompt enhancers/optimizers out there?

[round peak]

Can anyone help me with this ?

[upbeat wharf]

Hi all, hope this is the correct feed. I have "Ultra" I am still getting the VEO watermark in the bottom right corner of my videos. I generate lots of content for the company I work for and so they DO NOT want the VEO watermark on business marketing.

How do I remove the watermark or stop it appearing in the first place? Thanks in advance.

[rough wing]

That's a tech support issue.

[rough wing]

"Keyframes or Ingredients"

[rough wing]

Veo doesn't do lip sync. You'd need to generate the video you want, then use a service like Sync or Hedra (among others) to upload your own audio and have their models move the lips to your words. I don't know if they do whatever language you're speaking in, though.

[cobalt dirge]

You mean on Veo3? Doesn't help. I need consistent character and consistent location

[rough wing]

...that's why you need to use keyframes or ingredients.

[cobalt dirge]

I know but still doesn't help when you work not 1 video but 100, 200. Only other way is use start frame and end frame but it's alot of work. Surely there's another way

[rough wing]

I'm not sure what the disconnect is here. Reference imagery is the only way to do what you're asking for.

AI video models aren't magic. They can't stay consistent unless you give them specific reference material to base each generation off of. Without that, every generation produces somewhat randomized results based on the text prompt.

[scarlet bronze]

Does anyone have a good flow for re-generating the same ai person other than using the same prompt?

It seems like it confuses ai generated person for real people... Then it's almost impossible to create the same character from a screenshot.

For example I take a screenshot of the video, edit it with nano banana, put it back in as a prompt but it seems to work 5% of the time keeps saying I'm violating policies...

I thought there would be some invisible water mark or something so it can detect it wasn't a real person... 🤷

[rough wing]

Make a character sheet, then use it either as an ingredient, or using NB to make keyframes. (as high def as you can get)

^ not mine

[scarlet bronze]

interesting ill try that thanks

[toxic flame]

do you have a tool to do that ? That's a real pain to change position with midjourney.... The omni overfit so easily the generation

[rough wing]

Yeah, I love the styles you can get from MJ, but it's just awful with details. I got some wonderful artistic scenes from MJ that I'm saving for use in style references going forward, but the buildings in the background are ALWAYS warped if you look at them, and its too much work to fix them.

I just made this today, for one of my main characters, using Nano Banana Pro (in Gemini, but I'll re-do it once the 4K NB Pro gets updated into Flow). I gave it an example of a character sheet, a snapshot of the head and shoulders of the character I'd made already (using 3D software, but that's overkill for most) and cobbled together a prompt. After a bit of back and forth... good enough for government work, as they say.

I'll DM you the inputs if you want.

[toxic flame]

thank you for your feedback, it's really appreciate ! Nice, thank you for the tips I'll give it a shot. For the moment I try to do :

A mini Japanese-style cartoon inspired by Makoto Shinkai and Studio Ghibli featuring my cat as a memento of him.

But it's a credit drain, and I'm wasting a lot of time and credit trying to make something coherent...

I tried:

• “jump to”: but if my cat closes its eyes and I have to continue, it will have different eyes even though I specified them in the prompt. If a camera movement ends on the frame, the next sequence will be weird and incompatible, poorly mixed

-“jump to”: but by removing the background (which helped me on MJ), here I lose 5 seconds of video before Veo imagines a background

And after 3 “jump to” commands, not only is my cat no longer the same, but the visual style is extremely different from the original style

-“extend”: not very convincing, the scene is really inconsistent for my use cases, the cat changes too much

prompt in json: bad idea, doesn't help and adds noise

I tried **ingredients **in fast Veo 3.1 (but not available for quality), it looked promising

What works best, where I have the least waste, is to create a character with MJ, generate lots of frames for the intro scene, and try to build on that. Even if I lose the first 1-2 seconds for the flashy color animation created by VEO 3 to appear, and my chat (created by VEO) has diverged too much from the one on MJ now...

Do you have tips ? Or is it too specific, training dataset too little to handle this you think ?

[rough wing]

if you make a character sheet of the cat in a bunch of poses and angles, showing the eye color, etc., then use it in ingredients, that might be the best option here.

JSON doesnt actually do anything. Get rid of all those brackets and such, and just have an in-depth prompt, but only if you actually need an in depth prompt.

If you start with a good keyframe, then simple prompts usually work better.

[toxic flame]

Yes, in any case, I've given up just now on this project.
I wasted $100 for nothing; the technology really isn't suitable for this at the moment.

The ingredients aren't available for Veo 3.1 Quality, and Veo 3.1 gives terrible results.

The prices are extremely high for something that lacks consistency, with random results that are costly, interface bugs, no seeds, and no character references. In its current state, it's terrible for anime

I'm going to test the competition, Kling and Hailuo. Thanks anyway.

[rough wing]

You're describing problems with all AI video. Veo has been super buggy this week, so it's a bad time to test it, but if the two best in most cases are Veo or Kling.

[keen adder]

Why i am getting "unable to export" error in Google flow when exporting video from scenebuilder i am using google ultra account. any solution?

[toxic flame]

True, but other video models do not apply these prices and are not from Google, one of the largest multinationals at the forefront of technology, I'm trying Kling O1 and 2.6 to see if there's better result

[rough wing]

You get unlimited generations. If you use Veo a lot on other platforms it costs even more. Things aren't magically cheaper because they're a big company.

[pallid canopy]

I think what he means is he is using VEO3 in FLOW and is having trouble with character appearance changing between videos. He needs to use keyframes or ingredients to video.

[mental coyote]

@arctic compass I’ve forwarded your message to the correct channel.

[scarlet bronze]

Can you do this for me too 🙂

[rough wing]

Sure. DM me. The prompt is kinda long.

[scarlet bronze]

Anyone have tips for accents? No matter what I seem stuck on NOrth American Accents :./

[solemn cloak]

I know but still doesn't help when you

[jolly drum]

Cinematic Lighting

[arctic compass]

Struggle with leg consistency/posing

"Hey everyone, has anyone had luck getting Nano Banana to isolate specific limbs for poses? I’m trying to generate an image/video where the character moves one leg but keeps the other planted (or still). No matter how I phrase the prompt or in-paint, it keeps locking up one leg or moving the wrong one.

Does anyone have a specific workflow or prompt structure to fix 'stubborn leg' issues? Thanks!

[rough wing]

Words are too imprecise. You could find an image with someone in that pose and use it as a reference. Or some kind of pose dummy app.

[arctic compass]

Yes got it

[round peak]

Hey 👋
While generating a video in Google flow and when i am adding in scenebuilder it is being smooth but when I am exporting i am getting to many mistakes can anyone help me out.!!!

[loud kiln]

Hi guys. I’m used with comfyui+wan flows and I’m migrating to full google stack. What would be equivalent here? I’m confused where to store my flows, creative sources, prompts. It seems like vertex + gems + my n8n doing part of the work. I can’t get it clear yet. Same for Labs tools that are not consistent (what is working what is not)

[toxic nova]

how do I make square images in flow?

[rough wing]

Can't yet.

[rough wing]

In theory you could prompt for it to have black bars on each side, and cut them out after downloading?

[toxic nova]

not videos?

square videos?

or both not possible

[toxic nova]

ok

[rough wing]

Both, yet.

[supple sparrow]

It’s easy just type it and say only use the main subject in the second image and add it near the road

[hot torrent]

please I wanna ask bout scenebuilder feature when I am using extend the quality of the image is decreasing

how to fix that please

[fluid elk]

Not sure how to follow this but I am desperately looking for a solution on this as well. I've added special prompts, tried using JSON scripts and nothing seems to work

[round peak]

Hey 👋
While generating a video in Google flow and when i am adding in scenebuilder it is being smooth but when I am exporting i am getting to many mistakes can anyone help me out.!!!

Has anyone experienced with this issues ?

[tough parrot]

Has anyone any prompt experiece for these kind of videos https://www.youtube.com/shorts/VZWq6LK7RL0

I tried few but results was trash...

If anyone can give me few hints that would be great, cheers.

[fluid elk]

yes this has happened to me, but I'm not sure what mistakes you are experiencing. Mine where because I was trying to shorten clips in scenebuilder, so I had to export separately and edit in Premiere Pro

[round peak]

I am getting extra cuts.... For say when I am exporting individual clip and then editing in editing software I am getting different last frame for that which is not there in the scenes which i generated in flow

[ocean lynx]

hello all

[rough adder]

I have your solution, at no charge.. no charge this time anyways. You need to use ingredients to video solved..

[formal tundra]

Anyone else having trouble generating speaking videos? They keep flagging everything

[primal folio]

Hello everyone does anyone use this extension for connect prompts video ? https://chromewebstore.google.com/detail/veo-automation-google-flo/fnmijgmnjpealnnadjpjilaanhhambeb?fbclid=IwdGRleAOt93RleHRuA2FlbQIxMQBzcnRjBmFwcF9pZAo2NjI4NTY4Mzc5AAEev0Oac161vVoXvVHIPVViDtab8DYXIbdvSRfYvcvoNlOnXN0Ec8vS3L5dvao_aem_PKf2G4lAEbnH_ljf2B4Pwg&pli=1

[modern surge]

When using Flow’s “Extend” feature, image and audio quality often degrade noticeably.
Does anyone know any tips or workarounds to overcome this issue?

[glacial shore]

Extend is short way to generate video for start frame and end frame feature.
I have an idea, it likes:

1. Start image -> Edit Image (with banana pro) -> get End frame
2. Generate video: Start frame and End frame
3. Continue to edit to get End frame + 1
4. Loop 2: Generate video: End frame and End frame + 1

Loop.

DO you think Flow extend feature did the same strategy ?

[modern surge]

Thank you for your response.
Continuing a video in that way seems like a good strategy.

However, in the case of audio, there is an issue where mechanical or synthetic-sounding noise gets mixed in when using the extend feature. In addition, even when generating a video using start and end frames, the audio consistency is sometimes not maintained.

[glacial shore]

Understood. Try to describe very detail for: audio_cues, audio_track in your prompt. You might try with the same audio_cues, audio_track in many extend prompt.

[modern surge]

Do you have any best practice examples that you’ve found effective for maintaining consistency?

[glacial shore]

fully json prompt from veo-meta

[modern surge]

Thanks. I’ll review what you shared and try moving forward based on that.

[glacial shore]

You're welcome, just remove some fields you think that it is not necessary

[rough wing]

Pretty much the same tactic for everything when you don't get a result you want at first: Look at your prompt, look at the result, add more detail to, and generally just try to improve, the parts of your prompt about what is missing.

[modern surge]

Thank you for the tips!

[rough wing]

Hit me up by DM if you have any other specific prompt issues, or here.

Since Extend doesn't allow for an end frame, you only have a prompt to work with. Upside of this way of doing things, you're basically practicing prompting, so you'll begin to learn how the model interprets variations of word and image combinations differently, and your prompting instincts will improve.

[lean oriole]

Hey im new any tips

[glacial shore]

I have thought that, user input a prompt to Extend then system will generate a new image base on some last frame of current video called end image, then system run [start frame] and [end frame] feature.

[rough wing]

Extend is based on the last second, not just one frame, so it captures the movement without tha stutter you get when you use start and end frames.

[fallen spindle]

Why why why why does flow insist on putting this guy in? Every time it's inventing this guy What's the black shirt and dark beard. who is that? Why is it not following audio instructions? This is really frustrating.

Tried variations of prompts such as: The camera slowly tracks in closer and closer to the pile of money sitting on the sink, as a gruff grovely character voice in another room with a thick exaggerated New York accent says, "I'll take this stupid microwave too", followed by the loud sound of an appliance crushed, and then the sound of some of its pieces falling on the floor along with the sound of huffy exertion as fast footsteps carry the voice even farther away mumbling hurriedly, "Merry Christmas, yeah, bye" as the sound of receding footsteps continue. Negative prompt: Never during this entire sequence do you see any person in the bathroom. The only visuals are the camera tracking towards the money on the sink. Nothing else moves or changes in the bathroom, everything is static. do not generate a person anywhere during this sequence. The bathroom remains empty, and the money remains unmoving on the sink. The money and all other things in the bathroom stay stationery, they don't move at all. All the audio happens out of view. you will never see a person beginning, middle or end of the sequence.

https://youtu.be/xyu9W5Z1Iy8

[fresh valve]

Does anyone know how to remove an object from a scene in a video from a later fram in the video? When I click on edit video it only allows me to remove something from the beginning frame. I need to remove something from the 5sec mark.

Thank you in advance.

[terse thicket]

not sure but i often just crop it in a video editor

or regenerate witht he same prompt

then you can splice the 2 generations if nothing else

[steady python]

Is it possible to change the camera direction like curve motion by focusing on the object at the middle

[terse thicket]

i have been trying to get a bullet time effect for weeks. but a lot of it boils down to prompt craft. "the camera slowly pans left to right focusing on the object with the background out of focus"

[limber goblet]

Does anyone know how to recreate this effect

[terse thicket]

"buildings collapsing in on themselves in the style of inception" might work

[sand sandal]

hey guyz, really need some help, i wanna generate a realistic vlog video with the given character in the main frame and should be very close to camera and should voicover and proepr lyp sync and proper scene changes in the background... need help in this facing a lot of issues i have wasted more than 1k credits in this

[terse thicket]

might be worth doing the "voices" as seperate generations and worry about first/last frame

i cant share links here but we used that method for a video we just released

feel free to dm me if you want an example

[sand sandal]

that would be awesome, please check your dm

[rough snow]

Guys, how do you create a prompt for the right dialogue? Sometimes it's wrong that the first person speaks and the second one speaks.

[indigo ridge]

hi, how we save cadit on flow

[terse thicket]

learn to edit video XD

[odd birch]

@lusty cipher Here is your forwarded video and message. This channel is the appropriate one for prompt help.

[lusty cipher]

Thank you. Highly appreciated 👋

[thorn crow]

Hey guys anyone can help me , how can I generate banana pro images in flow in a 4K resolution?

[ivory crow]

Does anyone know how to precisely call camera movements? When I use the prompt to perform an "orbit up" movement, it doesn't work the same way as when I select the option within the "edit" menu. There, it executes perfectly. (That edit menu only appears in the Ultra account.)

[ivory crow]

From what I understand, when you download the image, the 4K option appears, but it's upscaled. It's not generated natively in 4K.

[rough wing]

Text prompting camera control in Veo usually requires a bunch of iterations. It's not precise.

[ivory crow]

Thank you, I thought that there might be some type of command that executes the movement order in a direct way. This is how it happens in Edit

[wise crag]

I have a clip with a black spot that I really need to brighten up. Does anyone know how to do it in davinci resolve? somone help

[marble cedar]

Mask it and then increase the highlights and feather according to your clip.

[wise crag]

i m tring but i m doing it wrong

[rigid ridge]

Thank you. Highly appreciated 👋

[rough wing]

Same basic camera commands (zoom, pan, orbit, etc.). It's just a blunt instrument.

[silent meteor]

Hi all. Old guy here burning through credits trying to get dialogue correct between two characters. Any hints?

[burnt star]

fully json prompt from veo-meta

[devout canopy]

How can I create video more than 8 Seconds in Flow

[molten parcel]

You don't, you create multiple 8 second clips and put them together. If you want to do this in the flow site hit "add to scene" go into scene builder and make another 8 second clip and it will have it together.

[prisma thorn]

Did you have any luck? I am starting to try this out.

[verbal bolt]

Before jumping in, think of a story or what you want to see. Then create some reference images and build up layers of images. Then, when ready, use the references and clear prompts to create what you want.

[prisma thorn]

I am using Opal. How do I send the text script to Eleven Labs in order to get the audio file? This is where I am stuck now.

[verbal bolt]

Given segments are only 8 seconds long, it's perhaps not too much to actually write out your script and that's what you use in Elevenlabs, it's what I do.

[vagrant grotto]

For my game mathventure I am trying to create either full on animations, or just nice frames I can use as ingame animations.
These are made with simple prompts like "attack animation", but I see a lot of morphing etc.
I would like to use multiple characters at once to get more useful stuff for my credits. But so far I get a lot of nonsense and it is too expensive for me to learn the tricks in prompting from scratch ...
Has someone a good resource with prompting tricks that are likely to work for my scenario?

[pure zenith]

Why do fine architectural lines shift between frames in Veo 3.1 image-to-video, even with a static camera?

When generating video from a still image in Veo 3.1, I’m seeing subtle line crawling / banding-like artifacts on fine architectural details (railings, beams), even when the camera is completely static and no motion is requested.

Two consecutive frames are visually different despite no intended change.

Is this expected behavior due to frame-by-frame re-synthesis, or is there a way to enforce temporal stability for static image-based videos?

The issue persists even in static scenes.

Is this a known limitation of Veo 3.1’s image-to-video pipeline (frame re-synthesis), or is there a recommended workflow or setting to enforce true temporal coherence / pixel stability across frames for static images?

Any guidance from the team or other users would be appreciated.

[proud jetty]

might be able to find something here...use the filters to narrow things down https://youmind.com/nano-banana-pro-prompts

[vagrant grotto]

Thank you - seeing a few examples that apply to my usecase, but they also make a point, that currently the lootbox factor is still too high to be of use to me :D. I guess for now it is NBP image gen and not video 🙂

[proud jetty]

Use a tool like TexturePacker or a simple "Sprite Sheet Slicer" online to turn the large image into individual animation frames for Unity, Godot, or Unreal.

[proud jetty]

thats how i made those 4

actually, i just took the 1st frame, used the prompt and the last 2 steps from the keyframe screenshot

since i didnt need it to be 100 perfect

[vagrant grotto]

ahh - I have no problem doing that, but also see that these do not turn into a clean animation ...BUT ... they are still awesome and I will work with exactly that kind of approach. fluid animations are optional for my use case, so I just need a 2-3 different attack frames and 2-3 different 'getting hit' frames. Thank you for sharing the prompt, really appreciate your input here! 🙏

[proud jetty]

np... could increase the prompt to 24 or 30 frames to maybe makeit cleaner

[vagrant grotto]

I have tried a ton of those approaches ... you can modify existing animation sheets and insert your own characters, you can't create clean ones from scratch ... at least not yet!

[proud jetty]

Yea...i used it to make sprite sheets for a Jaws parody im making on the side....luckily, sprites are alot easier to edit

[lapis pond]

this isnt to hard to achive you have to use extend video to make longer clips. thats it, then you can take the sound, thats often is bad and change the voice in elvenlabs, an example: https://youtu.be/5VIfkMbztYg?si=NsOsfmkC6W12FIXt

[modern surge]

I’m exploring whether time-segmented scene control is supported when using Veo 3.1 through Flow.

Since Flow outputs an 8-second video by default, I’d like to know if it’s possible to specify different actions and camera movements for distinct time intervals within that single clip (e.g., 0–2s, 2–4s, 4–8s).

Does Flow expose any mechanism—prompt-based or otherwise—to guide temporal scene transitions, or is the recommended approach to generate and stitch multiple clips?

[tight burrow]

all this can be done with timestamp prompting

[ 0.00 through 0.02 ]
[ 0.02 through 0.04 ]
[ 0.04 through 0.06 ]
[ 0.06 through 0.08 ]

but i advise to use only 3 timestamp

[modern surge]

just put this text in the prompt? Oh I see. Thank you for your tip!

[tight burrow]

[ 0.00 through 0.02 ] prompt
[ 0.02 through 0.04 ] prompt
[ 0.04 through 0.06 ] prompt
[ 0.06 through 0.08 ] prompt

[modern surge]

So if it’s 0 seconds, it’s 0.00, and if it’s 2 seconds, it’s 0.02, right?
Not 0.00 and 2.00.

In that case, do you know whether it can also distinguish milliseconds?

[tight burrow]

yes , no milisecond

only second work

for better continuity use 3 timestamp

[modern surge]

Totally understood! Thank you @tight burrow

[tight burrow]

[ 0.00 through 0.03 ] prompt
[ 0.03 through 0.06 ] prompt
[ 0.06 through 0.08 ] prompt

this is better

[tight burrow]

enjoy

[marsh otter]

guys can you tell me how can i fix this problem

[odd birch]

Hi there, please do not ask the same question across multiple channels. That is considered spam. We have the #1238621511401734254 where you can post questions like this and the team reviews those threads. Thank you!

[rough wing]

It doesn't work reliably, but this came from an official Google blog post about Veo:

[00:00-00:02] Medium shot from behind a young female explorer with a leather satchel and messy brown hair in a ponytail, as she pushes aside a large jungle vine to reveal a hidden path.

[00:02-00:04] Reverse shot of the explorer's freckled face, her expression filled with awe as she gazes upon ancient, moss-covered ruins in the background. SFX: The rustle of dense leaves, distant exotic bird calls.

[00:04-00:06] Tracking shot following the explorer as she steps into the clearing and runs her hand over the intricate carvings on a crumbling stone wall. Emotion: Wonder and reverence.

[00:06-00:08] Wide, high-angle crane shot, revealing the lone explorer standing small in the center of the vast, forgotten temple complex, half-swallowed by the jungle. SFX: A swelling, gentle orchestral score begins to play.

[modern surge]

Oh, thank you! I’ll need to take a closer look at that blog. As always, thanks a lot.

[rough wing]

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

[mental coyote]

Hi @forest valve 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!

[warm coyote]

flow thinks this sexual content. i cant animate it. any tips?

[upper badge]

u must change the clothes buddy, it is considered an explicit content.

[marsh forge]

Hello, do you have an idea to make a fixed shot, the generated video always shows movement!

[rough wing]

Try starting your prompt with 'Static shot of...'

[marsh forge]

I tried and nothing works x) it always moves forward

[rough wing]

You'll still get movement sometimes, but it works. The prompt expander they use screws up simple prompts sometimes, especially when you dont want much to happen.

[coarse wharf]

Please try to explain that when we generate a video, what do we have to write in the prompt to make the voice consistency of the character, what kind of prompt do we have to give, especially For character voice consistency.Please try to tell me as soon as possible.

[rough wing]

If you want voice consistency, you'll need to use a third party service, or just keep generating until you get lucky.

[coarse wharf]

Still you can tell me what we can write for veo 3.1 audio prompting follow where can understand the language of veo 3.1 prompting

Because I always try to avoid the need to use any third party AI, a little bit will do but sometimes it frustrates me, that is why I was asking you.

If you could suggest even a little bit about audio, it would be better for me to know this and it can be useful in video generation.

[stuck folio]

if the videos are not too long, in my case under 1 minute extend video feature pretty much nails it, the voice consitency, I can only imagine it will get better from here

[spark notch]

Question
Im unable to download a short video fron flow scene builder.
Anybody know how to download video from it

[versed wave]

Hello, i'm trying to do some videos and i'm doing my best in prompting and explaining everthing but if i did multiple videos with the same character it's ruined with the movement and the face and everything, would be great if some can explain to me how to avoid that as any project i'm doing is getting ruined.

[vestal moss]

Hello 🙂
I’m currently trying to generate videos in Flow using the AI tokens from my Google AI Ultra plan. However, I’ve noticed that asset-based video generation doesn't seem to support the 9:16 aspect ratio.

Does anyone know a way to create 9:16 Image-to-Video content using these plan-based tokens instead of using a separate API key?

Note: I’m looking for alternatives to the direct Gemini interface, as the generation limits are too low for my needs.

[spare carbon]

it support. who say it doesnt ? you dont see triger to change orientation ?

hey people,
If I have photo of people holding hands in a circle,
how do I explain the camera movement so that it circles around these people ?

[vestal moss]

When creating a video using assets, if I set the video to a 9:16 aspect ratio and click the submit button, an alert appears saying that only landscape mode is supported.

I know it might be hard to read since it’s in Korean, but that’s what I’m seeing. I have two Ultra plan accounts, and the same issue occurs on both.

[spare carbon]

does image is in Vertical or Landscape ? prepare it berore generating video

[vestal moss]

that image is Landscape but I already tried Vertical before, but didn't work for me..

I just tried Vertical and it still doesn't work

[coarse wharf]

Still you can tell me what we can write for veo 3.1 audio prompting follow where can understand the language of veo 3.1 prompting
Because I always try to avoid the need to use any third party AI, a little bit will do but sometimes it frustrates me, that is why I was asking you.
If you could suggest even a little bit about audio, it would be better for me to know this and it can be useful in video generation.

[vestal moss]

The feature shown is 'Frames to Video', but I think I'm referring to creating a video using assets.

[coarse wharf]

Still you can tell me what we can write for veo 3.1 audio prompting follow where can understand the language of veo 3.1 prompting
Because I always try to avoid the need to use any third party AI, a little bit will do but sometimes it frustrates me, that is why I was asking you.
If you could suggest even a little bit about audio, it would be better for me to know this and it can be useful in video generation.

[devout bridge]

@ocean tree hi! Did you solved the problem with downloading? I'm in the same situation. If I click to download the videos in 1080, it takes too long to upscale and download. Did you find any solution?

[rough wing]

That's not broken. Upscaling just isn't a fast process.

[meager sandal]

i want to create a ghost rider animated video how can i do please help

[fathom epoch]

Can VEO3.1 fast generate POV shot video?

[long wraith]

I'm having the same issue. I'm trying to convert my story into a video, but I can't seem to keep the characters consistent.

[versed wave]

man i'm starting with character and after just 3-4 scenes it's get ruined it's become another character even the lighting and everything in the video is getting differnet and changed

[coarse hill]

have you tried the T-Pose paper-doll method?

[coarse hill]

I have issues when it comes to weapons. My character has a VERY particular weapon and the way it operates and is wield the model can't accurately replicate

[versed wave]

i'm doing normal videos not even a lot of effects or some thing very high but it's keep getting ruined after just couple of scenes

[versed wave]

would be great if you can share it

[coarse hill]

its pretty tedious but I ensure I have my Character in a pose, with their arms stretched out. I pose them Front, back and two side profiles. then I create a short video of them turning around. I also make sure that they're on a "green screen,"

or a solid background.

Things will still get ruined but its a little more consisent , but you also gotta think in about 6-8 second intervals.

[versed wave]

yeah that's is good method i think i didn't try it before

[versed wave]

ah okay so no point of it

[coarse hill]

no there's a point. There's always going to be "drift" as they call it. its invevitable.

[versed wave]

i think Ai didn't get this advanced

[versed wave]

understood i think but this method can be done every video ?

i thinked about method which is just create 1 scene and then use it 2-3 times and then create new one in the same video

which should be good and give good results

[long wraith]

I did. I managed to do it once. It picked up the right ingredients as I uploaded them. But they do extra movements that make them look weird/out of place. And it's for a romance story, so yeah, it won't work.

[rough wing]

If you have a good character sheet, then use that to make keyframes or as an ingredient in Veo, you can usually get consistency.

[versed wave]

How i can make good character ? without losing the face the movement or giving it 3 or 4 hands?

[rough wing]

"character sheet" like this:

Also works for things

[versed wave]

Understood where i can make it or what tool ?

[rough wing]

Nano Banana

*Pro

[versed wave]

Wow no way man you are doing great job with it

[versed wave]

how you export it in this

what prompt you are using

[rough wing]

First just make the character. I make recurring characters in Character Creator 4 (a 3D modeling app), but you can do it in Nano Banana. You need to start with a good image of them, looking into the camera.

[versed wave]

yeah man you are expert

[rough wing]

"Generate a fully photorealistic character design sheet based on image 1, using a layout like image 2, with sections showing three full body views (front-view, side-view, back or rear-view), one up-close shot of her shoulders and face, and facial expression examples filling the rest of the sheet. Use the first uploaded image for the face of the character, but have her wearing a black turtleneck and black pea coat."

[mossy moat]

is veo 3 working smoothly rn or nah

[versed wave]

bro how many years experience you have

this is insane

[rough wing]

Nano Banana was invented this year, haha

[versed wave]

i mean i'm new to this it's just been few weeks

[rough wing]

But I've been playing with AI image apps since 2021. Dabbling with 3D for years before that.

[versed wave]

yeah as i said you're expert

[mossy moat]

its a prompt u can get from AI just need to phrase it well

[versed wave]

i mean the character thing as i never managed to do 1 character before with this realistic details

@rough wing can i send you dm?

[rough wing]

Once you have that character sheet, you can use it to make that character in any pose and scene to make keyframes out of for 'Frames to Video', or use as ingredients in 'Ingredient to Video'.

[rough wing]

Of course

You'll never get 100% consistency in your generations, but depending on how specific your aim is, you can usually get what you want within a few tries.

[versed wave]

understood

[versed wave]

check dm please

[versed wave]

yeah i know that

[digital spear]

Yes sir

[low oracle]

your name remind me minecraft bro

[digital spear]

[rapid sparrow]

I'm agree with your statement,it's always a small mistake

[digital spear]

Yes sir

[versed wave]

Oh okay

[digital spear]

Supp bro

[rapid sparrow]

Ender man,ender dragon

[versed wave]

XD

[versed wave]

I'm doing great

how are you doing?

[digital spear]

Good sir, u?

[versed wave]

doing good hope you have some good projects bro

[digital spear]

Good luck with your project, bro.

[versed wave]

Thanks man u2

[digital spear]

U too bro

[rapid sparrow]

How are you man

[versed wave]

I'm doing great

how are you doing?

[drifting mauve]

Any ideas on how I can get the bottles labels to be clear and readable?

[rough wing]

Have only one or two of them, and make them bigger.

[drifting mauve]

They're 10ml bottles 😬

[gentle grove]

Has anybody had difficulty with finding download locations?
Download locations for VeoFlow.

[iron fern]

why flow DROPS the quality of video ,when using the extend feature .it drops the visuals quality .Anyone else noticed this problem.can anyone guide me what should i do?

[molten parcel]

Use frames to video

You can even try ingredients if you are trying to get a consistent character but I wouldn’t use text to video at all for extending.

[stuck folio]

you just gotta do a couple rolls eventually you get it in my expeerience.

[low oracle]

Maybe more specific with the prompt

[rough wing]

You'll get fewer issues if you have fewer, and zoom in a bit so the bottles are bigger. Fine details get fuzzy with diffusion noise.

[rough wing]

Extend often loses consistency, especially if the character looks away or something, but overall quality tends to stay roughly the same.

[spiral cave]

Any luck with fast paced action scenes? Feel like it’s hard to get something that feels natural/exciting

[clever nova]

Hi @dim pulsar, we wanted to remind you of our server guidelines, stated in #start-here・📍 ⁠: “Currently, our team can support engagement in English; the discussion in this server will largely remain in English for now.” Please keep in mind that non-English discussions and content may be removed. Also this channel is for the Today’s special. Please visit the #server-directory・☎️ ⁠for a channel guide. Thank you!

[rough wing]

Speeding up a clip can be done in editing. Slowing is a bit harder, because you cut the frame rate down and if you slow it too much you can get choppy.

[spiral cave]

Yeah usually i just generate different angles to mask certain frames. Ive gotten a bit of results by just targeting 0-1s as the action shot and then 1-8s as all black frame. So you get a clean action, but way too much post editing

[rough wing]

I'm not sure what that means, but couldn't you just tell it to do a lot in those 8 seconds, then if it doesn't happen fast enough, make an 8 second clip 4 seconds?

[spiral cave]

Take a fight scene for example, if you allow veo to use all 8s it typically just fills in the rest of the fight scene, lets say i give a description of a singular kick for example, it tends to fill the rest of the scene with multiple other stuff i dont want/need. So by giving it directsion of a kick only for 0 seconds to 1 second and fill the rest of the scene with a blank black frame. you get the clean kick shot youre looking for

veo isnt quite there with good choreography yet in my experience but seeing if anyone has good experience.

I tend to use json formatting for scenes but found for action shots specifically natural language works best

[rough wing]

JSON doesn't do anything, but structured prompts with the same text can be helpful when you need that level of detail. I usually spend more time making the input images, and animation prompts are less complex.

[spiral cave]

JSON for me just allows me to structure stuff a bit better for organization
yeah reference images are good, i use way more now with ingredients it really helps for consistency

[rough wing]

I know what it's for. Take all the code bits out and keep the text, and you get the same substance, without all the extra punctuation. Same result.

I asked a while back. The model ignores all that non-text stuff. It understands the organization and substance of the text.

[rough wing]

Ingredients has been getting better and better. Was pretty buggy at first.

[warm coyote]

json prompts are like christopher nolan films. just pretentious nerd stuff. does nothing. just type what you wanna see.

[spiral cave]

Yeah !

[rough wing]

Structured prompts can be great, if more simple prompts can't get you there, but all those brackets and whatnot are extraneous.

[versed wave]

@rough wing To be honest, this man helped me really, really much. I would recommend anyone to mention him if you want to get help.

[mystic wedge]

does anyone know of a way to get flow to render things on an alpha channel?

[silk dock]

How can I make a Pokemon NatGeo style video? There's a lot of different YouTube channels that do this and most videos look like they could have been made by the same creator.

Here is an example. I checked the SynthID and other similar videos by other creators and they were made using Google Veo
https://youtu.be/oGlQvFnDMZY

[warm coyote]

so veo cant do super slowmotion right? it ignores the prompt

[rough wing]

It uses a prompt expander in the background that alters prompts. It's very hard to get it to do subtle movement, including slow-mo. It always tries to add motion you don't ask for, for some reason.

[warm coyote]

a big model like veo should be able to do superslowmo. what is the problem here. if i prompt a guy jumps over the roof in superslomo, it should just do it. kling does that everytime. but overall kling is just too expensive to use on daily basis. i want veo to do it damn

[mystic wedge]

👍

[harsh inlet]

How to use this ?

please me to how to whisk general in discord ?

[sleek marten]

Just tell it to put something infront of a consistently lit green screen with hex 00FF22 should work depending what you have in mind

[potent cliff]

You could always export and change it in Resolve?

[rough adder]

Json prompts don't get rejected as much. Higher level of getting success, no matter the content

[rough adder]

It's called Hallucination....does it a lot, thus the repeated attempts to find a good one

Good news, the next upgrade is far superior! Yeah! 😆

Takes me about 10 attempts to find a good one

[rough wing]

That's usually what I use for editing.

[wise kite]

is there a way to get an integration of n8n into an opal app or firebase project?

[rough wing]

This is a channel for Veo Flow

[ember finch]

I want to generate an add for my product, can anyone help to tell how can I write the prompt which give the desisre resuslts

[wise kite]

Yes...sry I just realised that.. I'll send it there