flow-general・🌊

[wet fern]

Please fix this bro

[woeful void]

Hey guys, why are photos easily edited in Google Gemini the first time,
but in Google Flow I have to repeat the same thing 10 times using different methods,with no demand result ?

Google Flow doesn't understand how to edit a photo, even for something as basic as this.
What kind of glitch is this ?

[wet fern]

NOT only that it cant even generate veo 3.1 fast videos using reference images

[mortal summit]

Each time you make an announcement I can’t find the link to event or video, it’s as though you intentionally fail to provide the direct link? Would you please focus on the task and share complete information and make a clear link to a class, video or presentation?

This is BS

[wet fern]

these mfs in announcement talking about partnership while we stuck here 🥀

[sinful lotus]

Is it just me that veo lite says "10 credits" (I thought the announcement said 5) and of course won't work with ingredients (bummer).

[gleaming lynx]

for pro they did 20 for fast

and 10 for lite

[edgy holly]

Gemini 3 thinks about your prompt before sending the prompt to Nano Banana.

[woeful void]

How to solve this issue?

[edgy holly]

It's not broken. They're different tools. In Flow you're prompting the nano banana. In Gemini you're prompting Gemini, then it interprets it before sending it to Nano Banana (you can see it "thinking" sometimes).

[glad yacht]

It has too many sensory filters. Many legitimate prompts and edits are refusals.

[crisp basalt]

as i was saying

i dont have anythig to say

[rapid sphinx]

Nano Banana 2 or Pro?
Can the speaker cover it??

[sour belfry]

Excellent masterclass session, thanks for sharing great tips!

[fervent parrot]

How do you handle when Flow continues to ignore direct prompts to not do something, such as, "The character is standing still, do not create any animation movements"

[gusty nimbus]

i use nano banana pro, though 2 is faster

[main saddle]

tysm for this feedback, and for joining!

had a blast

[crisp basalt]

by not mentioning it

[fervent parrot]

I mention it because it will produce an animation if I don't. It's a catch 22

[elfin nexus]

thank you very much @gusty nimbus and @tacit halo !!

[gusty nimbus]

this happens to me sometimes and it can be super frustrating! first step is to adjust the text description. such as "the character is paralyzed" or "the character is frozen" using different adjectives and states can go a long way

you can always use LLM like gemini to help you come up with a different type of description

[crisp basalt]

or prompt it stands perfectly motionless

[gleaming lynx]

going back to the technical and lighting extracting question

I prompted gemini to analyze this image

using the output prompt it generated this image (i dont care about the objects and their placement)

its just that the lighting is no where near the reference image's

[fervent parrot]

Other examples are "Don't produce and audio" because it will automatically otherwise

[crisp basalt]

which one is generated

[gleaming lynx]

second one

[crisp basalt]

let me try

[gleaming lynx]

I really have been trying for a few days but cant prompt it to extract the lighting style and the camera etc

[gusty nimbus]

i think the lens flare is the biggest thing changing the image, so definitely negative prompt away the flare. let me look at my extraction prompt

[wary ledge]

For those interested, they were done a few months ago but am doing fresh deep dives now and will share when done

[fervent parrot]

I've tried a million variations 🙂
The failure is not for want of trying. I just hate burning up credits on trial and error.

[gusty nimbus]

who wants me to post my style extraction prompt here?

[gusty nimbus]

it can get super picky sometimes, i would say that half the time there are some flagged terms in the text prompt. other half there's something in the input image that's the problem

[edgy holly]

I've tried so many variations of that, and never gotten it to work. Blew through several dozen gens once, got frustrated, hopped over to Kling and it got it right the first time.

[gusty nimbus]

so start with the text adjustemnt first bc that's easier

[edgy holly]

It's just one of the things Veo isn't good at. Each model has weaknesses.

[spiral loom]

Quick question - where will the recording of the masterclass be posted? So good! But I missed the beginning 😭

[crisp basalt]

well

dont ask me whos that man

[glad yacht]

I try to restore old family photos that's 60+ years ago. Flow keep rejecting uploads, minors detected... Warning ⚠️🥲

[full flicker]

That hasn't been decided yet. Keep a watch though. It will definitely be posted.

[gusty nimbus]

https://docs.google.com/document/d/1edDgquVcyGtpsQgNI2apsOYr_K9NOa3n-AYvsJWJ2to/edit?usp=sharing

[spiral loom]

@full flicker thanks 😊

[edgy holly]

I tried that with the first Nano Banana. Worked in Topaz, though.

[gusty nimbus]

definitely helpful to test out a different model if it's not working for a specific shot; i use LLM/gemini to convert my prompts so it meets the character limits of other models

[nova bay]

How do we enter the event or are we still waiting?

[edgy holly]

Just ended. They said they'll post a recording later.

[nova bay]

I thought he mentioned another

[glad yacht]

Sometimes I upload my selfie for editing, it would trigger flow eminent person detected 🤣

[wary ledge]

Your roles are conflicting in my opinion you are telling the AI model to be a human role and a AI

[gusty nimbus]

feel free to edit it and test it out, you might be right!

[gleaming lynx]

Yo thank you dude

I will play with this

[fervent parrot]

Thanks, I mostly work in tandem with Gemini and Flow, using Gemini to guide me, to analyze flawed output, or help craft my prompts. I'm new to using these tools and figured that was the best way to learn and improve. It's a great suggestion.

[gusty nimbus]

you can upload video to Gemini along with your original prompt and prompt it to analyze why the output failed

[gleaming lynx]

Did you just put the image in and ask it to generate this exact same image or something?

Cause i want to use this cinematic setting, the lighting and the camera style as the main theme for a very different video.

That's why I need a text prompt that works.

[crisp basalt]

try Single directional sunlight entering from a side window at a low angle, casting elongated soft-edged shadows across the bed. Strong highlight-to-shadow contrast with rapid falloff. Warm golden highlights contrasted with cooler ambient shadows. Subtle volumetric light rays and atmospheric haze creating bloom and light diffusion. Light wraps softly around subjects while leaving parts of the room in deep shadow. No global fill lighting.

[gleaming lynx]

thank you i will try this out

[crisp basalt]

or try : A vintage, cinematic aesthetic with a warm, desaturated color palette dominated by deep amber, sepia, and muted earth tones. The lighting is soft, low-key, and predominantly artificial, suggesting an indoor evening or a deliberately underexposed scene, with strong directional light creating long shadows and dramatic contrast. There is a noticeable film grain or texture applied, giving the image a retro, analog feel. The composition is presented as a triptych, or a vertical sequence of three cropped shots. The overall mood is intimate and moody, achieved through the low fidelity of the visual rendering and the heavy application of warm color grading.

[gleaming lynx]

ok

how did you prompt it?

[gleaming lynx]

i mean what prompt you wrote to get this prompt

[edgy holly]

You're secretly a Kardashian. I won't tell.

[gusty nimbus]

using [image 1] of man typing on laptop in bed as base image

recolor and change the lighting of [image 1] to match the visual aesthetic of [image 2[ multi-image

remove teh lens flare

negative prompt: no lens flare

[VISUAL STYLE]Film Stock: Analog 35mm motion picture film. Moderate organic grain. Vintage 1970s aesthetic.
Lens: 35mm prime lens. Deep depth of field. Subtle edge softness.
Lighting: Deep ambient shadows. Harsh highlight roll-off.
Color Grade ("Vintage Amber Cyan"): Heavy teal and orange split. Cyan-tinted shadows. Lifted milky blacks. Warm amber highlights.

[crisp basalt]

i used whisk actually

[edgy holly]

Break his prompt into sections. Keep notes on those different sections. When you don't get a result you like, try different synonyms and only change a few words. Keep notes on which words and phrases get better results.

Those results change with each model update, so you have to keep changing your notes.

[glad yacht]

Has anyone imitated mj style successfully with flow?

[gleaming lynx]

oh yeah this is wonderful

[crisp basalt]

yes

[gleaming lynx]

noted thanks man

[glad yacht]

Can we see some examples

[edgy holly]

That habit is by far the most important thing you can learn if you want to progressively get better, and it's free. Just find some notetaking app you like, and use splitscreen to run it in a sidebar (I use Notion), or in a seperate window if you prefer an app (like Obsidian).

[gleaming lynx]

will start doing that

[wary ledge]

@gusty nimbus i rewrote your prompt https://docs.google.com/document/d/1T7bs8UytDhFbX_Q0wk9JGD2HdeLhGxsF1O8WSIt9M1Y/edit?usp=sharing

[edgy holly]

Unless you really set up something like a notebookLM or Claude Project with a bunch of examples and prompt notes on a specific model, don't rely on LLMs for prompting. They give a lot of bad advice.

[gleaming lynx]

yeah after a while I went into the loop of asking gemini to update the prompt blah blah

and ended up with something completely differnet and useless

very long too

[gusty nimbus]

i have a claude project for prompting and it works just about as well as giving a template to gemini, but i think it's because Claude vision isn't as good as Gemini vision

i bet calling Gemini vision inside of a Claude app would be the best of both worlds

[edgy holly]

Yeah, they don't have secret information. They just go look online, and what you find online is often bad, or outdated for old models, or for different models entirely. It's a mess.

Other platforms have good instructions. For some reason, Google never does that (not just AI).

[gusty nimbus]

this is great

[sinful lotus]

Ah, got it...thanks.

[edgy holly]

If it's just for prompting, you wouldn't need vision capabilities. You'd need to collect some in-depth tutorials that are actually good (hard to find), and they'd have to be model-specific (not Veo 2, or Veo 3, but Veo 3.1 only). Then as you use it, save generations that worked well, build an archive and re-upload it periodically as a source file. It will see patterns.

I don't have one for Veo, but I have one for Suno in NotebookLM like that. Took me months to get it to work well, but it does not (for v4.5 and v5, anyway).

[gleaming lynx]

Yeah I think I can finetune it to make it work

[sinful lotus]

Hi again. The last few days, under Pro level, I've been getting failure rates on Nano banana 2 up to 90% failure (something went wrong) in the afternoon, like now. Veo 3.1 has been failing a lot, too. Even simple prompts... just checking to see if there's something I'm doing, or it's just in a bad mood as it gets sometimes. 🙂

[gleaming lynx]

mine has been working fine

did you write to google?

[regal spoke]

Is there a website where I can get ideas for making a commercial video for a car dealership? Or has anyone here worked in this field?

[cyan terrace]

O veo 3 n esta gerando vídeo

[sinful lotus]

The usual place to give feedback isn't there (like when it fails for a defined reason). This error I'm getting says "failed. something went wrong" (most of the time immediately)...and you can't "give feedback". I have tons of black windows. Upscaling fails from time to time, too (just did, then worked the next time). This goes on for several hours and then starts working again. Prompt complexity doesn't matter. Hmmmmm.

(sorry, prior was about huge numbers of failures on Nano Banana 2 and veo 3.1)

[pulsar coral]

Is that a Dr. Who

[pulsar coral]

Yeah I think upscaling it and putting at Whisk could work too, basically same thing

[pulsar coral]

If it starts working again later, then your account just went out of video/image generations for the day earlier, Flow just doesn't display the correct message for it.

And yeah that counts for the upscaling too, from what GPT told me Flow also uses the Nano Banana 2 or Imagen 4 model to upscale images, it does have a daily limit too, which will reset around the same time you generated an image successfully too.

The upscaling model is more specific or uses certain prompts tho, it's also at Vertex somewhere.

But yeah my brain went out of coffee credits yesterday when trying to find the exact one

So yeah point is Flow leaves you guessing what's going on saying "Upscaling failed" for hours and hours

[sinful lotus]

So you're telling me even though I have 2,500 credits I bought (and let's say I have 2100 left), there is some magical threshold it hits that it says "no more today"?
Or maybe it's a "per hour" thing? Because I just generated another couple of nano banana 2 images...and 2 veo prompts after all the failures last few hours....

(and it seems to be getting better)

[pulsar coral]

If you were generating some images fine some hours ago, yeah. It does have more than magical, an invisible wall.

Unless it's been going on for like 48 hours straight, then you'd better report it at labs-bugs-feedack

[sinful lotus]

Just the last 3 or 4 days.

[proper drum]

That only happens with first and last frame… If I use ingredients the video doesn’t have transition

So isn’t something about training data

[pulsar coral]

If it happened at 24 hour intervals or so, it'll be fine later again.

[sinful lotus]

Yes, it will be fine in the morning (same thing every day last 3 or 4 days), it follows the same pattern

[pulsar coral]

Oh right, if you were using Nano Banana Pro that one has even a shorter limit, might wanna check if Nano Banana 2 isn't working either.

[sinful lotus]

I've only used nano banana 2 in the flow screens (haven't tried nano banana regular)

[pulsar coral]

Pro just uses more juice and it's like %50 more expensive for Google

I don't like it much for some reason, I think it pedals to the metal a little too much

[sinful lotus]

Ok, so if I'm doing a lot of nano banana 2 test "stills" (before I generate a video)... I'm taxing the system and it's going to start throttling me (even though it says nothing of this).

[devout ledge]

The revised prompt refers to 4 output fields, but I don't see them defined. Is your revised prompt meant to be followed by the first version?

[sinful lotus]

I asked one of my Gemini AI assistants...looks like you are spot on.

"Google’s 2026 infrastructure has shifted away from fixed quotas to Account-Level Dynamic Throttling. Even if you have a massive bank of purchased credits or a high-tier subscription (Pro/Ultra), the "Flow" pipeline essentially looks at your real-time behavior.

Complexity Penalties: If you are running high-complexity prompts—like the hull-warping, particle-pelted ship exteriors we’ve been working on—the system flags you as a "high-load user."

Peak Hour Squeeze: Reports specifically mention that during weekday afternoons (US time zones), the "standard" limits for Pro users can drop from the advertised 100 images/day down to as low as 20 to preserve server stability for enterprise clients."

Thanks for highlighting MegaBuster.... I had no idea about these advertised limits and throttling. Easy to adjust now that I know what's causing it. Thx!!!

[desert horizon]

help

[sinful lotus]

My favorite failure of all (LOL). Try again later. The server is busy.

[desert horizon]

The problem is paying for it and not being able to use it. 😮‍💨

It has occurred frequently

[sinful lotus]

Yeah... the story of my life since I discovered Flow a year ago....

[regal spoke]

Has anyone here ever worked on making a video for a car dealership?

[ashen plinth]

Does anyone know how to prompt for a shot shorter than 8 seconds in flow please?

[tepid cedar]

From my experiences, it's easier to just have a longer shot and make an edit in post.

[ashen plinth]

Thanks, but unfortunately the 8 second length is restrictive. I’ve found there were less faults on a shorter prompt before they changed it.

[pulsar coral]

We all just learned it by painful sheer osmosis

[unreal viper]

Hi! We're so glad to have you here. To ensure our moderators can keep the community safe and everyone can join in the conversation, we ask that you please use English in the public channels. Thank you for your understanding!

[pulsar coral]

It's fine they just needed help on videos and said they'd go do it in private.

Gee the south american guy speaks spanish too what are the odds.

Anyway I found out I learned enough prompting to just need a $5 flow account

[wary ledge]

scroll down:
OUTPUT_SCHEMA — 4 lines, exact prefix syntax, zero deviation:

Film Stock: [Capture medium + specific stock/sensor aesthetic + grain structure class
(fine crystalline / medium chunky / coarse organic / digital noise pattern) +
scan artifacts (dust, scratches, base fog, gate weave) + processing anomalies
(push/pull processing evidence, cross-process color inversions). If grain is
ambiguous: INDETERMINATE: [observable texture descriptor only].]

Lens: [Focal length estimate in mm range + lens class (prime/zoom, optical formula
era) + focus plane depth + depth-of-field envelope (shallow/deep, focus roll-off
character) + measured optical aberrations: barrel/pincushion distortion %, vignette
gradient (center-to-edge ΔEV), chromatic aberration (fringe color, axis), flare
type (veiling/ghosting/lens flare geometry). INDETERMINATE permitted per sub-parameter.]

Lighting: [Primary source type (tungsten/daylight/fluorescent/LED/practical) +
color temperature estimate in Kelvin + contrast ratio classification (low/medium/high
key, estimated stop range) + directionality (clock-position, angle of incidence) +
fall-off gradient (rapid/gradual, Inverse-Square-Law compliance) +
fill ratio estimate + secondary/practical light presence.]

Color Grade ("[2–3 word aesthetic vibe name]"): [Color pipeline identification
(cross-process/bleach bypass/print emulation LUT/digital grade/scan-native) +
temperature bias (Kelvin shift warm/cool + directional channel) + luminance curve
shape (lifted matte / crushed blacks / S-curve / linear) + shadow tint (hue + chroma) +
highlight tint (hue + chroma) + midtone color cast + halation presence/absence
(color + radius) + black-level status (open/crushed/chemical base) +
skin tone rendering if present (hue shift, saturation, luminance).]

[pulsar coral]

But I did start with a $20 one, learned a lot that month

So barely feeling the limits atm, I'm done with the images in two or three hours, got a Deviant account and Flow is really useful there

So yeah best you can do is use your plans to try and learn and improve and you'd get stuck less when the limits hit

.... in fact I'm not even done uploading all the Flow images to Deviant with just a $5 Google plan and haven't used my Gemini bonus images, it does give more than one can chew once you get used to it

Cause yeah reminder your same Flow account links you to Gemini

[devout ledge]

Duh. Thanks for the clarification.

[wary ledge]

No problems👍

[pulsar coral]

Not to mention there's the rumor Whisk won't be taken down, so that's another happy day reason. Even tho I did just learn Flow's Nano 2 can prompt very similar to Whisk's Nano 1, and once you know that you can somewhat get hybrid images and eeeeh I'll see if I can post something safe enough later to show it.

[proper drum]

lol

they can just send the same text to Gemini and do the translate

[pulsar coral]

But I'd say it's pretty awesome

[rigid nest]

Whisk did that for me in past 🙁

[edgy holly]

M-F during work hours in North America and Europe are the worst.

[edgy holly]

The first thing doesn't = the last thing.

[edgy holly]

Image gen doesn't use credits.

[edgy holly]

They should pin this, and have it on repeat every ~3 hours.

[edgy holly]

Only a fake one 🙂

[edgy holly]

Prompts don't change settings, like clip length.

[edgy holly]

It's been 8 seconds ever since Veo 2 released early last year.

[dry frost]

I'd shop there.

[pulsar coral]

Yeah his Gemini explanation works better than mine I think, we could just screen it too.

[sonic rover]

cool

[pulsar coral]

It's good, tho it could use

[Fixes: No unnatural clipping/intersecting/overlapping. No occlussion errors.]

That way the sword slashes would actually hit something

[sonic rover]

I generated it using Seedance.

Its all-round reference mode can refer to 12 images, while VEO can only refer to 3.

[pulsar coral]

I shoulda caught on that, couldn't really spider sense it this time

[dry frost]

The frame rate tipped me off

[sonic rover]

where？

[pulsar coral]

Specially cause all of the Chinese gacha game aura, that's always a good indication it's Seedance. Flow's Veo tends to be more american styled

Like if watching that Wukong movie and it's videogames

[dry frost]

Typically most things in VEO are 24 or 60 fps, but your clip is far smoother, felt low triple digits at worst

[sonic rover]

And veo will Europeanize human faces

[pulsar coral]

That's what I'm saying yeah, that or americanize them

[sonic rover]

I really hope veo4 can catch up with seedance. Ultra infinite generation is really cool

Currently, if you are not a premium member of Seedance, you have to queue for 8 hours to generate a video

[pulsar coral]

Just send them a lot of feedback on what you don't like right at the video and they'll get on it

That's usually why they allow free generations daily too, we're supposed to help train the model

[crisp basalt]

veo 3.1 lite cna do extend feature now

[dry frost]

Sick! For 5/10 credits a pull and frames reference I bet lots of cool things can get made!

[sonic rover]

veo, I can only use it to generate this kind of video now

[sonic rover]

But I feel that the quality of the generated videos hasn't improved

[dry frost]

Likewise with myself, but video generation is still a new thing compared to all the other outputs AI has done

[frosty niche]

I thought it was more like 12 hours..

[pulsar coral]

Yeah this a typical locked first person angle with no transitions, seems to be what people use Veo the most for

Probably cause it works great for reels

[sonic rover]

This kind of video, veo can barely generate

[dry frost]

The classic violation of hinges, never stops me from smiling haha

[pulsar coral]

it's like she crawled into a smaller room Alice in wonderland style. I think I'd need some good 20 attempts to find a solution to that

[sonic rover]

At this point, a Gemini or Claude is needed

[pulsar coral]

From fixing it at Nano it would probably go something like

[Fixes: Professionally-correct physics, realistic perspectives, coherent object/prop/building scales, depth cues, correct proportions, professionally correctly assembled spatial layout. Perfect natural human anatomy and morphology despite what other prompts say, no miniature objects/beings/props/items/objects in the background everything must be in a correct size ratio that the character's full outline could normally use. ]

Point yeah it really ain't that easy and the model should be doing it itself by reasoning

[sonic rover]

I usually tell Gemini the general idea and ask him to help me generate reasonable camera movement scripts, but basically, the first video clip takes at least one to two hours to adjust

[pulsar coral]

Then again maybe she did drink a magic shrinking potion from Alice in Wonderland, what do I know from what's going on there

Or could be a Narnia thing

[sonic rover]

Haha, most of the audience are interested in the gold inside the cave.

[sonic rover]

Yes, many viewers have also told me that this video resembles The Chronicles of Narnia

A video queue for 40 minutes is too slow🥹 🥹

I asked Gemini which platforms on the market allow unlimited video generation upon upgrading to a premium membership. He replied with Flow, Runway, and Grok. I tried Runway's unlimited generation mode, but it was very slow, and there was basically no comprehensive reference model. Grok is useless for fighting videos (I feel it can only be used to generate funny animations)

[pulsar coral]

Yeah most sites giving "unlimited generation" are actually a trap like they mean they'll keep you waiting for it., Veo used to have unlimited but people were abusing it, so it only uses credits now. Tbh the actual strong point of Grok is the NSFW, same as Wan, Seedance is indeed been showing to be easier to use for long detailed cinematic scenes.

Wan and Grok mostly work better with encased precise prompts and a reference image.

But Flow's VEO works better without reference image and a coherent scene sequence.

In fact Veo can be jawdropping if you just let it work by itself with a single person talking.

Again, surely being used for short reels with useful stuff, which sells anyway.

[sonic rover]

Yes, I have received private messages on Tiktok asking me to sell videos

In a second-tier city like China, AI animation companies are everywhere. I'm a bit afraid to go there, as I'm worried they might close down at any time

I made this video before, but few people watched it

[edgy holly]

GEMini 4 update coming?

[sonic rover]

and

and

[regal spoke]

I started one, this is the beginning

[shadow dove]

A 10-second Hollywood cinematic action scene. A giant green muscular superhero jumps down from the sky in slow motion into a crowded student protest in Bangladesh. Dramatic lighting, dust explosion and ground shake on impact, students holding banners look shocked and emotional.

After landing, camera shakes and zooms in on the superhero’s angry face. He stands up slowly and speaks with a deep, powerful voice.

Voice (Bangla, deep male, emotional, slightly angry, cinematic echo):
"কোটার জায়গায় কোটা রইল, মেধার হলো না জেতা...
শিক্ষার্থীদের কাজে লাগিয়ে টোকাই হল নেতা!
জয় বাংলা... টোকাই সামলা 😑"

Add cinematic explosion sound on landing, heavy bass impact, crowd noise, tense background music, dramatic boom, echo voice effect.

Style: ultra realistic, 4K, Hollywood movie scene, slow motion, dramatic atmosphere, epic composition.

[sonic rover]

Wow, I don't understand. I generate images from midjourney and put them into flow for generation, but even in 4K download mode, they are still a bit blurry

I really want the clarity of wallpaper type

[frosty niche]

Might be heavy traffic, this was happening to me last week, & now that sora is sunsetting theres mostly likely going to be an uptick in traffic.

[blissful abyss]

How does this look

[pulsar coral]

You have to forcibly end your prompts with things alike

[Complete image's quality: Professionally remastered in best known quality, HD ultra high resolution, sharpest definition, denoised, unblurry, clean lines, High contrast, fix any jpg artifacts, 4k detail, fine textures, clean edges. ultra detailed, fine textures, sharp focus, high frequency detail, 8k detail]

Even the same chat models are now recommending this, even tho like I was doing it by instinct before

Tho if you will use it on reference images might wanna change remastered for "Everything must remain accurate true to reference image without any creativity freedom."

it's much easier to upscale once everything is clear enough from 1K, because 1k is the best an image would usually look like already.

The model won't do it by itself maybe because it could mix prompts up even more than usual.

[gloomy verge]

Hello. Can I ask about style of images something? I created this illustration, but now, I want create other illustration, but in same ilustration style, only character and evironment will be other. How to preserve only the illustration style from an image, not the character and surroundings? Is it possible? Thank you for help.

[rugged glacier]

elementary request, but still moves the camera💩

[gleaming lynx]

😹

[rugged glacier]

I had 600 bonuses, but I spent 400 bonuses to stop the camera from moving.😭

[sonic rover]

"Oh, I thought even if I didn't add it in the prompt, the model would still default to 4K, etc."

[pulsar coral]

Try

[Camera/perspective/composition: Locked to reference image, no angle panning, no transitions.]

But yeah, it should still be understanding you with that other prompt.

it might be thinking there's an actual static physical camera literally inside the video and that's what doesn't move.

[oblique bough]

Salut

[pulsar coral]

I think this one's safe enough. This sorta is an example of a hybrid prompt image of original Nano Banana and Nano Banana 2, asking Nano Banana 2 to use Nano Banana 1's image style.

It's mostly reflected on the body type but not much in the art style.

I''ve got way better ones but yeah they're not this safe.

Have you ever seen a sky without light pollution? Nano Banana 2 has seen it, and it's even much better than that.

Been practicing this because of the user who said she could't generate the mono-kinis, beachwear whatever, can do it well now.

Not to mention the cash grab

[gleaming lynx]

finally got the aesthetic right

[pulsar coral]

Why are Nano 2/Veo making people seem like they were casted for Dr. Who

No wonder that other guy said they looked european... which is close enough

[gleaming lynx]

are you talking about my image? cause I prompted it this way

[pulsar coral]

Yeah, it looks great tho

Specially the mirror reflection, I do often mess that up

Not even wanna try that atm

[gleaming lynx]

yeah I basically spam generated like 64 images

as you can see it for the driver seat side positioning wrong, and in the second image of the grid the car is completely open

using multiple images i got all the frames I wanted though

0 credit image generation really helping me out

[pulsar coral]

well yeah whatever works I guess

And yeah the driver seat changed... again european stuff cause they drive by that side

[fading dirge]

not going to lie, veo 3.1, aside form audio, feels so outdated compared to other video generators

[minor badger]

Has Flow changed the policies? I can't generate normal images reusing characters i have created inside Flow in simple prompts, no violation, just basic things. It's completely unusable on my end

[gleaming lynx]

💯 its soo trash im genuinely crying

[pulsar coral]

Other video generators just had a way more generous free phase, which might seem like a money loss but is actually used to train the models for way less cost than using professionals and massively faster due to volume of users.

So they advanced much quicker than VEO. Eitherway you might just always want to send feedback about what you liked or not after being done with the video, that helps like you have no idea.

Veo still has it's strong points tho

[gleaming lynx]

yeah dont get me wrong i still prefer it over anything available to me (dont have seedance 2 access)

its just that there a lot of things that I would love to see improved in the new model

[pulsar coral]

Another thing one's gotta say, the video generators that allowed at least certain degrees of NSFW are the ones that advanced much faster, except maybe for Seedance. All the constant stumps and blocks also massively hinder progress, since it also blocks every other type of user intentionally or not.

Sometimes you gotta do what you gotta do, and you're either in the business or you're not.

[regal spoke]

Niceee

[gleaming lynx]

imagen 4

[minor badger]

this is worse than nano though

But overall why has veo got so bad? the quality is absolute garbage now. what have they changed?

[gleaming lynx]

yeah dude cause its old lmao

[gleaming lynx]

can you send a comparison video?

[minor badger]

look at that!! what the hell. It used the be the best

[gleaming lynx]

😹 yeah this happens to me a lot

[minor badger]

what happened?
I have made multiple entire short films with veo a few months ago and they turned out perfect. but this is unacceptable

have they lowered the quality on purpose in the anticipation of veo 4 coming out soon? so people will be "forced" to switch

luckily i haven't stopped corporate job to become a full time AI creator lol

[pulsar coral]

Being fair, that character could need to be prompted to turn 360 degree only to one side, it might have been prompted only to "turn", and yeah VEO gets convulsions from that

In the first video no one was turning, in fact most video generators have huge probs with characters turning

[-Pointers: Make the animation coherent and unhurried with way more than enough frames to be fluid but don't repeat any frame and keep all frames at the same speed. Keep body's anatomy accurate and morphologically correct to any body's sides or in relation to poses.
When rotating the body can only do it until a 360° finishes, the whole body can only move at the same speed as the backside and can't unlink/flip/pose anatomically/morphologically unnaturally.]

Scene: The character dances to a popular existing k-pop intricate choreographed non-simple acrobatic routine with related music. must include at least one twirl towards the same side.

Identity of character: True to reference image, same facial structure, same face, same props.

Camera angle/Composition: Locked to the same one from reference image, character must remain full body in focus.

Background: Coherently related to scene. No transitions.

[Video Quality: All professional. Remastered 4K HD ultra high resolution, sharpest definition, denoised, unblurry, clean lines, High contrast, fix any jpg artifacts.]

Yeah gotta go out of the way to learn all that tho, which is a hassle

It should really understand that by it's own logic

[minor badger]

dude, i appreciate your comments, but you don't need to explain me how to prompt. I've made successful videos. I guarantee you I can tell the quality of the veo model has gotten bad

[pulsar coral]

Yeah but in those examples it's old known probs even more advanced stuff alike Grok has.

And one can patch by prompting, I'd say it has worse probs

Alike how it doesn't know what vocal distress is

[minor badger]

haha it has quite a lot of issues for sure

I had put video creation on pause for some weeks now, but I'm really shocked today to see what's going on

[gleaming lynx]

i never tried grok video

is it good?

[minor badger]

where's seedance 2 yet?

[desert horizon]

Welcome to the team

[wet fern]

wdym

oh

[desert horizon]

It's impossible to generate any video.

[wet fern]

thank goodness its not just me

google labs gotta lock tf in and start fixing their stuff

[desert horizon]

Yes, I thought it was only happening to me too until I joined Discord.

[astral fern]

And I mean this in the nicest way possible..

With SORA2 closing is VEO gonna get an update because holy moly the difference between the two is night and day.

[full flicker]

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

[fossil python]

Downloading the extended clips as single video is no longer working?

[blissful abyss]

Thanks

[wide ether]

https://tenor.com/view/tommy-boy-comedy-chris-farley-stressed-upset-gif-16379051

Can we just get a switch to allow us to shut off music and just have raw audio

I'd like to generate videos with audio, devoid of production music

[radiant leaf]

I have questions...I can use a child image for veo creation, but if I'm trying to upload the image as an ingredient for Flow is telling me that it goes against the policy. I tried even changing language as I wanted to extend a video making the child appear as an angel, but is doing the same, or again, creating results that would be considered the Twitty bird of grandma's social messages.

[sinful lotus]

Yeah, what would be awesome is to have a "seed" table that we can assign to things like the scene, the vocalizations, the music, etc. It takes quite a number of words to define and keep consistency.... and even then you have to be seriously careful not to have these items drift. Even adding a few words to a vocal text (VEO in this case) can cause the seed to drift (or so it seems).

[gleaming lynx]

yeah I deleted my message cause nano banana isnt a diffusion model

but there definitely has to be a way where I can change stuff without changing anything else

[sinful lotus]

LOL!!!!! I have this problem all the time... "You should see my list of negative prompt instructions....something like "No music, no music, pretty please, please stop adding music, I love music, but not on my video when I have somebody speaking, please stop, stop stop "..... 🙂 🙂 🙂

[edgy holly]

Negative prompting isnt effective in Veo.

[wide ether]

yeah veo ignores negative prompts 🙁

[wet fern]

HAS IT BEEN FIXED

[modern oasis]

Is this working for anyone?

[polar mural]

I am recieving this same fault.

[mossy badger]

im having this problum while im trying to download created images on 4K is there anyone here who can help me to resolve this problum

[sinful lotus]

I've had this problem from time to time.... the only thing I could do was "wait" and eventually it would do it... (in my case, it was in VEO or Nano Banana 2)

[hollow perch]

Same issue for me for last 10hours.

[versed marsh]

Same issue

[gleaming lynx]

istg if i get the currect nano banana pro model locally and open source there is no stopping me

[pulsar coral]

Only when you've practiced prompting enough, just like at VEO the encasing works well enough there. But if not you gotta prepare for some nice uncanny valley. It does NSFW way too easily tho... which sadly did advance things while it was very free

I don't think Google's VEO is ever gonna take that approach

[gleaming lynx]

yup, but idk how a random prompt i took from twitter generated n i ps

[pulsar coral]

VEO's nice little educational reels are very healthy tho..

[cobalt junco]

i pray my library does not blow up when imagefx and whisk import lol

[crisp basalt]

.

[gleaming lynx]

😭

im taking on this project now

sometimes gemini randomly generates these bangers in the middle of a conversation and I love them

[pulsar coral]

Yeah just talking to it and telling it to avoid doing things more often has the same effect tbh.

Unless it's a reference image with a kiwi and a banana, then it goes Brazilian nuts for some reason

I swear I'm still at it with that one

[pulsar coral]

Yeah you got it right, it's in this kind of retail product oriented stuff where it shines best... which makes sense from a sales point of view but

As it happens with Nano Banana 2 when it tries to fix your images by comparing it with existing images of characters and faces from the internet, more often than one would think people much rather do their personal ideas and characters.

[gleaming lynx]

dude this video is garbage
its just really funny, the tone and movement

damn i could not have prompted it even if i wanted to

[pulsar coral]

I'm so gonna redirect people to this no music prompt

[pulsar coral]

Well, it's corporate garbage for sure... but idk the dude seems to be moving ok

One of the first things people do to mock AI is dart to count a hand's fingers... so if that's ok you more often can call it a day

The asian girl did disappear tho, but that was a cut

[pulsar coral]

Button is there already, but doesn't work yet. Edit, nvm button's gone again. Well it'll be there when it'll be.

and the little doomsday message is back.

[cobalt junco]

lol i was like what did i do

[pulsar coral]

I was trying to say "button" and that was your name. Can't be the first time it happened.

[cobalt junco]

wont be the last either 🙂

[pulsar coral]

Well I got a counter offer.

[cobalt junco]

it works you opt in and they do it later

[pulsar coral]

"Thank you for being a Flow user! Image generation is moving to Google AI studio, our (g)old place for still being able to use both original Nano Banana (Gemini 2.5 flash image) and Whisk's "Precise mode" which is just the same thing. You can create in Flow until you hit your daily image generation limit so wait till it resets tomorrow year 2026. To keep your media, remember you also have an automatic limit for upscaling."

[autumn scroll]

What's the wrong please help me 🥺

[fair cave]

I wanted a video, I added the photo but at the end it changed facial features . Right is the video, is different from the photo, how it can be corrected?

[pulsar coral]

If you keep getting that message over and over more likely you just hit your daily limit, and that might happen even on Ultra plan while you have enough credits. You can try again tomorrow at around the same time you successfuly generated a video today and it should be working again. Flow just doesn't display the correct message for when you run out of daily videos and images.

[autumn scroll]

What's the wrong please help me 🥺

[pulsar coral]

Like I tell ya, but yeah someone might have a second diagnosis can't hurt to try.

[autumn scroll]

I change the email account and still error

[pulsar coral]

First take a reference image of the guy if you don't already have one, then add at the top of the prompts something like

[Identity of character: True to reference image without any creativity freedom, same facial structure, same body, same props.]

Tho if it doesn't work, it might be conflicting with other prompts and you might have to try a clean slate.

If the camera changes rooms you can also lock it with prompts alike

Wait lemme try that again

[Camera/perspective/composition: Locked to the same one from reference image, character must remain in focus, no angle panning, no transitions]

Well, that last closing message displayed at Whisk seals it, that's all folks, been fun tho, so gonna move on now cause 30 days of it is more than enough, I guess it was unsalvageable but I do need NB1. Will be around tho. And yeah I do mean flow not whisk in the meme.

[sour nebula]

Hi, I am new to Discord and my English is limited.

Draw Image is not working for me in Flow.
Select Image works, but Draw Image shows no brush, stroke, or mask.
This happens in Chrome and also in Incognito mode.

Is anyone else seeing this?
Thank you.

[proper drum]

so we now can use voice (from available options) as ingredients...

[bright heron]

"Only image files are supported." when I try adding an mp3

[bright heron]

how so?

[proper drum]

not for upload and just for Ultra during "experimental"

[bright heron]

ah, i'm on pro plan

[bright heron]

maybe its a regional beta too? where are you located

[proper drum]

Panama

[turbid lintel]

congrate flow team , you guys solved the voice issue with the new update ... thanks you very much

[cobalt junco]

if you don't choose a voice can it only use those voice models native?

[burnt mason]

I want to play with Flow but I only gave Pro. Would it be possible to do a scene with three people using Flow?

[frosty niche]

Can you use your own voice, and then restyle it with an existing voice model?

[upbeat sierra]

We're glad to hear you like it! Very much value the community's feedback

[upbeat sierra]

Voice Ingredients is Ultra-only for the initial release, but definitely hearing hte feedback that Pro tier would like this as well

[upbeat sierra]

Voice Ingredients work best with a single speaker / person in the reference image. We are avidly looking to support additional use cases

[upbeat sierra]

Voice Ingredients allows user to pick one of 30 provided voices. You can 'restyle' it by additional instructions in your Veo prompt!

[main saddle]

thanks ummerr you legend

[tepid cedar]

been real "fun" trying to work in flow tonight.

[hollow hare]

Go touch grass bud

[crisp basalt]

well that actually gagged me for a bit but

the voice ingredient works well , used it on one of my carbonara video back then 👀

[hollow hare]

What in the actual crime is this food being made, awesome creation tho

[crisp basalt]

tho i have a question , is there a proper guideline when it comes to prompting the tone and style of the voice?

[fading dirge]

the extenmd feature is useless in google flow, the voice changes ( even when choosing the new voice select feature), the ambient sound changes and the quality change makes it pointless

[crisp basalt]

i think they already stated that lol

[fading dirge]

google stated that?

[crisp basalt]

yeah in the changelog

[fading dirge]

oh dang lol. where do you see the change log?

[wintry solar]

flow failed

[upbeat sierra]

it should show up directly on flow.google when you next visit the site

[upbeat sierra]

that's no "fun"

please start a thread #1238621511401734254 and share as many details as possible so we can look into this. it's generally best to post there as it is easier for us to follow up on than in this chat. things that are helpful to share:

• prompts tried
• model used (Quality, Lite, Fast, etc)
• type of account (Ultra, Pro, etc)

and we'll do our best to resolve

[crisp basalt]

actually, you guys really need to update the changelogs page 😅

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

[fading dirge]

oh ok, update showed up for me earlier talking about the voice select update but i didnt read it because i rushed to try the voices

[upbeat sierra]

This is awesome! Will share it with the team ❤️‍🔥

[upbeat sierra]

Understandable. In regards to extend, but the model doesn't support that at this time. We're working on this, but in the interim, you can re-use the same voice ingredients across scenes / shots to try to make longer videos, like the one @crisp basalt shared

[upbeat sierra]

yea....fair enough 😅

[crisp basalt]

yea the best workaround right now is to use the 'ingredients-to-video' workflow for your next scene. That's the only way to reuse the voice 🙂

[frosty niche]

Is that some bacon-infused pasta! Whoa, nice.

[lunar mantle]

hello, does anyone know why flow fails to download 4K images

[wintry solar]

flow didn't failed it evolved into lite

[tepid cedar]

no point. user issue. it was a matter of wording and tagging.

[broken kettle]

I'm Ultra how to open Ingridient feature

[wintry solar]

how much you spending

[broken kettle]

thanks fam it's vietnamese from my side lol

[broken kettle]

half price for 3 months 😄

[crisp basalt]

Same!

It’s a really good discounted price

[broken kettle]

better if you got family to share

[broken kettle]

how to pick voice like the demo video

[edgy holly]

Push the plus button on the bottom left of the prompt box, and find the tab for voices.

[broken kettle]

would be great if we could TTS the voice separately

[edgy holly]

That'll come eventually. I've only seen one model that can do that, and it was recently.

You can accomplish the same thing on the back end, taking a video you like, going onto a service like Hedra or Sync, and it can lip sync to uploaded dialogue.

[rapid sphinx]

Hey @main saddle
Any info on the recording of the flow workshop??

[timid sand]

hey! anyone know how to add childrens images in flow ?

[arctic kelp]

Hi , I selected a male voice, but when the video is generated, the voice doesn’t match what I chose. Has anyone else had their output voice differ from the selected one?

I’m not sure if this is caused by my prompt or something else. If anyone has a reliable sample prompt for consistent voice results, could you please share it?

Thanks a lot!

[glad yacht]

its forbidden 😅

[harsh flare]

same issue happening with me

[timid sand]

TKS

[frosty niche]

https://tenor.com/view/touch-grass-datagram-grami-gif-17326686380942026919

[deep fox]

Same here. Or I don't know how it works

[rapid sphinx]

We're able to select voices now?

Yes saw a screenshot
Got my answer\

[vernal remnant]

has anyone noticed that after this recent update, many of the previous characters are no longer working, even if the characters were created inside flow, it now says " Failed: This prompt might violate our policies about generating prominent people. Please try a different prompt or send feedback" 😒

[fast cloud]

This is an Adobe inference... Smoking!

@upbeat sierra can I get an upgrade from Pro to Ultra in order so that I can provides sincere feedback on the 30 voices feature. All I need is 7. Thanks a mill'y

[bleak pivot]

i had this last night, it was fine for ages, then all of a sudden it refused to create anymore images using the the ref image i was using.

Also, this update is no different to prompting the voice you wanted. just now we have a list to pick from instead. and its still pointless because we can't keep the same voice in extend feature. which is probably something that most users need! (including me!)

[vernal remnant]

Moreover, it does not work when there are two characters in a scene or more, it just breaks, or everyone has the same voice 😂
And I cant seem to understand, why on earth would a voice update break our previously used characters.

Even If I create a new character, if it looks too real, it says prominent people, if it looks a little fake, it goes through, WOW! 🥲

[potent belfry]

OK since we are all going to Flow now I have one problem I CAN not remove music from my video's I mention it in my prompt to remove it and it still add music

[uneven sand]

Does anyone know why this keeps happening?

[bleak pivot]

Flow is broken lol

[upbeat sierra]

In regards to extend, but the model doesn't support that at this time. We're working on this, but in the interim, you can re-use the same voice ingredients across scenes / shots to try to make longer videos

[upbeat sierra]

Right now, the feature works best with a single person in reference image and a single speaker. We are working hard behind the scenes to improve the quality and support more use cases

[bleak pivot]

ok, if your working on it then thats great, thank you for the tip/update 🙂

in my videos I requir longer scenes than 8 seconds, so having to create separate clips just doesn't give the feeling and cinematic style im looking for. so I'll just have to wait and hope Flow sort it in extend 🙂

thanks

[upbeat sierra]

we hear you, will pass the feedback on to the team

[bleak pivot]

ok great, thank you. also, if they can make it so we can create longer than 8 seconds, that would work too 😁 15-20 seconds would be fantastic!

[upbeat sierra]

sorry to hear about the trouble. please submit a bug in #1238621511401734254 with as many details as possible (prompts, model tried, examples) and we'll try to look into this

[ionic agate]

Built a Flow video about Flow . Tips and Tricks and a brief review. my posting agent messed up the title 😐

https://youtu.be/nDD-hk5eTCg

[safe karma]

Anyone know where I can find the video record for the presentation that was done on wednesday? I was able to be present for most of it but missed the final 20 because of a meeting.

[uneven sand]

Just asked the same but still no answer

[harsh flare]

I can't download images in 4k it says upscaling failed ,
Is it happening with you all too ?

[gleaming lynx]

i think i saw someone complain about this yesterday

[ionic agate]

https://youtu.be/xB5MqhFLrTM

[gleaming lynx]

shared something new that I have been working on in flow gallery

[mental apex]

Anyone try the new voice stuff for a long video of one person just talking for example a podcast or something?

[ionic agate]

new voice stuff!!!! where, where! This is so needed

[sonic hawk]

available

[uneven sand]

Are you doing this only with flow? extend clip or one clip at a time? If I may ask

[gleaming lynx]

i rarely use extended clip

i generate 8 second clips and a lot of the times i trim them too

when i want small movements like head turning, i deliberately add things after the head turns so that it doesnt hallucinate stuff or make the movement too slow cause there is no option to make like 3 second long clips

[uneven sand]

Ohhh... that's a good idea... My biggest problem is to make them speak in Portuguese PT and keep voice consisteny-.. yours looks great

[gleaming lynx]

thank you bro

yeah i had a german client once, i used elevenlabs to swap the audio

also I didnt prompt the voice too much, it gave me knocking on the door sfx once and I just used that part in the correct places

and trimmed the weird part where it originally generated it

[uneven sand]

Very nice work... are the video getting a lot of views? I imagine so

[gleaming lynx]

lmao no

this is just a spec ad i started working on

clients mostly run AI UGC meta ads

i am trying to transition to something like this, a little cinematic storytelling

[uneven sand]

I like it...

[unborn bolt]

whats the best way to ask flow for a character to be aislated from background, so i can use photoshop remove background without getting some background leaks through hair of people?

[gleaming lynx]

thanks dude it really means a lot
cause i am having a lot of self doubt lately.

thinking maybe only people who generate AI videos find mine decent
and normal people wont

i guess you too are into the AI video scene but still

[uneven sand]

You're not alone in that... BUt I really liked it.. they are instructional, no slop and they are very well done... Maybe you should try to go for9:16 and go for instagram or Tik Tok...

[gleaming lynx]

i ask it to make a detailed character sheet of the character

[gleaming lynx]

thanks man,

yeah I have different ideas for portrait videos, will work on them after making 2 landscape ones

[subtle cedar]

(if this should go in another channel please kindly let me konw) Good afternoon, thanks for any time you take to respond. Does flow have image interrogations? Whisk provides interrogations at time of image upload which can be used to parse syntax and determine the most accurate/precise referential language to use. If whisk and imagefx are being terminated, and FLOW is what everyone is being pushed to, I thought it would also include features and functionality that whisk provided such as image interrogations. I have looked everywhere and am unable to determine if FLOW can provide image interrogations or if there are any viable Google-based image interrogators that can act as a replacement. The desire for google-based interrogations comes from the need to use the syntax google image generators prefer, so we speak with the same language when creating iterations. If the interpreter reads the image as a "square" and you start referencing a "box" it could error out and not know to adjust the square, because it's being interpreted as a box- which if we had the verbose image interrogation we would know to reference it as a square. (Syntax may refer to word choice, grammar, word order, phrase structure and sentence construction. )

I would be excited to know it's in the works with some sort of ETA so i can dedicate time and resources to other tasks. In the meantime, by April 30th, a good solution needs created.

Image interrogations play a vital role in our image and video generation workflow by (using the preferred syntax) essentially creating handles-of-manipulation or touch-points for an LLM to modify. Without interrogations, users are forced to create their own touchpoints or handles-of-manipulation, guessing the syntax, which leaves a lot of room for error. This ultimately costs more generations, when guessing vs using an interrogation w/an LLM as a springboard for iterative prompt design.

[rain rapids]

Invalid JSON payload received. Unknown name "referenceAudio" at 'requests[0]': Cannot find field.`, status: "INVALID_ARGUMEN

this comes when i generate veo3.1 normal with audio

[cyan dragon]

I have problem from few days when try generate with nano banana pro
always saying im trying to generate too quickly...

[ionic crown]

Same, anyone know the problem?

[rain rapids]

yes

[ionic crown]

Why is that?

[cyan dragon]

another problem

Most of the time i just fight against policies for no aparent reason and errors

[bright schooner]

how to assign specific voice the new feature? i cant seem to upload mp3?

[gleaming lynx]

please tell me this holds weight

[olive thistle]

Does tagging image actually work? Do you just need to type @ and then the whole filename of the image?

[unique abyss]

I've had a chance to play around with the new voices feature, here's quick video that show the process, and also has a short six shot film, showing the same character across six different shots.

https://www.youtube.com/watch?v=wZoAD8uFqFw

Some observations:

To have the character in short film speak in a consistent British accent, I needed to not only include the image and voice, but describe him (male, british professor from Oxford, 30 years old, from the 1920s) or the model would revert to flat "accentless" General American pronunciation.

Although the guidance was that this works best with one character in the scene, I was able to do a reaction shot from a different character, where only the speaker's hand appears on screen.

I was please that the voice could have a range of emotions, and was pretty expressive.

I had to create a fair number of shots (20) to get 6 that I liked, without spending much attention to filter for visuals--I'm sure the ratio would improve with experience, but at this early stage I wonder if this might be better suited for very short works of a minute or less, rather than 5-10 films. (That said, I'm almost certain to want to experiment with a five minute film, even if this is early and experimental.

[frosty niche]

What in the 4th dimension is this?

[lyric swallow]

i'm just getting started checking this out, can anyone point me to a good tutorial?

[edgy holly]

You can't upload voices yet. It's brand new. You can only use the selection of voices they have.

[edgy holly]

If you find some, share with the group. I was going to make some, but some mods get real angry when you're too helpful to too many people.

[unreal viper]

hey! we actually love to see tips being shared! tutorials are a great thing to share and we love to see others sharing what they have learnt!

there are however some rules around self promotion and advertising that we like to make sure people are following!

an example of something we would moderate would be: i made this video, go check it out and subscribe to my channel and like it.

an example of something we would not moderate: hey guys! i found/ made this tutorial. i think it can be really helpful if people need help!

this follows our #rules-and-guidelines・📋, if you want to learn more about advertising in the server please see this message: #rules-and-guidelines・📋 message

[wintry solar]

which tool shall i use which use my voice real time making generative video

[crisp basalt]

oh oops.

[dry frost]

👀

Happy accident?

[crisp basalt]

i think so? did they ever allow generating anything involving IP marvel character?

i can confirm that was in fact , an accident , i tried same prompt LOL

4k

is it not taken directly from the movie ? omg

the flow team reading this on their way to kick my butt rn

[dry frost]

AHHH!

[crisp basalt]

so i decided to try push it a little

lmao

the prompt was only : create

isnt that dialogue from wanda

google and flow team is prepping to kick me rn

ok thats enough pushing for today!

[dry frost]

https://giphy.com/gifs/isekaimeta-nervous-sweat-sweating-4QgiErmjZiPESRcKYt

I didn't see or hear anything, must've been a glitch hehe

[crisp basalt]

https://tenor.com/view/monster-versus-vs-alien-monster-vs-alien-gif-7403118861599528686

[frosty niche]

https://tenor.com/view/good-morning-fall-hang-gif-26723790

[dry frost]

-# Yeah let's keep the chat scrolling
So those new voice pack options huh?

[frosty niche]

Yeah lots of options i hear..

[dark dirge]

What am I supposed to do?

What's the best way to handle this?

[viscid marsh]

Question for the community: does Google Flow allow for uploading storyboards? If so, what are the best tools to use for this? I'm finding my videos to be too far off my vision for my liking and would like a way to visually direct the tool.

[edgy holly]

You can upload any image you want.

[raw oak]

I can't seem to generate in ingredients. it says I have reached maximum ingredients but there is nothing there

I keep getting this message when i try to generate in 3.1 lite

Does 3.1 lite support image references?

[weary cargo]

just refresh

[ionic crown]

the refresh does not help

[proper drum]

same as fast and normal

[raw oak]

Weird I got a message saying when I use voice and image reference it’s not supported. So I guess can we use our own image for the new voice over

[proper drum]

Oh

yes seems ingredients doesnt work, but frames works

[vestal vale]

hello

[gleaming lynx]

no

[edgy holly]

You asked about image.

[astral fern]

Hey Ultra users, do you guys have priority when generating using Nano Banana Pro and Nano Banana 2? How fast does it usually load for y’all?

[crisp basalt]

ultra user here, honestly not much diff. it’s mostly server-side. lately there’s random times (idk when exactly) where demand spikes and everything slows down, so everyone gets hit

[astral fern]

Ugh. I’ve noticed it’s typically in the morning. Between 7AM to 9PM EST.

I’ve been wondering if it’s because it’s late in the evening on the other side of the globe (China and India) and if they just flood it with requests.

It’s really frustrating 🙁

[crisp basalt]

it gets worse on weekend tbh 😬

and the weekend is where most of the flow team/google staff are off work

[astral fern]

Do you have unlimited Nano Banana 2? I can’t really find good info on limits and credits for the tiers.

[crisp basalt]

i don’t really use the image model that often, but yeah I couldn’t find any clear info on limits either, especially for the image model. but i guess this one?

[astral fern]

I’m a free user, I run into that infrequently after generating 4 images lolol.

[crisp basalt]

have u subbed to pro before? if u havent, i have one offer left for the google pro plan 4 month trial

[astral fern]

I have, unfortunately. I got a free month last year, so a worse deal 🙁

I wish they would just be more open and honest about stuff like this. It keeps me from purchasing it because Idk what I’m getting.. maybe lots of generations? Maybe nothing at all? Who knows

[crisp basalt]

well, u can always create a new google account

[candid bay]

it's not unlimited per se, but the limits are so high tier that there's a low chance of hitting them

[crisp basalt]

and claim the one month trial

[candid bay]

speaking from experience

[crisp basalt]

(speakin from experience btw

[candid bay]

and I mean easily over 100-200 images

and im lowballing with that

[astral fern]

No issue with the fact that you have the same credit card?

[crisp basalt]

🤔

well i didnt really use same credit card when i was claiming the trial everytime i create new acc

i used some kind of e wallet payment method

[astral fern]

Sorry, words are hard and I’m sick lol.

[crisp basalt]

i remmeber someone here got hit with limit and they did made like 200 images

pro plan i think? i forgot but yea

[candid bay]

im on free plan lmao. and i easily go over that

[crisp basalt]

oh non ono dont worry, the thinking emoji was just me saying "wait youre right"

[astral fern]

I’m in the free plan. I usually get capped out on 40-ish nano banana 2 gens a day.

[candid bay]

but i know limits are dynamic and they change based on demand and other stuff

[candid bay]

only??

[blissful abyss]

Is there cap for flow image generation

[astral fern]

I use it to generate some inspiration for Planet Coaster 2 builds.

[candid bay]

over 100-200 roughly or more for free users

[blissful abyss]

nice

I dont think I can hit that limits

quit alittle too much for me

[astral fern]

Yeah 🙁

it’s annoying. Sometimes a little more, sometimes a little less. 40 is the average for me.

Imagen 4 and Nano Banana 2 I haven’t hit a cap on.

Nano Banana Pro is usually around 40 for me

[blissful abyss]

a day

[crisp basalt]

safe to assume that the limits change based on.. hour or during peak hour

[blissful abyss]

Me too

[candid bay]

oh! you mean nano banana pro... i was talking nano banana 2 lol

[crisp basalt]

they did mention that ultra users get high priority for nano banana pro model only but it wasnt very clear

[astral fern]

I miss Imagen3

[blissful abyss]

But I noticed that my generations fail as I go closer to my limits

[astral fern]

https://tenor.com/view/waah-waa-sad-wah-waaah-gif-25875771

[candid bay]

Imagen 3 was the goat

[crisp basalt]

also if anyone havent done it yet

[astral fern]

Imagen 3 had this, stylized but realistic look that Nano Banana Pro almost gets.

[blissful abyss]

For what It depends

[blissful abyss]

I guess it depends on the prompt

[blissful abyss]

I prefare nano banana 2 for realism

[candid bay]

candid realism. idealized, beautiful but not too perfect that it looks bad

[blissful abyss]

But nano banana 2 trying

[astral fern]

I’m haven’t yet because it says you can’t use ImagenFX after you transfer and I’ve been relying on ImagenFX when Flow is being dumb

[crisp basalt]

you can still use em

i can still generate stuff on whisk and imagefx after i opt in

[blissful abyss]

It depends on prompts

Flow is my best others are just uhhh

[astral fern]

Nano Banana Pro (NBP) is sooo close, but just not quite.

Nano Banana 2 (NB2) always takes prompts literally in my experience

[candid bay]

nano banana 2 is nice and realistic and accurate but sometimes people come out looking a bit too average or ugly lol

[blissful abyss]

Do u do influencer images

[echo inlet]

You mean perfectly normal and not false?

[astral fern]

Imagen4 was the worst about recycling faces. It always spit out the same 4 men’s faces

[blissful abyss]

Maybe u can just have a board of your main character in different shot then use that for reference anytime u want to create something with that character

How does this look

[astral fern]

Which generator is it?

[blissful abyss]

flow

nano banana 2

Thats why I said it depends on the prompt

[candid bay]

Not really. A bit too average actually. Handsome men and beautiful women in prompts come out looking too average or lowkey ugly

And I mean with simple prompts. Wasn't the case with imagen 3. they looked beautiful but in a normal, believable way. Imagen 4 made them too idealized and fake looking

[astral fern]

I used the same prompt for both images. This is NBP

This is NB2

NBP captures the art style a lot better in my opinion

[candid bay]

looks clean and nice

[candid bay]

what game is this?

[astral fern]

Planet coaster 2

[astral fern]

Simple prompt: “Built in Planet Coaster 2, the entrance to a section of a park themed to South America.”

[echo inlet]

it's not really a test for NBP or NB2 is it?

[astral fern]

I didn’t say it was.

I generated a few dozen images in both engines a few weeks ago, I preferred the way the NBP images came out, the NB2 images tended to look cartoonish

[echo inlet]

What I'm referring to is that you're asking it to just replicate graphics within a game with a very simple prompt. So, its interpretation is that the style of the game is cartoonish because they are, so what do you expect.

[astral fern]

I gave it a simple prompt intentionally. I wanted ideas to base it off of, not a perfect replica of it

When I made the structure in game I used the NBP image as an inspiration not a blueprint.

NBP looked like screenshots from Planet Coaster 2. NB2 looked like illustrations of Planet Coaster 2’s art style

This is NB2, it just doesn’t have the same art style as PC2

[crisp basalt]

lets goo

[astral fern]

My usual prompt layout will look like this:

Style: (What the image should look like), disposable photograph with slight yellow and discoloration

Scene and Subject: (What’s going on in the image) outside at a park, captured mid-fall, an overweight elderly Asian woman wearing impractical shoes tumbles down a flight of concrete stairs as her grandson laughs while standing at the top of the steps.

Lighting and Pallete: (How the lighting should look and if I want specific colors to pop). it’s a bright and sunny day, natural sunlight shines from above. Gray concrete, red and yellow autumn leaves.

Cinematography (Where and what the camera should be doing) low angled camera at the bottom of the stairs, slightly Dutch angle, wide framing captures the scene

If it’s a video I’ll add in:

Actions Beats (these are what your characters should be doing, and bullet pointed.)

1. The old lady trips over her highly impractical shoes.
2. The old lady tumbles down the stairs, her grandson laughs at the top of the steps.
3. She lands at the bottom, hard.

Audio (what you should hear in the video). *The elderly woman curses in Mandarin as she tumbles, grunting in pain with every thud. Her grandson laughs obnoxiously.

[crisp basalt]

waittt this one is better

[crisp basalt]

When i cant access project genie and flow is the only app i can try simulate the project genie-

also hope u dont mind me trying out all your images 🙂

[astral fern]

You good you good. It looks good though thank you 🙂

NB2:

NBP:

[crisp basalt]

oh

nb2 really exagerated those "realistic" details

while nbp generates just right amount of realistic

[astral fern]

That’s why I like NBP

[wintry solar]

which tool shall i use which use my voice real time making generative video

[astral fern]

NBP

[blissful abyss]

Thanks

[astral fern]

Oil on canvas with NB2

Vs NBP

[astral fern]

Also, Sora will be missed. Here’s a comparison of VEO 3 fast vs SORA2

VEO

Sora 2

[gloomy verge]

Hello, can I ask, if Flow support negative prompt? What I don´t want in image?

[astral fern]

You can write out:

A picture of a cat laying on a pink blanket. Do not generate a white carpet.

The “do not..” being your negative

Or you can hit enter a couple of times for a new paragraph and write

Negative prompt: White carpet, puppies, fireplace

[gloomy verge]

Okay, thank you, I try it.

[gleaming lynx]

did anyone see the openAI models in arena

[gritty cloak]

?

[blissful abyss]

arena.ai u mean

[gleaming lynx]

yeah

[mortal finch]

SORA 2 works?

[brisk cosmos]

Hello i need help

I have clicked archive in a collection of loads of pictures I created and it does not appear anywhere

[astral fern]

No, it was an old video I had generated in the past.

[quiet wadi]

hello

if I purchase the Google AI plus subscription how many times on a day I could generate images on nano banana pro

[analog radish]

Did anyone experience drastically worse voice and sound behavior from the models?

[vernal remnant]

Yes, definitely, the same prompts, do not produce same results as before, the sound crackles, or goes to a high pitch, sometimes like the mic is maxing out.

[undone jolt]

Are there any commands to simply force Flow not to put in any audio at all?

[unique abyss]

No, but you can just strip the audio out in any editor. That would be nice for some one-shot videos though.

[quiet wadi]

what kind of suscription you recomend me

the plus or the ro

I want more images of nano banana pro

[lilac citrus]

is there any way to have access to the masterclass that was the other day?

[edgy holly]

More of an intro, but they said they'd share it at some point.

[lilac citrus]

Sweet thanks!

[subtle cedar]

where are image interrogations in FLOW? (eg: drop image in, get plain english description of image like in whisk)

sometimes it's hard to tell the subject matter in an image when there's lots of variables at play. the interrogations essentially give handles/anchors to manipulate.

[pale locust]

Hi Everyone, I had a collection in Flow and I have clicked on "Archive"

Now I cannot find the collection anymore. Any suggestion?

[potent belfry]

Is there a way to scream at the people who created VEO which Flow uses?

[pale locust]

I think there is a bug, I'm also facing the same issue

[iron matrix]

where do you see an archive button?

ahh dang, is it on collections?

thats my bad... I'll remove!

[pale locust]

yes, sir

in collections, click on Archive and never seen it again xd

[north wigeon]

Anyone know how to make ai characters dance like this https://www.tiktok.com/t/ZP8g1ph2k/ ?

[crisp basalt]

i know and its not veo/flow

[north wigeon]

Can you please tell me what tool is used?

[crisp basalt]

Kling

[north wigeon]

any specific kling mode?

[crisp basalt]

Motion control

[north wigeon]

thanks so much ❤️

[wintry solar]

which mode is free at kling

[crisp basalt]

Yes

[wintry solar]

image to video is not free

[crisp basalt]

Yes

https://tenor.com/view/donald-duck-sleep-gif-12073235593935901891 weirdo

[wintry solar]

night night

[twilit harbor]

Y9

[unborn ocean]

https://www.youtube.com/watch?v=bJxFOMehcRU&t=254s

[gleaming lynx]

switching to kling for video till the next update

[cerulean bone]

Wow lot of role, how to get that?

Also want to ask..

1. Is google labs flow didn't have any event? Such as giveaway or airdrop 😁
2. I didn't found flow on google play, is it unavailable rightnow?

[iron matrix]

1. We do not as of yet
2. Its currently just a website, go to http://flow.google/

[wet flame]

Switching to Kling is a good move tbh

Kling 3.0 is goated

[cerulean bone]

Oke.. So how about the role? like beaver otter snail etc

[warm orbit]

hey everyone, was there an audio update to ultra users in flow? i saw the update announcement but it's gone now🤔

[zealous topaz]

Hey Guys! l've been having constant issues generating images in Flow for the past few days. I keep getting an unknown error, which makes it basically impossible to use Nano Banana within Flow right now.
Also, the upscale option is constantly failing. Everything was working fine until last week. Who can I contact to have my account checked? The feedback option directly in Flow isn't working either, it always gives me an error and won't send. Appreciate any help!

[wet flame]

This is the exact reason why I'm not switching to Flow from Whisk

[summer kayak]

Why is Google VEO adding watermarks to my videos on an ultra subscription??

[wet flame]

ironic isn't it

[uneven bane]

This is also the same thing with me. I’m pissed.

Why the hell would I switch to Flow? Image Fx is amazing as it is, this thing sucks

[gleaming lynx]

VEO is really outdated
kling vs VEO

and I think NB is also going to get topped by new image models

google needs to release new VEO ASAP

@high fulcrum<@&1366831844502274180>

I find Kling omni better than 3.1 quality

[sturdy merlin]

HELLO EVERYONE 👋👋 👋 .
my new song is released .
It is a wholsome Feel Good Musical Video .
NO Rush , NO Tension , NO Pressure
Just pure calm and beautiful Musical Video
https://youtu.be/H2v-k5rondI .
when you guys get free time please check it out and let me know your opinion on this song please .
I HOPE everyone will Love this Musical Video

[robust egret]

how do I make changes in flow without decreasing the quality of the picture?

[wet flame]

tbh Flow has some awesome realism but the failing attempts are just nuts

[winged plinth]

Veo always quality.

[wintry solar]

tbh veo did great job, was that kling paid?

[wide ether]

Serious question: How the heck are we supposed to generate videos completely devoid of music and random production sound effects????? All this crap being added to the videos is SO annoying! I just want to generate videos as if someone was capturing a video candidly .... no production crap added to it ... it seems either music is added or random production sound effects are added ... HOW CAN YOU TELL IT TO NOT ADD THAT STUFF???

[celest lichen]

dude, take a youtube tutorial. every problem you face = youtube tutorial. easy

[wide ether]

link(s)?

[celest lichen]

bro, are you lasy or smh? cant you search it by yourslrf?

and also, it depends what youtube tutorial you need based on your problem

[wide ether]

problem is plain and simple: I state that I don't want music: "Audio: A quiet house, soft fabric movement and her movements on the table, no music" .... first thing I hear ... music blaring 🤦🏻

[pulsar coral]

Yeah, that one and every other Flow issue users have been complaining about for the past two months keeps getting brought up daily by users as seen above, which even if stronger prompts could control it shouldn't be that hard to do beyond [Music: Do not play any music at all]. Problems also include all the complaints about error messages giving the wrong causes leaving people to guess what’s even going on, with no word from staff.

Not to mention, not a single thing has been done to address feedback from users who were brought over from Whisk asking to add the features to make Flow more like it, or better yet not to take Whisk offline.

And now, to make it even better, I’m 100% sure they’ve increased the censorship for Nano Banana 2/Pro to the ceiling, because none of my old prompts that were working just fine two days ago work anymore. Now I’m going to have to waste time rewording them for no good reason, even though it was actually really tame stuff like that beach Kagome image I posted the other day. And Whisk receives them just fine.

At this point, it feels like they’re either sabotaging themselves and making the tool unusable on purpose, or trying way too hard to make it fail for some strange reason… or simply that most people running this Discord aren’t actually in charge of Flow’s decisions at all.

[crisp basalt]

ai nowadays makin ppl need to be spoon fed everytime its crazy 😭

[celest lichen]

then, i think its a flow problem. you should report it in #1238621511401734254

[pulsar coral]

idk I man I saw that music disabling prompt you had to do the other day... and in the rest of video generating tools it's not that hard to do beyond asking for no music.

After seeing that I don't even wanna try and will just do videos with music.

Which are nice anyway

Cause you're like really good at this and if you had to prompt it that much, what chance normal users have.

[gleaming lynx]

yes both paid

dont tell me you like veo's output better

[rough elbow]

Hey guys, anyone keep getting "Upscaling Failed!"

[crisp basalt]

kling is better

[rough elbow]

Is there a fix for this? I've found the regenerating the image sometimes helps but now no matter how many times I regenerate the image it continually rejects the upscaling download

[crisp basalt]

was the image generated/edited using nano banana ?

[rough elbow]

Yes Nano Banana Pro

using google flow

[crisp basalt]

wait try this workaround

click on that said image u wanted to upscale, it should bring up the "editing" ui right?

[rough elbow]

Yes

[crisp basalt]

try prompt "same. no edit" in the prompt box , and then enter, it should generated same exact image without any changes

then try upscale the \photo again

[rough elbow]

Yeah that's exactly what I do

Regenerating the image seems to work

I think it might be an internal rate limiter, if you do the regenerations too fast they block it

Thanks for the help anyways @crisp basalt appreciate it

[crisp basalt]

yeah it could be but im glad the regenerate the image workaround working for ya

[gleaming lynx]

did you try save to project

then upscale the exported image?

[balmy surge]

@gleaming lynx I can help

[gleaming lynx]

with what?

[balmy surge]

with whatevers going with the save situation

[gleaming lynx]

i am not facing any problem

[balmy surge]

oh ok do you want to be friends

@gleaming lynx name Grayson Graves
Phialdelphia US

[olive wraith]

Hi

[gleaming lynx]

dude stop doxxing yourself

what do you do for work? are you into sales & marketing
or content creation?

[balmy surge]

@gleaming lynx content creation sorry i doxxed myself

[wet flame]

Ah, that explains why my prompt "violates" their policy

It was really nothing explicit but it just results in that

[pulsar coral]

... yeah these last two days there surely was a massive increase in censorship for Nano Banana 2/Pro, which is funny because they'd usually wouldn't go out of their way to work during the weekends.

But here we are someone really got bothered by whatever snowflake thing enough to work even during holy week.

[wet flame]

ah, as we speak, there we go

[pulsar coral]

There also was an exaggerated increase for blocking copyrighted material for sure, like it took me this long to make this, but yeah I still know how don't worry.

What should we work on today? Fixing the error messages people complain about? Nah man. increase censorship bar to %200 and let's call it a day.

[quick breach]

hello world, how does new queue system works in flow? I get noticed that "You can queue a maximum of 25 generations" while i have 7 generations running
thanks in advance

[novel leaf]

Is Veo 3.1 fast still unlimted on Ultra Plan?

[analog radish]

Yes there is a version of "fast" which is unlimited

[novel leaf]

Thank you sir!

[frank tartan]

hey, can soemone tell me if this is normal :

limited for pro but no usage

[grave sparrow]

https://youtu.be/tvwD_RlbMho

[slender bone]

fake larper

[slender bone]

whats wrong with this?

[frank tartan]

?

[fringe spindle]

Good morning sir, I don't know u

This is a big thing that will revolutionize the world, spoiler: I discovered water and oxygen

If you want to buy 0.02l of water you know where to find me

[slender bone]

i dont get this joke

[quick breach]

why no one cares?

love is dead here

[undone summit]

Hi fam. I wish you all a happy and healthy life filled with blessings!

[brazen oar]

https://youtu.be/mw2BLJwcGVE?si=GJF_dn5rp3bGzd-I

[fringe spindle]

It's not a joke

[slender bone]

https://tenor.com/view/dexter-james-doakes-erik-king-stare-staring-gif-1343162792415382650

[neon raven]

yo guys, is it me or ingredients to video is faster than frames to video?

[fringe spindle]

Kebab

[fringe spindle]

Yes

[slender bone]

😋

[slender bone]

yeah it is

[fringe spindle]

[fossil bridge]

is flow down for anyone else

when I generate an image get to 67% and prints error failed somethign went wrong

[subtle cedar]

Where are image interrogations in flow? Eg: drop image in, get verbose text description. We use these for iterative prompt design on so many levels. I'm willing to pay monthly to get them back somehow. They were in whisk. long workarounds w/ai studio or other are acceptable. we just need to drop an image in and get as many data points in plain english as possible to use as handles for manipulation. the interrogations define the syntax which are the handles.

[pulsar coral]

You can put it at the labs-bugs-feedback area.

Cause apparently the excuse not to listen to users feedback (which are more like desperate pleas at this point) is that it's not being posted there and they don't read much around the other areas.

[pulsar coral]

Great now we need to go back on topic. But thanks for lighting the mood man. Just mad about all the new censorship from Nano Banana 2.

Cause now I need to rewrite like 20 prompts.

Yes I'd love to be touching some grass instead, but I work with those images too.

Funny thing is this just made original Nano Banana even more relevant again now, cause all prompts Flow's Nano Banana 2 is rejecting still work just fine for original Nano Banana.

And at Whisk by default, in fact they still work better there. I've spent a month REALLY trying to get used to Flow's Nano Banana 2, and I did just about get the hang of it just for them to change everything again.

But I also learned to make cute food menus with little gingerbread men tho... seems that's mostly what Nano Banana 2 will be good for now.

Then again not so sure tho, what is this guy so "happy" about if you get what I'm saying

Is that a third leg or what, seriously NB2 can't even do this.

How can I write it not to do that, considering if I prompt it not do that it won't make the image either.

[frigid pier]

Why does Nanao banana fail so often?

It's really annoying..

[pulsar coral]

idk the errors could be whatever now, not even paying is going to save you from not getting anything from it more often now.

Tho if you're just making cookie monster menu images it should be working

He probably just likes biscoffs a little too much, who am I to judge

[pulsar coral]

I already did the meme part today, now it's your turn to put it back on topic.

I'mma go try to fix my NB2 beach prompts

[fading dirge]

If veo 4 is as good as nano banana is for image gen, its over for every one

[pulsar coral]

There are so many interpretations to that atm.

[crisp basalt]

https://tenor.com/view/grass-gif-6112271. x

[pulsar coral]

Gimme a sec

There now we're still relevant

... man I'm just tossing images at this point, and yes they do cost money. Just gonna sleep it off and charge back tomorrow.

Admittedly NB2 catches on to prompts this easily... not much science to it

Tho for some reason some people seem to have a lot of trouble getting images to do this sort of thing... sometimes it's just incredibly logical with it's understanding

[crisp basalt]

https://tenor.com/view/that's-what-i'm-talking-about-maya-choi-obliterated-that's-what-i-mean-celebrating-gif-5158548592563686051

[wet flame]

Sometimes you just gotta make the prompt simple

[crisp basalt]

^^^this

[wet flame]

I've tried some selfie shots in Whisk, but I need to prompt more to get the pov right (the selfie pov, with the hand on bottom of the frame). But NB seems to understand "selfie" more, so the prompt for the hand isn't really needed

Less is more

Although Flow sucks rn but thankfully image generations are free, so you can trial and error from the simpliest prompt

[sand thorn]

hello

[wet flame]

hello

[sand thorn]

so i made some new FGO Servants

[gaunt vessel]

anyone help?I was doing my project on flow an hour ago,it stucks,so i change my VPN site

am i banned?

now i can''t login

[frigid pier]

i am sure veo 4 will shock the world.

but the best image model will be GPT image 2

[blissful abyss]

A bottom left is meant to appear that says go to project

[tepid zephyr]

Give me suggestions

https://youtube.com/playlist?list=PLhWDUs4MpRSf6M7LQWwVyq10JClaC9_2x&si=c3Upw0Et6m05SYoP

[modern salmon]

Hi, I have ultra plan for Flow and still I have been getting veo watermark on videos I generate. Any solution for this please?

[crisp basalt]

It means your country is one of the selected regions required to follow AI regulations, so Google made the watermark visible for Ultra users located there.

[wintry solar]

https://tenor.com/view/stitch-lilo-and-ohana-hi-gif-1354856903718586695

[autumn night]

if I get ultra - does it reset everymonth? or stock up ?

[gaunt vessel]

Thanks but i didn‘t find it

[crisp basalt]

im not sure but yes upgrade to ultra

[fossil python]

It's reset to 25K every month

[split raft]

didn't know

[autumn night]

is falied generation normal?

[wet flame]

yes

[wet flame]

It should be reset to 25k every month, and extra AI credits will stay if you haven't spent them all

[cloud marsh]

Migration of whisk to flow, can we still use whisk

[gleaming wyvern]

Hello. how does poeple get free accounts ? or at least very cheap ones ?

[wet flame]

Flow is free to use, aint it?

[gleaming wyvern]

but you only have 150 free credit

i bought once 45k credits ultra account with 20USD
its lasts for 2 weeks
but the official price in website is 150usd

[wet flame]

That's for generating videos

Image alone is free of charge, although 100 images daily

[gleaming wyvern]

yes i want to generate videos

[wet flame]

Then yeah, you have to pay

[cloud marsh]

i hope it won't be paid by 30th April?

[crisp basalt]

lol the third party reseller

[autumn night]

quick question anyway does veo 3 follow timestamp like kling 3.0

[wet flame]

one way to find out

[frigid pier]

bro please fix the nano banana on Flow

What are developers doing?

the platform is not working properly over several days.

This pisses me off

Can anyone generate images on Flow now?

seems like all Nano banana generations fail now

[lime quiver]

Apparently, there are no developers left—just bugs.

[novel plank]

Small complaint on Flow: I did video with the new Lite model with reference images a few days ago. It worked fine for me. I guess that feature is now not available for whatever reason. That's fine. However, I was trying to use it this morning, loaded image, and it looked like the system accepted the image. Loaded image, loaded instructions, got back random people and style (because I had an image reference for it to animate).

I had thought I forgot to add the image and that it was my fault. And I tried several times with credits to get it to work only to realize after it finished uploading the image I wanted to include and tried to apply, that it tells me in a little popup if I hover over it that references aren't available for the model. But it didn't stop me from continuing even though it appeared I had a reference loaded. I would rather it stop (gray out the go button?) if a reference is loaded and it can't use it so I can know to inspect rather than it try to continue and I'm wasting credits thinking it's working. Hope that makes sense. And I hope you'll bring it back if you get it working because it was working nice for me. Thanks!

[crimson garden]

bruhh (fkin useless most of the time)

[crisp basalt]

what...?

[frigid pier]

yea this is insane..

i can't generate any images on FLow

Fix it please? fk...

[subtle cedar]

Thank you for the tip on where to post feedback. I will not post in this channel regarding the feature request.

[iron matrix]

hi again wrigley... where are you seeing this?

[proven roost]

Hey! Could someone please help? The main screen has been loading for me since this morning for more than 3 hours now, as well as my credits in the top right corner. I tried everything – clearing cache, cookies, logging out and back in, incognito, different browsers – everything is still the same. Is it the server problem?

[rigid nest]

Where can we add feedback or feature suggestions? I have few proposals, Please vote if you like:

1. Ability to move Collections to Other Project
2. Batch Copy of Photos
3. Rearrange Photo Grid

[winged plinth]

The #labs-bugs-feedback channel is your best bet for this!

[iron matrix]

are you able to open the console and send us what you are seeing in the console and network tab there?

[proven roost]

Let me know if this helps please

[proven roost]

Here's the network tab

[stuck snow]

I was trying to animate an old bicycle ad and it said there is a minor in the photo so its against the rules?

[iron matrix]

pinging via DM!

[wintry solar]

going okay, how about you

[frail kindle]

Hey… I’m pretty frustrated right now. I deleted multiple projects and can’t figure out how to get them back if that’s even possible.. I did a lot with them and I’m not sure what to do..

[marsh inlet]

Hello, I need some help with my flow. I have an active Pro subscription, but I’m having trouble generating images. An error message keeps popping up since 4 PM yesterday. I tried again at 8 AM today, but it’s still not working. Does anyone know why this is happening or how to fix it?

[frail kindle]

Am I screwed or can I restore delete projects?

[pulsar coral]

You can't restore them.

[frail kindle]

…. 🙁

[pulsar coral]

I did ask for a recycle bin feature to avoid this but there's been no word on it. Flow's UI is just full of empathy-lacking beginner traps.

[frail kindle]

What do you mean this website doesn’t have that feature?!

I lost at least a month of stuff because I was frustrated with things not loading properly.

Ain’t that great…?

[pulsar coral]

You're not the first one to lose a bunch of images, but you get used to not falling for it again.

Just this week Flow also had a fake "Archive" feature also eating people's images to some unknown void. At least that's disabled now.

[pulsar coral]

If it was working fine some hours ago maybe you just hit a daily quota for images, Flow sends the wrong error message, it should be saying something like "You ran out of image generations for today, come back tomorrow 🍌 " or something alike but instead more often says you're going too fast. You should try again tomorrow around the same hour it was working fine the last time.

Also you can check if swapping from Nano Banana 2 to Pro or vice versa gives you some more images. Your Flow account runs out of Pro images much faster cause they cost more.

[marsh inlet]

Oh, I see! I tried switching to the Nano Banana 2 model and it worked. I thought I was generating too fast and it would reset in a few hours, but it turns out I just ran out of my daily quota. Thanks for the help!

[tacit bridge]

Compute isn’t the real cost. Bad product thinking is.

[frail kindle]

Oh.. crap

[simple stone]

What if when Whisk is closed at the end of this month, Flow experiences troubles that cannot be repaired for a long time? FUBAR 🤣

[wintry solar]

https://tenor.com/view/hello-hi-hy-hey-gif-8520159980767013609

[frail kindle]

I still can’t believe I made myself look utterly stupid.

[tough salmon]

if whisk is being discontinued and moved to flow, flow must build a background feature for image generation

[pulsar coral]

Nah man, its been happening to a lot of people, happened to me a couple times too

[frail kindle]

Really? Yikes…

[split raft]

no need to worry about that. Not everyone know everything, just keep learning

[clear glacier]

what happend to flow ? did you guys had this ?

[exotic meadow]

How to get unlimited plan for whisk ?

[frigid pier]

Flow is trash

I paid for credits but cannot generate any videos there.

Fkin fix this service plz

[pulsar coral]

Yeah and it's visibly starting to get less crowded around here, at least from users who generate images, there's a bunch of white names in the chats because they left more likely angry.

And the ones who stayed, only have complaints...

[clear glacier]

brruuuhhhhh

[frigid pier]

yea i experience this too

fkin annoying. unusuable..

[kind dirge]

Hello. Did flow censor image really hard? I was doing image of a character wearing bra and panties doing pose at the mirror and notice it's randomly adding clothes to the character or cancelling all 4 image.

[crisp basalt]

oh?

does that mean 4,6,8 are finally coming to flow

[wet flame]

I do images like that most of the time 🤝
But it's definitely stricter

Oh well, not quite actually

[wise locust]

Why did they get rid of draggin images? It makes organizing collections so much harder now. There's not even a way to move images to collections. You have to copy and paste them now.

[spark hornet]

I'm fighting with this for the past couple of days .... I can't use image generation at all !

[pulsar coral]

I can still drag images, I'm on PC tho.

[brittle pelican]

WHAT IS THAT? Why there is always PROBLEMS?

[pulsar coral]

Almost all my old prompts still work the same now, just with some rewording. The level of censorship was fine before—raising it this much has only made it more difficult for users to generate average normal images too, so Google just sabotaged themselves for like the 10th time this year for no good reason...

at this point I'm just curious how much longer will everything last lol.

And I'd say pretty much until the Chinese, Open Ai or xAI give a better image generation options... and they're getting pretty close, while over here it's all pedaling backwards.

With Nano Banana 2 as it is now, once Whisk is gone I don't think many will have much reason to keep paying at the Google platforms for image generation at the least... the image models will still be useful since can one still pay for those anywhere else.

[wintry solar]

disagree

[pulsar coral]

Well it's not as good, but they will keep trying.

I'm pretty sure they'll get there, if their video platforms are better now, very sure they can also do it with images eventually.

[pulsar coral]

But as for Google's platforms, I don't see how with all the complaints, it's gonna be a miracle if it keeps being markeateable, maybe they're even losing money by now.

I've been at the xAI stuff and yeah, I'm sure they ARE getting close.

Might be hard to believe but in some little aspect or the other, it's already somewhat better than Nano Banana 2.

In fact I think if they'd fix the problems with depth perception, it'd be useable enough. Can't be helped if there won't be anything else really, and seems we're getting there.

Nano Banana 2 itself does also have heavy depth perception probs anyway, so I imagine it can't be easy to fix.

[crisp basalt]

nano grass when

[wise locust]

I used to be able to on PC but not on laptop. Now I just can't in general.

[gloomy verge]

Hello. Can I show you my very first creation by Flow? It is my very first video and I don´t know very good english, so, video have some mistakes.

[gloomy verge]

My short movie have visible some mistake. I don´t know, if I can relase it. Is it some problem, when there are some mistakes? Thank you for help.

[near gate]

Damn it. We are in the "Value Extraction" phase of this app. This week alone I've seen degrading intelligence of my prompts. I hope I'll be able to finish my project!! Oh the risks of dark patterns is high!!

[neon coral]

this sometimes happens, but i thougt it was me bru

[crimson garden]

just came here to ask if thats planned to be added in the future.. would be nice to have also 15s like Kling

[old bay]

I'd second this suggestion but I'd rather have a parameter letting me choose the duration of the clip, 8sec, 16sec etc...

[crimson garden]

yeh that would be ideal

[sinful lotus]

I've been getting this odd message while working on my sci-fi film that causes generations to fail. After asking for a generation of a scene, it immediately fails and says, "Failed. We've noticed some unusual activity. Go to help center for more info" (which provides nothing specific). I can't even then run a prompt that says something like "create a dog," and it will fail the same. I have to log out and re-login to get it back to normal. Any idea?

[crisp basalt]

i made a bug thread post about this , head over #1238621511401734254 and look for "Flow reCAPTCHA Token"

[sinful lotus]

Wow, thanks Wrigley. I didn't know about the labs-bugs-feedback area, but then again, even if I did, I had no idea about this reCAPTCHA token thing (and no idea what that is)... thanks for posting it there... it makes me a nervous wreck every time I see it!

[frigid pier]

I am so fkin annoyed by the error "We noticed some unusual activity"

I cannot generate videos nor images in Flow!

I paid my money and why i can't use it properly

[crisp basalt]

i was worried too when i got that error , usually when that kind of error shows up it means either my acc is being flagged or something but im glad im not the only one facing this problem

[crisp basalt]

recaptcha is basically a small security check that websites use to make sure you're a real human and not a bot 😅

so in this case, flow is using it behind the scenes buut somehow its misfiring and thinking normal users are “suspicious activity”, which triggers that scary error

so yeah nothing wrong with our account, just flow being a bit overprotective right now , too overprotective i'd say!

[sinful lotus]

Fascinating, and thanks for the detail! (No idea how you figured that out). It seems that if I log out and re-login, the problem goes away for a bit and then comes back eventually. Not a great workaround, but so far 100%.

[frigid pier]

so we cannot figure this out?

i cant make any videos now even tho i re-logged in

[sinful lotus]

Gosh, that is frustrating (ss8722)... I wish it had cleared for you as it does for me. It does seem from not-so-scientific observation, though, if you wait maybe 10 minutes, it might subside (it did a few times for me that way as well). In short, I'm also not sure if this is some kind of weird "reporting the wrong error" and VEO is just overloaded - perhaps mangling this captcha thing - (I'm struggling with it this morning to maintain proper accents by speakers, which is a sign of cloud overload, in my experience).

The other thing you might try is to start a new project (I know, potentially painful). My AI engine that helps me with all this stuff (Gemini) told me it may clear metadata stuff and clear the problem.

[old bay]

Hey, i've had this before, it usually comes up if theres a copyright flag. Maybe see if there's anything sensitive. Particularly with known entities like nasa or Federal agencies or other official things, I had run into this issue when I used the name Holloway for something and apparently theres someone out there with a social media account called holloway which I think is dumb, because its also the name of a university and a road...

[old bay]

Oh actually I get this often enough as well when i[m generating clips and it fails for no reason, usually when I make 4 at a time, sometimes when I make 8 at a time. I think its high usage and there arent enough resources to process.

[frigid pier]

yea

more and more people are using AI now and these big companies cannot handle the requests.

due to the high cost of computing power...

i think this proves that the current AI tech is total bubble. bs..

it's too expensive to generate things.

[old bay]

Just accept that theres high usage at the moment, sometimes it takes a while for it to render seamlessly.

[sinful lotus]

Yeah, thinking more and more, it's some kind of system overload (resulting in a wrong message):

Even this simple prompt failed (in addition to when I did one that said something really simple like "make a dog" 🙂

"The man turns to the dog and speaks with High-fidelity spoken dialogue with a clear British-Indian accent stating, "To be perfectly honest, I’m at a complete loss. It is—without exaggeration—the most bizarre thing I’ve ever seen!" delivered calmly with zero pauses underlaid by a loud rumbling engine sound."

[frigid pier]

AI business is not going to work until the scientists figure out how to reduce the computing power per tokens.

[old bay]

Sounds like you're doing something fun there. Cant wait to see the final movie. I'd love a larger library of voices and accents to save having to use that high-fidelity accent specific part of the prompt. Get some voice actors in and give us more range google and not the corporate generic library we have at the moment.

[old bay]

need more powerrrrrr!

[sinful lotus]

Yeah.... voices I don't have to burn all kinds of prompts and careful language to keep them consistent 🙂
I had hoped for those "voices" in the latest announcement they had might get there (at least in part), but my menial "pro" subscription apparently doesn't qualify yet to even check them out (beta went only to ultra I'm told).

[old bay]

yeah man, I hear you. I paid the ultra, give it time, they'll roll it out to the other tiers at some point.

[sinful lotus]

LOL, maybe if I had just ponied up for Ulta, I'd be ahead! I probably have $300 of failed prompts a month (just kidding, but sometimes it feels that way!!)

[old bay]

I'm doing lower tier model renders for cheaper anyway and I'm glad they brought back the lite low priority option which costs zero for me at the moment. It was painful when I ran out of credits last month, really boned my progress. Not sure if that options available, but you're probably using the higher quality models for better outputs. Not sure if its possible but maybe one day it will be possible to ram out a low model version of everything then render up with the higher model? anyway, its still early days so i'm sure it will get better over time.

I'm not doing anything high quality and happy with 720p renders, I output the final at 4k because it gets passed through the youtube processor anyway for upscaling, and i'm more interested in story and less about beautiful photorealism.

The audience will always accept a poor medium with a good story

Beauty and the beast was released at the cinema unfinished with colourless sketches and it still became one of the greatest animations of its time.

[sinful lotus]

A great story is worth 10 million high-quality renders!

[pulsar coral]

I imagine people now have to try much harder just to get a single image because of all the censorship and copyright flags and that added to all the other recent error messages, and that probably led to a spike in rapid clicking—almost like spam—since users keep retrying over and over. Because what else.

And that's also affecting people who use Flow slowly. Just another side effect of Google's very bad planning.

…I think they should really sit down and fix the existing error messages so they clearly state what’s causing them, instead of constantly tweaking things in Flow and Nano 2 and creating new error messages.

[pulsar coral]

Well at least we can still make memes... sort of

Like this worked only 2 outta 12.

[spark hornet]

I literally been using the same prompt style for weeks without a single problem, this week I wasn't able to generate a single image because of "policy" .... I tried playing cat and mouse with the prompts just to have one successful generation out of 50 tries ..... my god the censorship went crazy this time !

[pulsar coral]

And there it is.

Also about all the talk about high end costs and computing power and system overloads and all that, idk Whisk was working just fine and no one was asking for something else besides adding Nano 2/Pro to it and some more aspect ratios.

They could've added some more cost to it and people visibly have said they would have payed to keep Whisk running as it is.

It's when they started with this Flow migration that all started failing.

[vagrant laurel]

Flow is failing?

[pulsar coral]

From what people have been posting, tons of error messages yes. However it might reach some accounts more than others. You might not be experiencing anything.

That or one gets used to it.

I at least haven't seen the recaptcha problem yet, but it's likely because I went back to re-prompt my stuff instead of trying to play gacha with it.

But paying users do say to have seen it and posted screens about it.

There's also all the massive increase in censorship and copyright, which I'd say it was useless to do, I at least can generate everything samey by re-wording/re-prompting so that was for nothing but to make people waste time.

And it's also stumping the creation of all sorts of other images for all other users too for no good reason, the censorship level was more than fine as it was last week.

[obtuse flint]

the censorship is too much now.

[dusty aspen]

😭 😭 😭 😭 😭

everyday problems

[bright heron]

It seems like no matter what words are in my prompt if I am doing x2+ generations in nano banana 2 or pro I only get one image from each

[kind dirge]

my mostly one or no image at all

[pulsar coral]

You just need to be more ambiguous and indirect—basically dig into a thesaurus-level vocabulary. I won’t say much more though, in case they end up patching that too. I’m pretty sure they’re paying attention in these threads despite claiming otherwise, and I don’t have another two days to waste dealing with rewording again.

[obtuse flint]

do they even address this stuff are we just chumps for talking and complaining about it

[unreal viper]

hi all! weve seen a lot of bug reporting happening in here! We hear you, and we're sorry this news is frustrating. We know Whisk [/ ImageFX] means a lot to you. We made this decision to give your creativity a better, more powerful long-term home in Flow — and we're committed to making the transition as smooth as possible! the best place to report your feedback to the labs team is through #1238621511401734254 , these threads are activly monitored and are easier for the team to see rather than it being buried in this channel! Thanks all!

[brittle pelican]

Well Well Well

Nothing work, again