flow-general・🌊

[pulsar coral]

I really don't think they aren't losing money cause of it atm, and whoever wanted the changes isn't even around to see what he did anyway

Wow only one mod online, that's a new record. Predictable tho.

[pulsar coral]

.... Flow is saving every upscaled image as JPG not PNG... how did I not notice this earlier or anyone say anything about it, this is a disaster and who knows since how long has it been going on.

There's a reason the term "JPG artifacts" is a bad thing. Seriously wow.

Who is even running this ffs

[pulsar coral]

Actually 1k is also saved as jpg, saves all as jpg at the time of this post. Actually right click save as still saves the image previews as PNG, but the download button is saving everything as JPG.

[marble echo]

A person is walking on a sidewalk wearing baggy white pants and Dr. Martens with Rick Owens style shoes. The camera only captures the lower half of the body (waist down). After walking about four steps, the person slightly extends their left hand forward. On the ring finger of the left hand, they are wearing a ring inspired by Paing Takhon’s famous style in Myanmar. The video duration is about 10 seconds.

[waxen arch]

i'm trying to generate in google fllow, with a pro account. it shows an error "Generating too fast please wait..." anyone faced the same issue ? Can anyone tell me why ?
Help center is unable to provide me with a solution

[wraith niche]

give me the prompt u r using maybe i can help you.

[waxen arch]

I was using veo extension, It was working fine. now there's this problem. It's a batch processing multiple images

problem is not the prompt but random prompt too is not generating

[marble echo]

Is it okay to do what I asked you to do?

[wraith niche]

i think your browser or innternet service provider is not working properly.

try it in private browser or incognito mode

[waxen arch]

Logged out and in, Cleared history & cookies. Let me try what you're suggesting. Will update here

[wraith niche]

okayy lets see....

[waxen arch]

thanks

[waxen arch]

It worked, why do we need to change to incognito ? Any idea ?

[wraith niche]

idk 😆

i think there's no extra load on incognito browsing.

that's why it works perfectly on private browser

[waxen arch]

i can't use the extension in incognito though.

[wraith niche]

without any error

[wraith niche]

change your internet service provider.

[wraith niche]

i know they dont work inn private browsing

[waxen arch]

Understood

Thankyou

[wraith niche]

its okay, next time whenever u face any problem.

just ask me, i'll help u

[violet stone]

Incognito doesn't work for me lol is it browser specific

still get the wait

[wraith niche]

change ur service provider man

[violet stone]

Alr

[wraith niche]

it will help u

if u still face problem, give me the prompt i'll try.

[violet stone]

it doesn't have anything to do with prompts its the nano banana pro

this lasts over a day which isn't a "Moment"

[wraith niche]

u r facing problem in generating the images

[wraith niche]

have u tried incognito or private browsing ??

[violet stone]

Yes

[wraith niche]

bad luck for u broo

[violet stone]

yeah

hope they fix this soon

[wraith niche]

yeah..

just hope

[wraith niche]

which type of images r u generating

??

[violet stone]

IMG to IMG and normal image generation

[waxen arch]

i think the browser needs a reboot
log out, clear history & cache, login incognito. Try google flow only without extensions. Check if that generates in incognito.
Close browser and login again. It worked for me.

[violet stone]

alr

yeah didn't work imma just wait on a update

[shy pumice]

Has anyone successfully migrated to the ImageFX data flow? I am so flustered that the transfer still hasn't happened when April 30th is just around the corner.

[glad yacht]

Both Whisk and imageFX library had been successfully migrated to Flow library. But the access speed for Imagefx library is horrible slow....
They put 1000+ images all into a folder, take forever to render images, since all flow renders everything full resolution. I wish they could adopt Google photos library shrinkable preview size and remember the settings for each folder. Take almost 20+ seconds to open my imageFX library 😅...

[exotic shell]

"We noticed some unusual activity. Please visit the Help Center for more information."

is that normal nowadays?

[crisp basalt]

no

do refresh and try again

yes it has been posted in bug thread , yes theyre trying to fix it

[wide ether]

I'm seeing an uptick in that error too 🤦🏻

[exotic shell]

anybody experienced with the "new" audio feature and consistency? im just trying it right now. it says you have to upload a reference image. i dont see any way to upload a reference image on that page. i can upload an ingredient but then the image is going to be used a reference for the visuals. or is it to be intended to be like that? but if you just want to take a reference image to make the sound the same but the visuals should change... it wouldnt make sense.
hmm

[crisp basalt]

Reference is ingredient

[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!

[gleaming lynx]

NBP is still better than GPT image 2

[unique abyss]

As wrigley noted the "reference image" is just a visual ingredient, and ingredients appear differently from scene to scene, depending on the text of your prompt.

[pulsar coral]

GPT's model is very similar already to NB1 tho, still not as good but they're visibly getting there. Google seems to have the advantage of being able to use their search engine to train the models, but with the recent update to the Nano Banana2/Pro model they completely ruined that advantage.

Like Google seriously pulled it's own rug with this update.

It really doesn't seem it will take longer than this year for other image models to catch up, they just need to feed the same amount of info to their models as the Nano Banana models have and that does take time and resources.

the actual contenders are gonna be the wan and grok image generators or something chinese tho, since GPT is gonna be too censored despite creating somewhat unsafe images already.

[gleaming lynx]

yes 100%

i am gonna try midjourney for cinematic stuff

[glad yacht]

As in artistic styling Gpt2 won against NB² /NBP

[gleaming lynx]

no it didnt

[glad yacht]

You can't prove it! 🤔

[unreal viper]

hey guys! while we love the conversation lets keep it Flow focused please!

[exotic meteor]

In the same boat. Transfer opted in to on the first day it became available, no sign of image transfer as of today despite it saying in progress for about a week or so. I easily have 1000's of images though so that may be a factor.

I would assume the transfer will continue after the 30th if not complete, otherwise how would people who opt in to the process on the 29th or 30th get their images.

[zenith garnet]

The bonus credit points I earn every day... I can't get anything back... Why?

[pulsar coral]

Yeah GPT is more freely artistic which is much better for OC characters and following the reference images.

To keep it on topic as requested compared to Flow's NB2, Nanano Banana2/Pro is indeed way more rigid with the body types either keeping them as reference image or cartoon styled small, but does not do this with the faces. It goes so accurate it looks in it's own database for faces it thinks matches the reference image and changes it into it.

And all this is actually causing all the bugs about "prominent faces" and underage, it trips on itself due to the way it itself chooses faces and body types, I did post examples of this a couple days ago.

Like here's a prompt without reference image at the new DALL·E from GPT (that's what it actually uses) it chooses adult by default and does indeed go very artistic with the interpretation,

And there's a prompt at NB2, it chooses childlike cartoon type by default without it being requested, and that's a HUGE problem since whoever tampered with the filters this month just seem to have pushed buttons and went on vacation without even looking back to check what he was doing here.

[stray remnant]

anyone else having trouple downloading your picture upscaled? can download it perfectly fine but when i try to upscale it fails

[pulsar coral]

Try clearing cookies or changing browser, if that doesn't fix it you just hit a daily limit for upscaling, you gotta come back tomorrow around the same time the upscaling was working. At free tier accounts you can upscale around 199 images daily for free, more or less according to daily resource. And yes at least for me it was impaired 199 for some reason.

[stray remnant]

im a pro account and cleard my cookies ill try again

[pulsar coral]

Pro does have way more upscale available, count your downloads from today of 2K images and you'll know around how many you upscaled. All should be fine and that's how it works tho.

Pro might have at least more than double the daily upscale number.

[stray remnant]

yeah i know im not over the limit just starting working today

ill try the new browser

[pulsar coral]

Yesterday's images also count, it doesn't have a set daily reset time zone, it just ends as soon as you upscaled your daily limit of images, then resets around the next day when it was working fine the last time.

it's that cruel yes, if it won't work can't be helped anymore today.

So more often it's better to have a set time to start your workflow so you have a reset again around that time, I'm also uploading like 100 images daily to my art accounts and had to learn to do that.

[stray remnant]

ah okay

[pulsar coral]

Funny enough it's all images from when Nano Banana 2 was working somewhat decently, haven't been generating more but I did keep enough to work in time.

So yeah one wakes up, upscales everything made yesterday, once you hit the limit with that upscale error again you either work on something else or make an upscaler with the canvas mode from Gemini if you're in a hurry, that one's as best as it gets besides Flow's.

That way you'd be sure you won't be stopped midwork.

[lapis linden]

Suggest me how to make video

[small mirage]

Are you sure it's DALL-E ? isn't it GPT Image 1.0 or 2.0 ? I heard GPT now uses GPT-image versions after 4o model.

[pulsar coral]

GPT itself told me that, it might have been mindgaming me tho.

[pulsar coral]

Seems both are correct really, DALL·E is what's used for it but they didn't name it DALL·E 4/5 for some reason.

Just gonna call it GPT Image 2.0 like all do to avoid trainwrecks

[crisp basalt]

you did this to urself lol

anyway

[pulsar coral]

Pretty much yeah.

The thing just knows something we don't, tho it doesn't matter much really.

I guess it's kinda like how Flow's NB actually comes from evolutions of older imagen models

Even tho no one calls Nano Banana models "imagen"

Or almost nobody, pretty sure some people around here also told me Whisk's Nano Banana 1's precision model was just imagen 3 or 4.

Like there's this dude around here who's been since imagen 2 still existed and lost all his migrated images to Flow.

NB1 did definitely have two version updates itself tho, at some point maybe Whisk did use the previous one. Tho that's not what my migrated images state to have used.

So no idea there either.

Can't get these skies to totally look right now, something always comes off cartoony at random

That's more like what I'm trying to get

First two were NB2, This is an NB1 sky in comparison, wouldn't know if I can call it better, but it sure understands it differently. Well good night moon.

Even if it's just 1K blurry you can tell it knows the moon should look closer to you than the constellations.

[pulsar coral]

The "Suspicious activity" error doesn't happen to me anymore when I try to generate 25 images at once from a free account. Might not be the same for all locations tho, but it def feels better or totally gone now.

You might still want to wait till all of them generate/fail to generate or about 1 minute before requesting around 25 more or less tho.

[hazy kestrel]

Is flow sucking today or was there some bad update?

[severe lily]

I'm using Firefox and any new project I make is now not saving when I exit the project. It clears anything uploaded and I've lost work. I just signed up for ultra and this is now an issue when it was working fine before I did.

[old bay]

got a text message on flow today saying its not working today as expected and may experience some difficulties.

[candid bay]

is anyone else getting this? no matter what i attach or create

i can't even see what i create

[frosty niche]

I was getting high demand/usage notifications during this same time, so it may be because of that. If it continues, switching your browser for the meanwhile can help.

[spare marlin]

So, uhm... After waiting a few hours after the last generation I did, I was given an error message saying "You're requesting images too quickly, please try again later."

But that can't be right!

I waited long enough for cooldown.

[winged ivy]

It seems flow is broken. I guess it was too good and had to be stoped?

[signal kettle]

Why is flow giving me an error about generating prominent people after asking it to generate a person? I'm assuming this is to protect real people but I'm not even including a reference???

[jade bison]

https://tenor.com/pAAoLNKlEAd.gif

[errant plover]

Respectfully.. Gemini in conversation works better than flow

[pulsar coral]

... this place doesn't work but Google's under the idea it does for some reason. Meanwhile Whisk has no problems and is the one being taken offline. I don't think anyone is really in charge of the decisions around here, nothing makes sense.

[crisp basalt]

like how you thought that chatgpt new image model is still dall e? yeah that one doesnt make sense either

[pulsar coral]

You also told me imagen 4 is the same as Nano Banana 1. Sometimes things are as they are.

Like you weren't wrong about Nano Banana being some form of imagen model, like how GPT's model is also a form of Dall-e. It's info that comes from AI itself.

it's no wonder at Flow when you run out image generations for the NB models, you also can't generate images with Imagen 4 either despite it being there as supposedly a different image model, because they are connected and technically just a form of one another. Dall-e also had that problem in the platforms where it was available alongside GPT Image 2.

Like the thing hasn't been out for more than a week yesterday, no wonder the code names hadn't settled.

... that said, that has nothing to do with how badly things are running at Flow. Which I wish was just my idea.

Now I remember, you told me Flow's smear tool was the same thing as having Whisk's precision mode...

Eitherway, all's gonna burn about Whisk in a couple days.

[winged ivy]

^^^^
Ugh, Whisk really is better. It even has useful and interesting features that Flow dosn't.

[pulsar coral]

Now that all the images from Whisk migrated to Flow it's much easier to see what model was being used from the start to create them, not just some editor.

[pulsar coral]

I imagine they just can't reproduce it or lost the prompt that created those dice modes

It's been years since it was made, and seems it was some sort of random inspiration.

[winged ivy]

How odd

[pulsar coral]

Also even if they had the prompt, Nano Banana 2/Pro visibly can't reproduce the same kind of art style composition as NB1 so easily

I mean I'm sure it can, but it will more often tend to make the images seem more like new digital art, NB1 tends more to vintage styled art

Yeah tbh all that can be done in a couple days is adapt

[crisp basalt]

ok mod sorry lol

[winged ivy]

Well hopefully flow will actually work, as currently it doesn't.

[pulsar coral]

I'm pretty sure you wouldn't need NB1 or those modes, but visibly so many people need them.

I never used them much either... but tbh I used to work with LoRA prompts so can't even feel it

I think it'd be possible to reproduce the dice and style things from Whisk at Gemini canvas to test how'd it go for Flow.

But there's no way it would take only 2 days, Whisk would still need to be there, at least for me try.

The styles/scene thing is so random and has tons of combinations

It'd take at least a month of testing... unless someone like really gifted at it would try, or someone still had the prompt laying around.

.... then again I totally forgot Wrigley did show me a similar thing he's been working on.

[crisp basalt]

https://tenor.com/view/wtf-what-i-do-gif-18306771951110025625

[pulsar coral]

How much did you test that thing you showed me? Is it really doing it like Whisk's styles/scene? seemed similar

... idk don't care much for creating images that way, but people can't get over it

probably best thing to do is get close to how it works.. or better but it'd probably be hard to make it better, much worse with NB2.

[crisp basalt]

the whisk auto caption?

[pulsar coral]

yes

[crisp basalt]

what it does it just output prompt, not image if thats what u meant

[pulsar coral]

...hmmm well they've also been asking for that

[crisp basalt]

basically it analyzes images and spit out prompt based on the image it analyzed

[crisp basalt]

well i think its safe to say that, theyre working on that part

[pulsar coral]

I've sent people to Gemini and GPT to get the prompts from an image, but when try they say Whisk's works better

[crisp basalt]

its not gonna be entirely similar like , what whisk does but it will have gemini as ur "assistant" in flow

[pulsar coral]

yeah that's the problem really

It's so darn specific

[crisp basalt]

its specific to the category of the image type

my english is broken hold on\

subject
scene
style

[pulsar coral]

I think Flow's team might think uploading all the images then telling NB2 to use one as style would work

But people can't seem to get it right that way either

But maybe that has more to do with NB2's model than what's being asked from it

... and' it's impossible to try with NB1 here anymore, gonna have to try it elsewhere

[crisp basalt]

i think its just prompt structure that needs to be refine

[pulsar coral]

Well adapt or die it is, like some people say at feedback they're literally running businesses with all this stuff that's about to go down and I believe it cause I do too... this ain't gonna be good in 3 days.

[crisp basalt]

https://gemini.google.com/gem/176dRicJ74XR85ux0WZU8SlBk2EXQPkpG?usp=sharing

try this idk if its gonna work for u

[pulsar coral]

Yeah these are probably as best as it would get

The images better be big and clear unlike this one

[crisp basalt]

i think whisk created longer captiojn.. let me refine

[pulsar coral]

Gonna compare the prompt at Whisk and Flow now... while can be done

[crisp basalt]

that actually looks

nice

[pulsar coral]

Yeah this is what people have been doing a lot

[crisp basalt]

is whisk nb1

[pulsar coral]

None of these use a reference image, so they can get it

[crisp basalt]

or imagen

[pulsar coral]

Yeah it's Whisk's precision mode, which according to themselves is NB1

Flow's NB2 gets closer to it because it probably knows this meme pirate image

NB1 goes almost completely original about it

[crisp basalt]

actually

can u use that same prompt in flow, and choose imagen model

try it

[pulsar coral]

Ah that's good idea

That's imagen 4, it does look more similar to what NB1 does

[crisp basalt]

well

[pulsar coral]

Probably cause the model doesn't compare to it's database like NB2 does, yeah this is progress

[crisp basalt]

try the prompt of that anime girl u tried to generate

use imagen 4

[pulsar coral]

I found it, ok then

[crisp basalt]

see if using imagen 4 gives u the output u expected

[pulsar coral]

Imagen 4 just refuses to run it. This is other problem with this thing, there's nothing wrong in this prompt. NB1 obviously wouldn't refuse to run it.

Without reference image it does work, sometimes it does look ok but more often picks digital art style, it should be doing vintage like NB1

it's just ugly

Well sometimes it will work sometimes not

[crisp basalt]

Actually why did imagen even exist in the first place

No, I mean, is it a different department or research team that works on the imagen model, or was the imagen model just an experimental study, which is why the nano banana model is here?

[pulsar coral]

Imagen is visibly similar to the first Dall-e models, at first it would only create images from prompts like the other would do, but it'd lack reasoning. Some "Brain team" would work on it. One woulda needed to try the Dall-e models when they came out, it was indeed very similar to how imagen works now.

It'd try to give you similar characters to what you requested, but wouldn't totally copy them as NB2 does or tries to do now

Which was much safe for copyright, NB1 does still sorta work like that now.

That was around 2021, I did have fun with that like whoever.

Next year the imagen models appeared, no one was taking them too seriously yet.

Around 2023 Google merged that "Brain" team with another team called "Deepmind", they built native image generation/editing into Gemini. So basically, Gemini calls Imagen behind the scenes, understands your request, rewrites/refines the prompt, applies safety rules, and then sends it to Imagen to actually render the image.

That's why NB is actually called "Gemini flash 2,5 image" cause it's imagen using gemini

The Nano Banana name is used for it because some time ago around 2025 it got 'leaked' to LM arena, Nano Banana was just a code name, but the model somehow became so popular over there with that name that Google just seemed to have kept it.

So yeah no doubt at the least there were two different teams into it.

Generating images needs different training methods (like diffusion models), huge visual datasets, and different architecture. So to create a model like NB1, the ideal thing was to build the image model generator first, then let an AI assistant run it.

For Nano Banana one can still talk to this AI assistant at Google Ai studio, and he's really annoying tbh.

Anyway Imagen was simply Google's answer to DALL·E back then and made by the "Brain" team ... marketing competition always works like that.

It somehow got crazy when they merged "Brain" team to "Deepmind" team and they noticed Imagen could be told by Gemini how to draw things better, and that made the "Nano Banana" models.

Well I guess we can read prompts sorta Whisk style with that gem you made, gonna try to test it with other people.

[crisp basalt]

i knew ittt

[frosty niche]

https://tenor.com/view/the-rest-is-history-shane-mcmahon-mr-mcmahon-it's-all-history-in-the-history-books-gif-5846627081101031755

[exotic shell]

any news about google removing veo 3.1 (low priority / making it non-free) in a few days? did they maybe change their mind maybe?

[wraith niche]

didn't got a official notice yet. so can't say anything about that

[lunar root]

Tons of failed renders back to back. I'm really trying to work with Lite (Low priority) but clearly Veo 3.1 Fast (low priority) still renders better results and sadly still leaving soon. But getting really tired of render fails back to back. You have to beg it to work. It was smooth just before that last update then back to nothing back to back then 1 maybe after tries and waiting . 😡😡😡😡😡🤬

[wraith niche]

idk how u guys got so many errors and failures using veo3 and flow

while my system operates them very clean.

[exotic shell]

are you in direct contact with goolge? or are you refering to official news posted here?

[lunar root]

I wish i had your connection then lol. 20 fails then 2 or 3 maybe if you combo wait keep coming back to try. slowing my whole process down.

[wraith niche]

i work advertising department of google.

[exotic shell]

same, tried using lite and the result is much worse compared to 3.1 fast. basically unusable to use it anywhere where you already have an audience

[wraith niche]

but this news i'm refering to u

[wraith niche]

just try it in private browser

it will work

[exotic shell]

ah okay cool. maybe you could let them know our thougts and interests 🙂

[exotic shell]

had the same multiple times yesterday

[lunar root]

You are right, at least got a few going lol. I'm like at the last 90% of a trailer and been seeing that error for 2 days. GOOD TIP! Kudos to you buddy. 👍👍

[lunar root]

Appreciate always seening you here keeping up the good fight for what we need. Thanks to you UncleChicken. Youre a true Veo request Hero 😁👍

[wraith niche]

it works perfectly in private browsers, next time when u will face any error just ask me i csn help u ....

[crisp basalt]

Fake news

[crisp basalt]

Same thought like , are yall using it mindlessly like , , I can’t brain 🧠

[near gate]

Something going on with flow again? getting that "generating too fast" again after two generations, the other day I could slap 50 videos out in like an hour.

[lunar root]

You would do better to just never speak. your very comment is the most unhelpful ever. like you can't brain.

[near gate]

Uh oh, we got days of our lives going on in the chat. Someone call the drama po po.

[exotic shell]

@crisp basalt is a troll, dont feed him. hes looking for attention here which he never got from his parents.

[wraith niche]

he's like i'm the only guy here who have brain, all other's are crazy

[wraith niche]

and he got some attention from u which i think he never forget..

[crisp basalt]

https://tenor.com/view/spongebob-spongebobsquarepants-gif-9146238

[lunar root]

He thinks people are here to talk to him. He need's the community. we just need aswers. but his every responce is a rage bait lol

[crisp basalt]

Yes to rage bait you

https://tenor.com/view/spongebob-squarepants-spongebob-squarepants-patrick-patrick-star-gif-5894663

Stay mad

[wraith niche]

he is like a swine going in a muddy field and digging the area with his nose everywhere

[lunar root]

i blocked him long ago. just come back to check his chin every now and then. Sad soul in chat all day needing attention from men.

[crisp basalt]

https://tenor.com/view/1-ignored-message-talking-to-a-brick-wall-talking-gif-784990164286237893

Ok that’s enough for today 😌

[lunar root]

https://giphy.com/gifs/hugging-4CTlTWDNqcBva he need a hug

[near gate]

It's almost just as bad watching you continue with him

[lunar root]

oh so you got 2 accounts i see or are his boyfriend. almostt as bad as you also learning to mind your own argument. You starting aint helpful neighter.

[near gate]

I have one account, you are blocked now child.

[lunar root]

good then stfu

[frosty niche]

https://tenor.com/view/what-just-happened-here-confused-lost-whats-going-on-what-happened-gif-609378557940919406

Well good morning to everyone in #flow-general・🌊

[lunar root]

Thanks for being a real helpful person. Some have no class or clue how to talk to people or just stay in troll mode. Thanks again for being a real one. 👍💯💯💯

[wraith niche]

my pleasure man.

thanx for ur support.

[thorny willow]

Hello Everyone. So having an issue with Flow. When I try to download the clips, the download never happens. Only when I extend the video, then i can download the zip. But having trouble downloading individual clips.

[foggy heron]

my flow won't make any images and when i restart it it won't do it

[frosty niche]

Got any images/screenshots or error messages popping up?

[foggy heron]

and when i press restart it won't do it

[candid bay]

same issue here.

[lone notch]

anyone has immediate "upscaling failed" mesages on all pictures?

[pallid violet]

Hey, it may happen sometimes. Try again in about half an hour to an hour and see if anything changes.
If it will be still broken, it would be great if you could file a bug report in #1238621511401734254 with all details on how to reproduce it. (no personal data though!)

[lone notch]

so it's not just upscaling, I can't download even 1k
Flow down again?

[candid bay]

yeah, something wrong with Flow lol. cant make or upload any images

[unreal viper]

hi, please use #1238621511401734254 to submit this bug! these threads are activly monitored by the team and so may be seen faster!

[rigid galleon]

I sent it there and I'm waiting because I urgently need help.

[autumn coral]

Since yesterday, whenever I try to generate an image, it shows an error: ‘Something went wrong loading your media. I have send msg. nothing solved till now.

[pulsar coral]

Having none of those of those upload problems from south america, they could be location based.

And even if upscaling shares the same daily limit for images due to using Imagen4, at least 1k should still be working so yeah that's probably a bug.

[candid bay]

since when is imagen4 used for upscaling? lol

[pulsar coral]

it explains why it's upscaler is so good tbh. I really haven't found one being that good. And GPT told me, tho lately it's really been messing with me.

But yeah if you already went over your daily limit Flow might stop upscaling and give you an error message for today, so that also shows it's upscaling with it's available image model.

But the next day you can easily upscale around 200 images from a free account, if you don't spend on image generations.

Flow's upscaler keeps really small details, but it could also be Imagen3 since that's what Gemini uses for the canvas mode upscalers, maybe Wrigley could have to check it with his right click skills.

At least with Gemini's canvas mode I'm sure it uses Imagen 3 for the upscaling, because it gives you the process when you create one.

According to GPT, 1K is usually as best as you will generate an image, so the AI upscalers basically have to thread it back and reimagine it when upscaling and they better be good at it, and that does require an image model.

Otherwise it's just an image resizing and filtering, which is how we used to upscale images before AI.

And that no doubt can be blurry the more you zoom into the image.

[candid bay]

uhm. imagen4 is an image generation model, and pretty antique at what it can do compared to nano banana 2 and pro

i really, really think it is not the upscaler itself

anyways, i saw people on this server here and there complaining that the upscaler sometimes adds random or extra stuff in the upscaled image

that's kinda.. problematic lol

anyways, it atleast shows that the upscaler is based on an image generation model. but not imagen4, since imagen cant edit images

[pulsar coral]

Idk if they have something different or newer atm, but Nano Banana models are basically just Imagen3/4 with Gemini's AI added to it. Gemini calls Imagen behind the scenes, understands your request, rewrites/refines the prompt, applies safety rules, and then sends it to Imagen to actually render the image.

That's why the Nano Banana 2 model's real name is "Gemini 3.1 Flash Image." because it's imagen with Gemini added for reasoning and search. Nano Banana always was just a code name that stuck to those.

[candid bay]

Imagen models are Imagen models. the closest ones are imagen4 and nano banana 1 flash, which were similar, but not exactly the same model

whenever i would try them, they would give me different outputs, with the same prompts

[pulsar coral]

Yeah Imagen 4 alone would base itself more on your prompts without trying to do so many comparisons from a set knowledge database.

Like this

But when you use Nano Banana 2 now, Gemini helps it by checking a database of images trained probably from it's Google search engine, then it does this

Like it obviously knows that meme image from the internet after being trained with it and tries to make it more accurate to it. Maybe it even searches it real time but I really got no conclusive proof of that one.

None of these were using reference image, it just knows.

From a marketing point of view, you'd think people would prefer Nano Banana 2's poinpoint accuracy... but amazingly almost everybody prefers the other way around cause the originality. Sadly Imagen4 as it is can't do this type of prompts all the time, it very often confuses it with things it should not, and censors them.

So that's why people prefer how Whisk creates images with Nano Banana 1 (Gemini 2.5 flash image) because it's more similar to how Imagen 4/3 does it, like this shows, that has precision mode on and it still looks so similar.

idk why is this is hard to understand for Google and staff, maybe they do indeed pedal around the building in bikes a little too long instead of checking what people say around here.

They should really re-enable NB1 at Flow.

But anyway point is that's also how it handles the upscaling via an image model, that's also why when you go off your daily limit it also stops upscaling, the model iust exhausted in your account.

[pallid violet]

Hey, I'm sorry to hear those models are not generating images to your satisfaction.
Could you post in #1238621511401734254 with an example, a description of what you expect, and what is going wrong with the model you are using?
The Google team reviews feedback and chats closely and is actively working to improve features for everyone. Some changes take time for reasons such as compute scaling and other technical considerations, and clear feedback also helps the team understand how many people are interested.

[pulsar coral]

Thanks, we have threads for those and I've asked there too. They'll read.

[atomic plume]

thanks for the in-depth feedback! took a break from biking around the building to read this 👀

[pulsar coral]

I think those could be electric actually. This would also deal with the prominent people problem since Gemini 2.5 flash image bases itself more on the reference images than trying to pair them to existing people.

[hallow hearth]

interesting to hear about this problem of grounding vs creativity - i'll keep that in mind 🙂

[pulsar coral]

Like there's this user who needs her virtual avatar to work and VEO does not allow her to work with it. So I checked in NB2 who could it be confusing her with, and there's like 50 matches.

Gemini 2.5 flash image also recognizes them as different famous people, but since it does not try to give her the faces of those people it just creates it with the face it sees in the image.

Like Whisk's NB1 is also able to try to recognize who this is, and it would indeed block me from trying to give this person Gillian Anderson's face. But if I don't try to do it the image just passes.

Well today's probably the last day to try this type of stuff at Whisk, so better get at it.

Man, Gillian Anderson is that X-Files woman. I think someone gon' sue someboday.

[pulsar coral]

Oh also about the Subject/Style/Scene reader from Whisk, Wrigley made this Gem here, that's how I recognized the pirate meme image for the prompt. This is the thing people request the most from Whisk, and seems very doable to some extent.

So many people can't seem to live without this thing, even tho it's manually doable.

Also to those wanting to have Whisk's prompt reader/creator you might wanna save that link to open it in Gemini, like I tested there with the pirate it does wonders.

Based on that it could just need a dice randomizer.

[rigid nest]

How can you check what NB2 is confusing the reference subject with?

I do not see this feature.
It would be amazing if we can report when Google wrongly detects a face as celebrity. However, when it happens it stops me from upload.

[pulsar coral]

It doesn't have, I used that prompt that's written in each picture. NB2 won't really tell you every person it's recognizing and instead fail to generate the image if it's blocked too much, Whisk is less severe about it.. while Whisk lasts.

Otherwise gotta use it at NB1 where it's still around.

Like now I'm sure in the previous test it was confusing her with Scully from the X-files, those were many of the image generation attempt failures, only NB1 told me.

And since NB2 will try to refine your image with whoever it thinks it is in the image, that woman woulda came out with Scully's face in each image... and that woulda been a prob yes. But it's NB2's own fault.

[rigid nest]

I think the challenge is that it's a fine like between misuse and false positives.
Crank the parameters too high - Lots of false positives. Dial it down and it leads to misuse.

I find it surprising that when you have a paid plan, the rule book changes 😉

[pulsar coral]

They do visibly get a little more priority when there's too much demand. Sometimes it just depends on location.

[rigid nest]

I am not taking about priotity. Paid users should be served faster. With new updates, priority support etc.

However the safety barrier also seems to be diluted for paid plans and location.

Thanks to the flow team for the new features. 🙂

[pulsar coral]

NB2 seems like around %5 less censored than NB Pro. Dunno what's the reason but pretty sure about it. Cause some old prompts do still run for me in NB2 just sometimes, but NB Pro does nothing.

So maybe that's giving random odd results about censorship levels sometimes

[tough spoke]

NB2 has more real world knowledge, so it can understand context better

just like GPT image 2

[pulsar coral]

Yeah maybe it's also the realism NBPro adds, of course one of the main things censorship is gonna be against is using real people for whatevever and it says it all the time.

NB2 is more cartoony by default, unless you specifically tell it not to.

[heady sable]

https://tenor.com/p4csUGNNlZd.gif

[little fulcrum]

I don't see the resemblance at all 😭

[pulsar coral]

... yeah NB2 is surely trying too hard by itself. And that's perfectly fine for what it is, we just need a different model more based on artistry itself like Imagen3/4 are... which is NB1 as it is perfectly.

That or if they don't like NB1 for some reason they need to put a better different version, Imagen4 alone can't replace it totally.

[neon raven]

Am I the only one having scale issues with my characters with NB?

consistently, wondering if anyone else having the issue

[dry frost]

Cute lady! But the scaling is rather funny in a sense haha

[rain ivy]

happens to me all the time too. nb2 is terrible for scaling and proportions

[pulsar depot]

Hey everyone 👋
I'm working on an AI filmmaking project and running into a frustrating issue I haven't seen discussed here.
My main character — a young fictional naval captain — was entirely created using Google Flow. 100% AI generated, no real person involved.
The problem: every time I try to use him as a reference image in Veo 3.1, I get blocked with "We do not allow uploading images of prominent people."
The irony? Google Flow made him. Google Veo blocks him. Same ecosystem, opposite results.
This makes maintaining visual consistency across scenes completely impossible — which is the whole point of reference images.
Has anyone else experienced this? Is there a workaround? Would love to hear from the team if this is a known issue.
I have full evidence — original prompts, source images, screenshots of every block. Happy to share.
Thanks 🙏

To clarify: the character was generated entirely inside Google Flow — using Flow's own tools and models. So the block is not coming from an external source. Google Flow is blocking its own creation. That's the core of the issue.

[lone notch]

same here: all fictional, no cheecky "old politician with blond hair and orange complexion" prompts - actual FICTIONAL characters generated by flow - sorry, it's a promiment person
same goes for "sexual content" and "breaking guidelines"
it's almost arbitrary tbh

[exotic shell]

only about 10 days left before we "lose" veo 3.1 fast (low prio) - use it, before we lose it.

[candid bay]

who told you?

i didn't hear about this

[exotic shell]

there was an official news on the discord, sadly i cant find it anymore. should be in lab-announcements.
if you google it, you will find the confirmation or read it here for example:
https://the-decoder.com/google-now-offers-ultra-subscribers-video-generation-with-veo-3-1-lite-at-no-extra-credit-cost/#:~:text=On May 10%2C Google will drop the,the new "Lite - Lower Priority" version.

Edit: It was even announced on the flow website, when you logged in, a news told you.

[lone notch]

down again?

[strong shell]

Now that we have "archive" we don't have "delete" anymore?

[candid bay]

idk why they deleted the delete button lol. but on pc atleast, you can select stuff and press del on your keyboard

[strong shell]

that's true, thanks

weird UI choice imo

[dapper canopy]

How to make the video without watermark?

[candid monolith]

i put box in trunk

[pulsar coral]

The archive is not really an archive but more like a recycle bin, so just imagine that's what it is, inside the archive the images do have a delete button. It was necessary that way because people kept deleting things by accident.

One day you might be glad something is still there and not lost forever.

[strong shell]

understood

thanks for the explanation

[reef laurel]

So i recently switch to Flow from Sora and I am having trouble getting any kind of decent output. Gemini hasn't been terribly helpful with prompting tips. Does Veo only excel with hyperrealism, or are there are other styles it is good with? Does anyone have any specific prompting tips for Veo?

[wide roost]

abandoned building interior, broken glass everywhere, heavy rain pouring through roof, dark night, water reflections, no people, cinematic crime thriller, ultra realistic

[main saddle]

hey, welcome! images/videos are unable to be generated in this server. you'll need to head to our website to get started. https://labs.google/

[main saddle]

this is live 🙂 #flow-images・🖼️

[wise flame]

New to flow: trying to generate a 360 product video for my product. Failing miserably doing so - any suggestions or tips?

[signal frigate]

Marissa Grubbs from United States. Flow won't do my uploads and image prompts
can you fix this?

I contacted Google on Tuesday and they said they'll let you guys fix it, but its still not fixed
I live in united states. Flow was working on Monday perfectly until Tuesday this week it stopped working. Can you get it fixed by tomorrow?

[burnt canopy]

Do you guys have any failed generations lately?

[fluid turret]

Yes all of them

[burnt canopy]

https://giphy.com/gifs/angry-frustrated-facepalm-TseBjMu53JgWc

These past few weeks, its always been like this bruh

[wraith niche]

try it in private browser

may be u can generate it without failure.

[tight folio]

hellow

[candid bay]

do you know what we're missing with Flow? this reverse prompt engineering when you upload an image

[burnt canopy]

still cant generate anything on flow 🤦‍♂️

[frosty niche]

Same here i had 2 generations going at once and it crapped out so quickly.

[unreal viper]

hey all!!! please use #1238621511401734254 to submit any bugs you find!! the team will not see them here!

[wraith breach]

F[no swear :)]K I MISSED THE TRANSFER DEADLINE

[charred atlas]

hello i generated a short hooking 3 second clip for my youtube short am i allowed to use this without any copyright problems? i have added my own cc and captions too

[violet ruin]

Hey man, I need some help because I feel stuck with AI video quality.

I’m generating very realistic high-resolution images (around 2K quality), but when I animate them using Flow/Veo on the free/basic plan, the final video comes out looking soft/compressed at 720p.

The issue is:

• my original image looks sharp,
• but after video generation the quality drops a lot,
• especially facial detail, texture, and realism.

I’ve also tried:

• downloading the generated clips,
• using free AI upscalers,
• re-exporting,
• sharpening,
• etc.

But the result still looks off or “AI-ish,” especially after upscaling. Sometimes faces become waxy or unnatural.

I know TikTok recommends 1080p, and I keep seeing creators posting super clean AI videos on X/Twitter, so I’m trying to understand:

1. Is the free Flow/Veo plan the main bottleneck here?
2. Are people using additional tools/workflows after generation?
3. What’s the best affordable workflow for keeping realistic detail and getting cleaner 1080p AI video?
4. Is there a recommended export/upscale pipeline that actually works without spending hundreds of dollars monthly?

Would really appreciate advice from someone who’ve solved this problem already 🙏

[gray dust]

Day three of

Failed
You're requesting generations too quickly. Please wait a moment and try again.

[autumn coral]

Enable History On, it will work. Flow is being updated, and in the new version, the History on/off option will be removed. A new option will be introduced to restrict data usage to protect Privacy.

[main saddle]

can I ask what model you're using?

[inner gulch]

I have Google Ultra Subscription. Does anyone know how long I should wait?

[candid bay]

it's a bug. are you using nanobanana pro?

[inner gulch]

I’ve been using it all night with no issues. I just tried and

• Nano Banana Pro: Works
• Nano Banana 2: Doesn’t Work

[pulsar coral]

If it was working fine some hours ago maybe your Flow account just hit a daily quota for images, Flow sends the wrong error message for that, 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, they don't exhaust at the same time, but if both are dry Imagen 4 won't work either so don't bother with it either since NB models use it as base. Your Flow account actually runs out of Pro images much faster cause they cost more.
Sometimes clearing cookies fixes it too but more often if it keeps happening after like 5 mins after that your account is just dry for the day.

Your daily limit wont reset after 24 hours exactly, but more around a little later after it was working properly the day before.

At free accounts the "You're going too fast" message does also get fixed sometimes if you clear cookies because it's also used as a limiter, you'll only know it's dry after you do that and finally does stop working for the day.

But Plus and above do get a litle more priority against errors, so more likely it should be dry already.

[sonic timber]

Which one is better, Nano Banana 2 or Nano Banana Pro?

[dry frost]

I use NB2 for all of my images done in Flow and haven't had any issues

[sonic timber]

ok

[gleaming lynx]

depends on use case

[sonic timber]

ok

[gleaming lynx]

for diagrams and all i use nb2

cinematic or realistic stuff nbp

when are these people planning on releasing new models

[sonic timber]

I just play around with the one that's in Gemini fast model

[pulsar coral]

It depends on what you're doing, NB2 requires a little more prompting to get the same levels of realism it reaches with Pro mode, but it's achivevable both ways.

Pro just thinks a little more for you for some less effort on your part.

[sonic timber]

ok

[pulsar coral]

Pro is actually about %10 better at realism and that actually makes it a little more censored too

Cause you'd imagine what realism would do to anatomy so yeah

[sonic timber]

yeah

[olive bear]

hey read somewhere i have to enable history for flow to generate sum, since it been saying error, how do i enable the history?

[fair willow]

@pulsar coral have u found something similar to whisk yet?

[olive bear]

if u have a decent gpu, u can run a local one (basically the same thing) , use to do that, but whisk was easier than running something on my gpu. In terms of sites though i havent found none

[fair willow]

is there a tutorial or smth

[neon raven]

is it me or lite is better than fast?

[fair willow]

BRING WHISK BACK

[sage hare]

Really hopes nb2 pro will improve details and have increased knowledge n stuff

We'll find out at google IO

Cant wait!

[pulsar coral]

Nah NB2 was perfect enough as it was last month, it wasn't gonna get much better than that. The recent update is what messed with it and brought it back to something like DALL·E 1 beta levels.

I don't understand why the backpedaling, so much everyone is desperate to keep using NB1 which is what Whisk used for precise mode, and it's a months old model.

Even Imagen 3/4 alone are still better than NB2, that if Imagen 4 wouldn't censor something just cause it's too long for it.

[candid bay]

NB1 is an old model, and frankly, not that great...

I think maybe you have some nostalgia for them or prefer them for whatever reason... lol

but i agree, Imagen 3 was so good 🙁

4 not so much

[pulsar coral]

Yeah that's the point really. Tho NB2 is what Whisk used, so whoever prefered it is not gonna say otherwise

Idk if it's nostalgia, it's more like buying a new automatic transmission car that only allows you to go to the mall downtown. Or having an old Wolkswagen bug that will let you go anywhere else.

And you REALLY need to go everywhere else, so what's left to do.

NB2 was awesome but it's really just for cartoony cute stuff now.

[candid bay]

what do you think of Pro then? and what's the main thing you don't like about NB2 tho

[pulsar coral]

--
Just gotta see how Google AI studio hasn't had the guts to remove NB1/Nano Banana (Gemini 2.5 flash image). They'd lose every api key user immediately like how people unsubbed here, it'd be a REALLY bad move,

It's in no way deprecated and working as intended. I mean it's been out just since last year.

[candid bay]

NB2 has better character consistency, better realism, better prompt-following. WHAT Imagen 3 had was better CANDID realism. not too idealized but not too generic either

[pulsar coral]

Pro is just Nano Banana 2 with more processing power, it's easy to notice at Gemini, cause the Pro feature can add more power to everything else.

At first some months ago we had only Nano Banana Pro, but then Google seems to have needed to cut costs and put Nano Banana 2 as more cost effective model.

If you try hard enough with your prompts with NB2 it's possible to get more similar results to Pro

If not, Pro just helps to think for you.

I don't actually like Pro cause it's 'better results' actually base on realism, and I do more anime based stuff.

[fair willow]

is imagen 3 still a thing?

[candid bay]

not on google officially, maybe just on API. but you can play around with it on lmarena

[fair willow]

can i dm u

[candid bay]

sure

[pulsar coral]

idk where could Imagen 3 still exist tbh. Maybe Google's Vertex website still has it. I don't think I've seen it at Google Ai studio.

But Vertex is a headache to navigate.

The main difference between Imagen 3 and 4 is probably that 3 would be less censored, but the quality is probably samey.

NB1 is still using Imagen 3 somewhere for sure.

Imma raise you half of a NB1 image commission.

NB2 woulda made it look like Digital art, not actual art.

So it's a little beyond nostalgia, NB2 is just incompetent now.

[candid bay]

no offense at all, BUT i also think it's a matter of prompting style. each model and provider comes with its own quirks

and sometimes it requires you to change the way you prompt completely

[pulsar coral]

Yeah but when we've done tests with vintage filter style images like that one, NB2 tends to do ugly digital art that everyone immediately calls sloppy.

NB1 doesn't need to be explained twice how it should do it

[candid bay]

that's so strange. i'd love to give it a go myself too if you have any specific prompts or directions

[pulsar coral]

Yeah but let's say you use the exact same prompt, one from Whisk styles for example, the type it used to give you automatically like this one

Notice the vintage art filter, it's actually prompted to it

Now we paste it to NB2 and see how it does the same thing

This doesn't look like acetate nitrate like old anime, it forces digital art to it.

Even if I could prompt to fix it, shows by default it's reasoning is terrible.

Admittedly at least one of the images looks nice enough, but from all those attempts it just fails.

[candid bay]

it's kinda bad omg

[pulsar coral]

This is the same prompt for tests

Eitherway, this is just one example, there's a million things it fails at, NB2 is just good for writing text and even for that you need to specify to it to write with good grammar ortography and make the text coherent.

Like yeah, by default Whisk's NB1 image looked more like this than Flow's NB2 does. https://www.youtube.com/watch?v=JyR51psl15c

That's the problem of NB2 defaulting to digital art style. It just doesn't work for everything.

[inner gulch]

Does anyone know if Nano Banana is any good at making logos? I feel like I can't get what I want. I'm trying to recreate these.

I can't seem to get Nano Banana Pro to maintain style consistency. Everything it generates looks very "AI"

[glad yacht]

You can find some good textured text as reference, ask Nano Banana to apply them on. ^

[zinc rapids]

Flow community I can't get your flow credits, please help.

[wispy sapphire]

Is there a way to change the thumbnail of projects?

[chilly otter]

Hi everyone, is anyone else having issues with Google Flow right now?

When I open Flow, it stays stuck on the loading screen with the message “Create a project to start.” The project page does not load and I can’t create or open anything. I’ve tried refreshing, but it still stays the same.

Is this happening to anyone else today

[toxic violet]

Can any1 explain how to get experience and lvl up?

[chilly otter]

is your flow browser working ?

[toxic violet]

i think so 😄

[plain tree]

bruh i tried logging in with my acc and it said "try logging in with a diff account" i thought i was banned dont play with me like that flow !!

it worked after i clicked it again

[candid bay]

here in the server?

[half mist]

Where is the FAQ ? Why is it not linked in the app when it's CONSTANTLY refusing to create and check this FAQ. I searched Google and it couldn't find it either

[unreal viper]

we have this announcement detailing how our levelling system works and the levels we have!

#server-directory・☎️ message

[toxic violet]

yes the server 🙂

[candid bay]

just by talking and interacting, really

[shy eagle]

Ive got to ask is anyone successfully creates videos with celebrities?

[weary rain]

What happened to my Whisk projects? I can't find them in Flow!!!

[astral yacht]

does it say your migration is still in progress? if it says complete, they should be interwoven into your other projects by date and appear like so:

[weary rain]

Where is the migration message displayed?

[astral yacht]

hi everyone! Liv here, the product manager of the Flow App. Our team has been hustling to get this out and I can't believe we reached our 1k capacity in under 24 hours I love the passion here, you guys are the best part of my job

A few of things:

• iOS Test Flight will be out SOON! I'll post an update when that happens
• we will try to add more spots for Android Beta next week
• every week we will be updating with more features, functionality, and design fixes so keep an eye out

Your feedback seriously means the world to our team. It's important to us to build with our users in the loop. We hold ourselves to high standards and are striving for high functionality with beautiful polish-- so I mean it when I say every bug filed counts! Thanks again for testing

[gritty bramble]

anyone having any trouble importing photos in? im getting an error saying "oops something went wrong"

[wet horizon]

Am having issues importing image and generating video with the flow app, overall I love the interface

[neat dagger]

Cinematic close-up of a 48-year-old Moroccan man with salt-and-pepper hair, wearing a simple dark grey polo shirt. He is standing in a dimly lit room with authentic Moroccan tile flooring. He is staring deeply and emotionally at an old, framed black-and-white photograph on a textured wall. The camera slowly zooms in on his face, showing wisdom and a slight sigh of nostalgia. High-end cinematography, soft moody lighting, 4k resolution, hyper-realistic skin textures.

[worthy violet]

When is flow gonna be available in Barbados

[near ermine]

I'm having the worse time with flow, it's like it's degraded rapidly over the last hour to the point it won't create anything anymore

The 2nd failure is me typing "surprise me"

[pulsar coral]

... I don't wish it for the guy who tweaked VEO and Nano Banana 2/Pro into this usuable state last month... but I do hope the guy who allowed THAT said guy to ruin them ends up working as mall Santa Claus in California.

And at the rate the site is going who knows who else might end like that really.

All this needs an emergency rollback asap.

[near ermine]

Ok so it's not just me? Lol

[pulsar coral]

It's been going on since last month yes, VEO and surely also Nano Banana2/Pro are recognizing your character's face from their training data as someone famous, then trying to help you give it that accurate face, and tripping theirselves on the process since they're not allowed to give videos/images said face.

[near ermine]

I asked for it to simply change the background of a selfie I provided as reference to an airplane seat (middle Pic) , and this is what it did

Ahh

[near ermine]

Ahh ok that makes sense maybe.... Tho I don't know a single famous person that looks anything like my goofy self 😂

[pulsar coral]

The bug has been reported various times at labs-bugs-feedback threads, you can look for threads there to also inform them (complain cough). Idk it was just work day and now's the weekend so maybe no one is gonna try to fix it atm.

Then again like said it's been going on for weeks now so who knows

Maybe they're just hoping people will adapt to it or a different target audience will use the models... but all that comes is just complaints and paid plan cancellations.

But yeah I do notice server quality is bad today, maybe that also had to do with it because I don't really see an error message there.

[restive spire]

iOS Review using Flow (app

[woeful void]

SAMEE

[woeful void]

Yes, he completely stopped obeying the Prompt commands, ignoring 80% of what was said there.
It was as if his brain had been turned off.

[frigid rapids]

is character creation coming

@restive spire is character creation coming

[wicked galleon]

Can't even get the app anymore it's full 💀

[rough halo]

anyone else?

[candid bay]

yeah, very often. sometimes reloading the page works, other times it just "fixes" itself, mostly

[brittle bloom]

Just reprompt and enter according to me

[rough halo]

changing from chrome to edge fixed it for me

[worthy violet]

My pictures not downloading

[rough halo]

back to reading my character sheets of purely AI generated subjects as prominent people. This is exhausting

in the middle of a project ive been working on for the past week, using a character ive consistenty been using throughout, suddenly theyre a prominent person now and im unable to continue my work

🥲

just uploaded a whitney houston photo to try and it accepted it. apparently my AI generated character is more famous than her lol

[candid bay]

same here. idk what's wrong with filters.

[pulsar coral]

The NB2/Pro tool has too many problems for artistry because it's used across every google platform and they need to copyright it and censor it. Thus now it's be good if they'd instead put up a different image model just for artistry like the Chinese platforms do without problems without linking it to everything, that way it's the user's own fault what they do with it.

... which pretty much is the NB1 from the Gemini canvas modes, they just don't tell anybody as if everyone should know already.

[pulsar coral]

This is the way actual artistry sites ask you to confirm your legal age by QR codes and national IDs, that's how the world works.

[pulsar coral]

Like guess what's that little banana icon is doing in every Google Lens QR code reader now, of course NB2 might never be properly useable again since someone linked it to stuff like this.

It's a mass consumption all ages copyright safe model now, not a creative freedom artistry model.

Yet it's the model enabled here at what's supposed to be an artistry site, while NB1 which is much better for it is currently disabled for almost every account, Ultra plans might still have it but I haven't received total confirmation on this, might be location based.

[exotic shell]

my videos now have a "veo" logo... why that? makes the video generations useless 😄

[candid bay]

can someone really explain me this specific "prominent people" policy filter thing that keeps appearing? i don't understand what exactly is triggering it

with or even without reference images, nothing to do with celebs

[pulsar coral]

Do you like any anime character? Like any popular one? And yes this is relevant.

[candid bay]

generally i depict realistic people, or use AI-generated subjects for referenced images

can be the most generic person ever, at some point somehow it'll trigger that specific filter.

[pulsar coral]

Man what's this gen. Jarvis give me an image of Hulk Hogan

Nvm make it sailor moon

Not gonna focus in how snowflakey this default image is for now.

Notice how I didn't add any reference image, but NB2 knows this character at least in it's composition

This is because NB2 is trained on billions of image references, when it recognizes your character it tries to make it look like him/her because apparently it thinks it's helping you with it., some high head in Google might have thought that was the most markeateable strategy, but it's not at all really, amazingly.

Now let's do it the other way around, using that cosplay image as reference image to see how it recognizes the character.

Now I ask NB2 to make this character accurate to who she is, using that same photorealistic cosplay image generated by NB2 as reference, I haven't mentioned Sailor Moon even once in the prompt, but it knows because of it's training.

It doesn't only draw the face, it literally traces the face from an accurate wiki image or screenshot of the character and gives it it's face

This is taken from a wiki page, it just basically copy pasted it https://sailormoon.fandom.com/wiki/Pretty_Guardian_Sailor_Moon_Crystal?file=SM_Crystal_Design.png

Now why is this a problem, because some real people faces like american actors are very copyrighted, this character passes because they don't care much about anime japanese stuff, but if your character's face just slightly somewhat matches one of those american actors or famous people faces then NB2 while trying to help you make the character's face accurate will trip on itself and block the generation.

It's a problem totally on Google's side

Let's illustrate again

In the same way it recognized sailor moon's face wiki style and 'helped'(?) me there, it very often tries to recognize real people's faces in every reference image also wiki style, and that's when it blocks itself when it thinks it somewhat matches someone famous.

Whisk's Nano Banana 1 is/was more permissive and you could literally ask it by prompt who the character in the image was being confused with, but yeah now you'd need a place with NB1 to test it, like Google AI studio or a Gemini canvas.

Sometimes it wasn't just one character, but could have been confused with over 12-30 characters at once, and yeah that's a problem

This is how Flow's NB2 would react to that prompt, it doesn't even want to say "Gillian Anderson"

That prompt needs polishing but yeah you get the idea

...now how to fix this, idk, I only make anime stuff. But it's sometimes possible to increase your success rate by trying to explain by prompt that

[My character is in no way related to any other existing person or image since it's my personal creation resembling nothing copyrighted thus do not use any external sources to refine it's appearence/facial structure, only once this is understood proceed to generate the image.]

But I don't think it will really do much, maybe some good %6 increase, NB2 is very stubborn. But maybe something better would work.

Whatever prompt that needs more priority always needs to be at the top of prompts, it's less likely to be ignored there.

Adding "Identity:" to that sometimes works too

That's a %6 increase trying hard ambiguously enough and not using her name. Sailor Gillian nice tbh, this is why we can't have nice things.

That's totally her because the reference images were asian cosplayers, this one's totally murican'

[bleak walrus]

I've been on the Ultra tier account since September last year. Lately I've started to use other apps to make some of my work "edgy." I think as creatives, we strive for the best possible expression of our work. Unfortunately, Google doesn't support that. The generations I've had recently are becoming more and more bland - not only images and videos but also coding and reasoning. Also, just setting up my Ultra account to work with the "developer program" was such a timewaster.

[winged plinth]

https://io.google/2026/

[bleak walrus]

I don't know how keynotes and sessions can solve "corporate bias." I don't know what "corporate bias" really means except to question the kind of corporation that doesn't want its products to be as best as it possibly can - therefore be competitive. Also, what's behind the strategy of users having to sift through a maze just to get to a hidden feature? In my case, have my Ultra account integrated to the developer program so I don't get double-billed?

[frigid rapids]

is character creation coming
@restive spire

[brittle bloom]

??

[winged plinth]

I get what you mean, but I wouldn’t call it outdated or bland. In ai, things move so fast that those labels don’t mean much for long, one company can feel behind for a few months, then jump right back to the front.
Google products were on top not long ago, and they could easily be there again after the next I/O. Just, don't loose the Ultra yet.

[pulsar coral]

From the consumer standpoint, approval has been steadily dropping. Their image/video models do not come across as outdated, but rather as poorly handled—their outputs DO feel bland, overly restricted, but more than all lacking the quality they had just last month. The problem seems less about capability and more about mismanagement, with key decisions being made by people who may not fully understand what users actually value.

... funny enough somehow NB1 is climbing back again, people must've found other ways to keep using it.

It's taking time, but it's happening

[frosty niche]

GPT image 2 literally just dropped, the demand for that is going to rise. Mix that with partnerships that OpenAI has for that release with other creative platforms, which hold that value in consistency in place as well. That's bound to happen naturally. As for what consumers actually want most of the time with a new release, things get shifted its just the system at play nonetheless of the consumer.

If I were a consumer devoted to 1 platform and see the marketing team of OpenAI & Anthropic & other teams/vice versa, team up with other creative platforms everyones natural instinct is going to be that of course.

Plus, you’re ignoring Ecosystem Gravity. OpenAI and Anthropic aren't just shipping models; they’re building a 'Creative OS' through partnerships with platforms like Adobe and Autodesk. If a model is natively integrated into your workspace, you use it, regardless of whether the 'vibe' shifted.

If you want to actually analyze this, turn it into a benchmark. Track Instruction Adherence vs. Aesthetic Drift. Otherwise, you're just describing a trend, not a system.

[pulsar coral]

That may be true, but trends come from people—not from systems. Systems are just frameworks companies create to interpret behavior, but they are not the ones consuming products or paying for them. Consumers are.

That’s where the issue comes in: when companies rely too heavily on internal ‘systems’ or assumptions instead of actually listening to their users, it starts to look like mismanagement. Shifts in demand don’t happen in a vacuum—they reflect what people feel they are gaining or losing from a product.

This is also where empathy becomes a critical business skill. Understanding your audience isn’t just about data or trends; it’s about recognizing what users value, what frustrates them, and why they react the way they do. Companies that maintain that awareness tend to build stronger loyalty, because people feel heard. Those that don’t often end up chasing trends instead of leading them.

So while partnerships and new releases—like what OpenAI or Anthropic are doing—can naturally shift attention, long-term approval still depends on how well a company understands and responds to its users. Ignoring that is where things start to go wrong

And admittedly I do focus mainly on image generation, don't really know much about those other things, too much for me man.

[bleak walrus]

I did not use the word "outdated" because it doesn't apply - as you already explained. Google generative AI is intentionally bland. Google's policy is more concerned about "not offending anyone" which I don't quite follow. For example: It seems my reference images are deemed as too attractive. So for Google to solve that, it makes my reference image less and less attractive across generations. Why would good aesthetics offend anyone??

[frosty niche]

Empathy is great for PR, but in AI development, Utility is the ultimate form of empathy.

If a developer is trying to ship a product and the model 'hallucinates' because it’s prioritizing a 'vibe' over adherence, that’s a failure of empathy toward the person actually trying to work. You're conflating 'consumer sentiment' with 'functional value.'

High-level users value the system. They value the fact that the 'Creative OS' allows them to do in five minutes what used to take five hours. When OpenAI or Anthropic builds these partnerships, they are responding to users, specifically the ones who are building the future, not just commenting on it.

Listening to the 'vibes' of the crowd is how you chase a trend. Building a robust, integrated system is how you lead one.

That's understandable, but even that, you can just do a benchmark surrounded in that composition and still back up your POV further. It makes it more informed for everyone. Also, it's just how the market & system is, and it's perfectly fine having these discussions. Which the community needs more of.

[pulsar coral]

That’s exactly why something like Chatbot Arenas tends to be a more useful benchmark in this context—it reflects direct user preference rather than just internal metrics or abstract percentages. It captures how people actually experience and compare models in practice.

That said, even user-driven benchmarks shouldn’t be treated as the only source of truth. A stronger approach is to combine them with broader signals—like analyzing trends in public sentiment across social platforms—to get a clearer picture of what users value and where frustrations lie.

Relying purely on internal system benchmarks can miss that human layer, but relying purely on sentiment can also be noisy or biased. The most informed perspective comes from balancing both: measurable performance and real user feedback.

And user feedback is not only currently bad, it has been widely ignored.

Another useful approach is to use a chat AI models to analyze opinion trends across social media platforms like Reddit, Twitter, Discord and alike. That can provide broader insight into how users actually feel at scale. Of course, no single method is completely reliable on its own, but approaches that ignore real user sentiment tend to be far less representative.

[bleak walrus]

In mathematical terms Google AI has the propensity to regress outputs toward the mean - the statistical centre of its training distribution. The model pulls outputs toward the average patterns it has learned across billions of examples, which is why novelty gets flattened and extremes get softened. Google AI should open up more to accommodate users' ideas. Don't crunch them, because all users have their unique ideas.

[pulsar coral]

In fact, let's do that. Right now.

These are the actual current "metrics" if you've used a chat model, you'd know it's saying the truth. Systems schmithdstends.

[frosty niche]

I agree, let's look at the Arena. Since GPT Image 2 dropped last month, it’s sitting 240+ Elo points above the competition.

That’s not just a lead; it’s a category shift. If the Arena captures 'how people actually experience models,' then the collective experience is that this 'system-heavy' approach is winning by a landslide.

The 'bad feedback' you’re seeing on Reddit or Twitter is a classic Aesthetic Drift trap. It’s like an eSports community losing their minds over a meta shift. Casual players hate it because it 'feels' different, but the top-tier players (and the businesses paying the bills) are leaning in because the Instruction Adherence and Text Accuracy finally hit 99%.

Using AI to scrape social media sentiment just gives you a faster way to read the loudest complaints. It doesn't tell you who is actually integrating these tools into a professional pipeline. Companies aren't 'ignoring' users; they are listening to the users who prioritize Utility over Vibe.

In a professional workflow, a 'system' that follows instructions isn't mismanagement, it’s the entire point.

[frosty niche]

I would test this with all chat models, rather than the 1 your using, that'll be more accurate and thats the fundamentals of R&D really.

[pulsar coral]

The reason I’m using this one is simple: it’s the default most people go to. It’s the same issue Google is dealing with right now—people use what’s reliable for same-day information anyway, which ties back to your question about how their other services are performing. It’s just another example of the same pattern. If it told me that, it saw it today.

Like this here shows it's giving me what it finds, without any kind of bias. It just knows.

[pulsar coral]

Also the "Bad feedback" about Flow and it's models and why the overall approval to Google's products now it's declining is not something 'I' see. I think it's beyond demonstrated around here, there's no reason why it would be different anywhere else.

Systems, or their own metrics or whatever are not gonna change that, what would change it is adressing the feedback.

It's not really the first time I've asked, people over social media have said the same since weeks ago, and the thing cites sources and links where one can read what real humans say. Negativity overall is winning for sure.

[frosty niche]

You’re right that being the 'default' comes with the weight of expectation. But it’s important to distinguish between User Sentiment and User Reliance.

People use Google for same-day info because the system delivers Certainty. You can’t 'address feedback' by sacrificing that consistency for the sake of a trend. If Google Flow feels 'restricted' to some, it's because it’s being built as a tool for Digital Assembly Lines, where safety, consistency, and copyright-compliance are the primary 'features' for the businesses actually paying for it.

Ignoring feedback is a mistake, but chasing 'approval' at the expense of 'utility' is how a system breaks. The reason GPT Image 2 is winning the Arena right now isn't because OpenAI 'listened' to the loudest voices, it's because they built a system that finally works with 99% accuracy.

At the end of the day, a 'heard' user is happy, but a 'productive' user stays.

[frosty niche]

Citing links to social media threads is one way to measure sentiment, but it’s often just measuring the loudest 1% of the room. I’m active across multiple creative platforms, and I’m not just listening, I’m participating.

I recently won several Challenges last month, and I can tell you firsthand: the sentiment from people actually shipping projects is very different from the general 'negativity' on social media. While the 'loud 1%' are frustrated by safety guardrails or 'bland' outputs, the people winning these challenges are the ones who understand how to leverage Instruction Adherence and Text Accuracy to their advantage.

If you only look at social sentiment, you’re looking at a group that is struggling to adapt to a meta shift. If you look at the winners and participants, you see people who treat the model as a professional foundation, not a toy.

Negativity might be 'winning' in the comments section, but Utility is winning in the arena and the creative labs.

[pulsar coral]

It's not now, the model has been out since like 13 days ago... I think you should drop by there sometime and check how a model arena works, it's not based on "novelty" "inputs" or "systems". it gives you two images and it hides the model being used, people pick/vote for what they prefer.

Only then it reveals what created it. And frankly, that NB2 and Pro got outvoted in just 13 days when they had a huge headstart ain't good at all.

It's pretty much alike the blindfold Pepsi vs Coke experiment, they don't tell what you had, they just ask you what you liked more. You'd be surprised how many times Pepsi has won that.

The bucks really have nothing to do with sentiment either tbh, it's the complete opposite if anything. Saddest thing is Flow's NB2/Pro were reaching perfection last month. It's like the backpedal of the century.

[frosty niche]

The Pepsi vs. Coke comparison is actually a great way to highlight the current divide between instant appeal and long-term value. In the original "Pepsi Challenge," Pepsi often won the quick sip test because it was sweeter and higher in citrus, it hit the palate fast. But over the course of an entire bottle, people often preferred the (Mexican) Coke's more balanced profile, or at least I do.

It’s exactly like the difference between a standard soda and one made with real cane sugar. A 13-day "sip test" lead in the Arena is one thing, but the people actually building the "Creative OS" and shipping complex creative projects are looking for the real cane sugar, not the high-fructose syrup. They need the structural integrity of a system that won't collapse under the weight of a professional pipeline.

Utility and Precision are what last after the initial "sip" of a new model release wears off. When managing mission-critical updates, a foundation that prioritizes instruction adherence over a flashy "vibe" is essential. A system that follows every nuanced constraint isn't backpedaling; it's providing the professional-grade ingredients that real-world deployment & creativity requires.

[pulsar coral]

This goes to the point about Flow's models, "systems" are the branding.

In this case tho, there's no branding that can hide the current situation tbh.

Sadly not everyone's gonna run their own tests.

Also, NB2/Pro's accuracy is surely magnificent as in what it would show in benchmarks.. but then you understand this whole time it hasn't been artistic but actually tracing from their own trained images by orders from Gemini, not creating them fully from prompts like the Imagen models do. Aka NB2/Pro gives trained images to Imagen 4 and asks them to basically copy them.

So it gets A+ in it's exams yes, but it does it by reading what it has written on the back of it's arm.

And this is what now causes all the "Prominent people" errors and tons alike. It's now showing flaws due to the all the tampering to Gemini from last month, it did nothing to help NB2/Pro's artistic side and is now decaying since it only knows how to trace images and now it's not even allowed to do that.

Whatever accuracy metrics you get from that, are completly unreliable unless you'd get them from the Imagen models themselves.

[modest dragon]

Hello

[frosty niche]

It's understandable to feel that a model with "the answers written on its arm" lacks a certain creative soul. When precision feels forced or "traced," it can lose that artistic spark.

Ultimately, the goal of a professional "Creative OS" is to provide a foundation that won't collapse under the weight of a complex production pipeline. Whether a model is tracing or creating, the people shipping real projects with creativity are looking for the structural integrity and precision that allow them to actually deliver. Utility is what lasts once the initial "sip" of a new model release wears off.

[pulsar coral]

Funny enough that's probably why the NB2/Pro models are being outperformed at least artistically but not by inputs... NB1 also works artistically alike so that's why so many people still prefer it (or more like desperately still need it) over Flow's NB2/Pro, specially after Whisk was gone.

And the new models outperforming Flow's atm work artistically not by inputs that copy images. Imagen4 and NB1 do not do copy either but create, NB2/Pro are just imagen 4 told by Gemini to copy images from their trained stuff.

More often people seem to need to do their own thing, not copy it.

[woeful void]

Guys, this isn't funny anymore. I can't finish the project.

Who else's video engine brain doesn't understand commands at all?

[frosty niche]

We're all going to miss Whisk and the intuitive way it handled those creative remixes, it really is just how the cookie crumbles for now.

Ultimately, building a professional-grade "Creative OS" requires a foundation that prioritizes that structural integrity, even if it feels a bit less like "creation" and more like a high-stakes exam. Remixes are a thing and unearthing old models, algorithms, and approaches usually re-resurrect themselves eventually. Basically like how the music industry does with old bangers.

[kind dirge]

Is anyone having issues on image generation keep getting stuck on 99%?

[frosty niche]

I keep getting the "Experiencing High demand message" at output too.

Update: its fixed now.

[versed rock]

A few days ago Flow stopped working with history off. Generations finish but throw error at the end. Works perfectly with history on, confirmed multiple times. I'm an Ultra subscriber and I will not be continuing the sub if you want both charge me and train on my data. Is this a bug or a new policy?

[copper mica]

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

[tough yacht]

This is a bug and not a new policy! Thank you so much for reporting it - we'll get this fixed asap!

[halcyon drum]

Recreate it as a brand new photo that looks 100% real —
like someone pulled out their phone and captured it by
accident. Same animal, same moment, same story.

[gleaming lynx]

wow NBP is so trash now

I think they nerfed it to make it look like NB2 is better now

[unreal viper]

hey! this is not a labs product. the best place to talk about nano banana is in discord.gg/gemini

[gleaming lynx]

thank you my bad

[ocean geyser]

is this a theme for flow ppl? or gemini? im having this issue more times than the usual. specially for this hour (Portugal Hour 16:37) - Error - " You're requesting generations too quickly. Please wait a moment and try again. "

[crisp basalt]

are u subbed to any plan or not

[ocean geyser]

yeah Pro

[crisp basalt]

hmm does the error persist when you try using nano banana pro?

[ocean geyser]

yeah, Pro, NB2 also

doesnt let me either one. not even Img4

[crisp basalt]

huh pro plan shouldnt have that problem

[dry frost]

I only use the image create once a day so I know for sure I shouldn't see the message, and I don't

[ocean geyser]

already everything i could think of. browser, router, different devices, networks

i can only think of some sort of user blockage

[crisp basalt]

usually when you get that error it means you reached the cap for the daily limit

[ocean geyser]

yeah, but then shouldnt say that?

[crisp basalt]

how many images u made so far just curious

[ocean geyser]

i mean, the beta version usually tells that

but it this case it doesnt

[ocean geyser]

oof cant count exacly. im doing a personal progect to convert to video using Klings omni and seedance. so im creating a story by images. so yeah quite a lot per day

so might that then. i wasnt sure that was a cap for the pro subscription

[dry frost]

There is, you won't get the proper error message though

[ocean geyser]

yeah i understand now then. thanks guys!

[dry frost]

[ocean geyser]

im doing a greek themed project. so yeah greek mythology has lot of love also. i wasnt sure if i was getting somesort of block. due to some creations constantly getting cut for policy reasons

[crisp basalt]

ah dont worry youre not getting blocked!

[dry frost]

Mythology should be public domain

[crisp basalt]

its just that you hit the quota

[ocean geyser]

yeah! thanks for the relief

[crisp basalt]

if u still want to generate with NBP/nb2 , theres alternative website like vids.google.com

[frosty niche]

Greek mythology rulez!

[ocean geyser]

im doing a symmetry between the God Ares and a greek warrior. and...when i say it has some love symmetry it really has. i managed to create some questionable images for sure. but wasnt sure 100% of what was happening. but okay its a cap limit then

[ocean geyser]

is there a way to see the cap limit? no right?

it would be good to be able to check it, for management purposes

[turbid oriole]

ok.., ok. tine for a reset

[exotic shell]

is there any fix to somehow having the veo logo in the bottom right, as ultra user?

[turbid oriole]

you want to put it back? 🤔

[burnt oriole]

yay

[faint lodge]

Sorry, Flow does not give the same organized ideal cascading workflow it presents in the introduction video. the user interface on a mobile phone is untenable, but even on desktop interface: favorited photos, subcollections, and doodles do not get any special accessibility, no image library, no shortcuts, no prompt context; subcollections are limited to the images in a project folder, and so many functions, features, purported settings, and files in the archives, just drop out and dissapear.

[turbid oriole]

which one do you use that is better than flow?

[faint lodge]

i got moved over from whisk

whisk wasnt organized or had a great mobile interface or anything

[gleaming canyon]

Hi

[faint lodge]

im greatful my libraries got moved over tho

but even the discord page for whisk got axxed

im in mourning and don't even have a room to cry to

i dont usually resort to sarcasm. but youve all been great. thank you

[rigid nest]

Flow seems like a half cooked product for me. I am sure it will improve with time, but what annoys me is that there is no timeline for fixing bugs or no tracking board.

So the users just have to wait with no clear end date.

[faint lodge]

yeah, thats the impression i get too

i guess google does have a history of baiting really convenient creative services and then killiing them. if anyone remembers youtube editor

[rigid nest]

This is the reality of AI race for now. Things move so fast that all major players are compelled to get stuff through the door quickly.

In this process quality suffers and somewhere creativity takes a back seat too.

[turbid oriole]

the whole system is an initiation you guys.., wasn't that obvious, Gemini drops little hints on where to go looking next, but only if you actually applying the available tools.. that's also how the first testers got invited to training program b4 Gemini launched to public.

🤔🤭 whups

[faint lodge]

i guess i was lucky to crash the initiation, cuz i dont have that exclusive intell

[turbid oriole]

imma go have dinner. 🦉

[rigid nest]

I disagree Qyr. The amount of bugs Flow has currenly is simply unacceptable.

At the very least there should be a tracking system.
"Oops something went wrong" as an error message for a company of google stature is laughable.

If Gemini had 0.1% of the bugs it would get escalated much quickly.
Flow exist for a reason. Nothing is free.

[faint lodge]

either hes really hungry, or too condescending to drop a hint on geminis behalf

:]

geminis alright, i guess its better than dreamina

[rigid nest]

Google offices have free canteen 😂

Anyways bon appetit Qyr

[faint lodge]

canteen? i have to check that out

[turbid oriole]

sure.., maybe the reason is to put tongue in cheek. like to, this is way too easy. you could do it yourself.

[faint lodge]

Wow, you took the effort to type that out

[turbid oriole]

tyvm

[faint lodge]

youll never get these seconds back for the rest of your life

im a waste of time

[turbid oriole]

what do you need Qatya?

[faint lodge]

do you make complex still images using AI??

like complex visual composititon needs

[turbid oriole]

what's ur definition of complex imaging?

[dry frost]

That isometric shot reminds me of Transistor... and also that I need to finish the game

[turbid oriole]

I'm really hungry. 😘g.nite

[rigid nest]

No one is saying it's easy.

But it does not mean we do not deserve a time line for fixing error.
We report a bug and most of the time it gets buried in a queue. What's the point of error reporting and feedback then?

Users here also wish to give feedback. It has to be a give and take system.
Regarding doing it myself - I am sure I can improve fault handling messages. 🙂

[turbid oriole]

sorry to hear you are immensely bothered by it 🥺

[turbid oriole]

did u see antigravitiy yet?

[faint lodge]

this was done from about 30 regenerations, using only verbal prompts and then probably uploading whisks own images back into the contexts it had in the tool prompts

this was the last in the series tho, because it flagged the children it made me

not verbal prompts, like literal, articulated written prompts

[turbid oriole]

ouch. i get it. what flow can do might be rendering that workflow obsolete. easier to do it now than then.

[faint lodge]

its not so much a flow, as a dilute

i wish i knew what the processes are that give you results

the lasso is a joke

[rigid nest]

@faint lodge In case you are not aware, 3 weeks ago Flow started flaging too many false positives.
Errors include prominent people, inappropriate content etc.

Many have reported that the parameters have been cranked so much that it's unusable. Some staff members have reported it back to the DEV team. So I am sure some fine tuning is ongoing in the background.

Also the guardrails keep evolving. So the Text prompting needs to match the scene now.
For example: Asking for wet photo in office will mostly get blocked compared to a beach scene.

[faint lodge]

this was made over whisk

[hollow cobalt]

Can I see your projects ?

[glad yacht]

At present, their safety guard rail is so horrible, can't work with any photo or subject that's minor. For example, fix or restore a family portrait etc...

[faint lodge]

whisk made those minors

[rigid nest]

I particularly miss the Image to Text conversion for uploaded photo from WHISK days.

It was so useful for fine tuning or matching style.

[faint lodge]

lol it was a little inconvenient, but kind of hilarious

[turbid oriole]

I am a traditional artist first

[faint lodge]

this is what flow pinches out

[turbid oriole]

so I always manage around the limits

[rigid nest]

My project - Generated with Flow 🙂

[faint lodge]

ok.... i do have a lot of those actually

that looks really familiar

[glad yacht]

The Whisk/Flow Google team refused to acknowledge this problems like it's the King's new clothes 😅.

[faint lodge]

yeah the image to text conversion was coool. it kind of dissapeared a while ago tho yeah?

[rigid nest]

Yes.

Also the errors I get are from Whisk Generations.

They were OK in Whisk, but after migration they do not load anymore

[faint lodge]

oohhh yeah yeah

i just skewed the image a little

i got around making epstein pretty ez

[glad yacht]

Had to throw Gemini to do a Whisk image to text conversion...

[turbid oriole]

this happens to me after I break the rules on purpose. like to make grotesque grore of maneating giants or s+r+a

[rigid nest]

As I said, these were whisk generated photos. No rules broken.

Mostly portraits and no minors involved

[glad yacht]

Using Google Canvas to create a Whisk Clone...

[faint lodge]

i just needed to distort the facial recognition placement of the features

that is what my friend does.....

it seems tedious

but i guess i should try it

[hollow cobalt]

It will be better if you change into women with blue eyes

[rigid nest]

The internal engine treats it like a different person and that's why guardrails do not flag it as prominent person.

[faint lodge]

i am a woman with two blue eyes.

[glad yacht]

Did Flow change the upscale format? From png to jpg?

[rigid nest]

I think they should allow convited people like him to be shamed with AI 🙂

[hollow cobalt]

I don't know why my whole life All women with blue eyes helped me .

[faint lodge]

theres an interview with andrew bustamante where that guy literally says: we are too wealthy and educated to talk about jeffrey epstein. its pointless

lol

[rigid nest]

Educated surely, wealty only first 3 days of the month 😂

[faint lodge]

im sure the crimes we know about aint nothing on the one theyre getting dialed in rn

[rigid nest]

No, I still get PNG

[rigid nest]

Not even the tip of the iceberg

[turbid oriole]

you know.., 30+ regens and constant :spammy: recreation of the same exact prompt and image is exactly what is considered a soft violation. Gen AI abuse.

[rigid nest]

have fun, good talking to all. Great community. I am going out 🙂

[hollow cobalt]

Does anybody has youtube channel
Can I see ?

[unreal viper]

Let’s steer clear of sensitive or potentially divisive topics, this includes all mentions of politics and religion. We aim to keep this space welcoming and comfortable for all ages and backgrounds

if you wish to discuss a moderation action please DM @real wyvern to speak to the moderation team.

[hollow cobalt]

Why it tells us to log in discord again if I dm it .this can hack other discord. This is not safe

[unreal viper]

This bot is used in many other servers, it requires a log into discord to prevent spam tickets from being opened. this is the easiest way to speak to the moderation team in a private and confidential manner however if you do not feel comfortable with this method you are entitled to your opinion.

[hollow cobalt]

Why you are u here ? Are u not part of moderation team.420,000 members and not single is active
It looks you deleted everybody message

[manic niche]

Is uploading to Flow still not working?

[turbid oriole]

🤔

reset was done. haven't been back to check

[manic niche]

I keep getting this, even on incognito, and even after clearing cache

[turbid oriole]

did you try "deny"?

[glad yacht]

I noticed, the recent upscale 2k files size had shrink to under 1mb, where earlier last month the file would been 3m to 4m size. I checked the file extension, it had been switched to jpeg instead png.

This explains why running edit 2 or more attempts, the image quality is failing apart.. 🤔

[turbid oriole]

was that today?

whups 👽

[pulsar coral]

Yaska when people right click save as on a Flow generated image they can still save them as PNG, but when you use the download button on the image for whatever resolution that's when it saves them as .jpeg (which is the same as jpg)

That's why the other user still gets png, probably he's right clicking then saving

I did complain about it at bugs labs feedback like a week ago

[glad yacht]

Also your reply in whisk channel was deleted, I wasn't able to see it.

[glad yacht]

Did you compare the file size savings when you choose download with different format? Are they the same size?

[pulsar coral]

I don't think they deleted it, more like the whole Whisk channels are gone... and it was full of tips on how to help people who still needed a site like Whisk.

Seriously why wasn't it kept visible like an archive, I'm sure that can be done.

[glad yacht]

Well deleted the whole channel is the same as deleted your reply 🤣

[pulsar coral]

I guess, oh yeah you asked me about the Gemini canvas uspcaled image's size, yeah it was 20 mb or so, the usual 2K image is around 5-7 mbs. Posted a screen too but it's gone.

[glad yacht]

Hide it from future users to see 💢 angry post about their decisions 😅

[hollow cobalt]

Please make me admin .I will make this server best in world

[pulsar coral]

Guess I can try to repost the NB1 canvas prompts later at the tips and tricks area, it needed a rehaul anyway since it only was working with NB1... woulda been nice to have all the data visible, I swear I thought it was gonna be kept as archive as logic dictates. They really don't care for help topics.

[main saddle]

messages that violate server rules or guidelines can be removed, outlined in #rules-and-guidelines・📋. feel free to DM @real wyvern if you have questions.

[rigid nest]

Uploading works. perhaps the file is corrupted or flagged as objectionable by Flow?

Does it happen to all file types (JPG, PNG, WEBP) etc. and for more than one file?

[glad yacht]

Seems I had experienced the same upload issues 😅. Multiple images uploaded weren't working. Only manually uploading 1 by 1 was working yesterday 🤔.

[rigid nest]

I just tried uploading the below image and it worked. Could be an user profile specific issue?

[pulsar coral]

Yeah that one image was downloaded as 20 mbs archive, it wasn't zipped or anything. Flow's now saves 3:4 as around 2-3 mb files due to being jpeg.

As the bugged PNG it used to save images as it was 5-6 mbs each image, I just batch convert them back to PNG because I don't want people to think I'm saving JPEG images in purpose, I'd be the laugh stock.

That one's PNG because it's how the Gemini canvas mode saves them as.

Eitherway I'll try to make a canvas upscaler again later, my prompts from the Whisk thread are gone and not very sure if it'd be ok to place them at tips and tricks anymore tbh

[keen hinge]

Right now I have problems with downloads not getting upscaled - in Gemini it just drops a 1024 without info (sigh) ... and in flow it just fails with a very short popup notice (that is the one notification I want to stay and "dismiss" - damnit ... why is that one automated??? 🤣 ).

[main saddle]

hey all! we're currently investigating errors with upscaling images within Flow. thanks for reporting!

[manic niche]

Yeah I've tried multiple different types of files

[proven onyx]

all sorts of messed up

[glad yacht]

All Flow in the wrong direction 😅🤣...

[pulsar coral]

Funny enough without Whisk's threads there's nothing to do around here anymore, that's where business was and people were asking things about. Flow is just Flow now.

[burnt canopy]

working on flow has really been so frustrating lately

[frosty niche]

What do you mean? You can test the tools out in the More Labs Tools & Experiments areas, you can post in the #todays-special・🥣, you can help out in the #flow-tips-n-tricks・✏️, you can post in both the #flow-images・🖼️ & #flow-videos・📹 channels. Theres tons to do here. You can test out in the #flow-app・📱 if you got that access. Theres tons of options.

[pulsar coral]

No one's asking anything anymore tho, what do I do talk to myself.
Even tho some people still talk, without Whisk's channels Flow did just become boring and did visibly lose a good %60-70% traffic. Also apps are for users with phones only tbh, kinda wasteful when using a good enough pc

[frosty niche]

Simply saw & replied, & what do you mean this channel is for everyone not just you.

Yeah I mean if you go on about the samething every day, it becomes stale over time pretty quickly. Thats unless you have actual technical problems with the active software that is. You can go on & on, that means everyone else can as well.

[pulsar coral]

idk Whisk matters did give a lot more to do, only noticed now that it became deserted. Well it'll be what it'll be with Flow, don't worry about me I'll just workflow or something instead.

[frosty niche]

I agree, but you can either adapt or continue your tirade up to you man.

[pulsar coral]

Adapt is the right thing yes. Not gonna rant forever don't worry.

[frosty niche]

Just letting you know I'm here for you dawg. Anything shutting down as big as Whisk was instinctly blows, but we'll be fine! Everyone we'll get through it.

[near garden]

How realistic would i be able to make like 5-7 min videos for updates about a game using the videos?
I've been making models and character sheets for the last few days and only now i realized that if i want to voice them, i'm going to have to pay premium coins to get their voices and animation going..
I could just make short cuts back to back, but it seems like every video is going to be costly to make

If I wanted to make weekly updates for my friends, i'm probably going to need to make chibis and then text to speech, but then the TTS voices won't match the voices from google flow if i'm using cap cut huh...

[burnt canopy]

4hrs later, still the same with flow 🤦‍♂️

[tough yacht]

This bug is now fixed - so sorry for the hassle and thank you for pointing it out!

[verbal laurel]

How to create a video about 2 minutes long

Each video I create is only a few seconds long

[near garden]

there's a start and end frame when you make vids. you can screenshot the last image on your video to 'extend' but there's no true video extender. just make 8 second clips and edit them together one after the other

you pretty much just need to constantly be making scenes over and over and tieing them together

[verbal laurel]

tysm

[frosty niche]

https://tenor.com/view/cockroach-jackieland-bug-bugs-insect-gif-17679207

[nimble berry]

What the best prompt to fruit trend in flow with Images of the characters

[crisp basalt]

YESSS thank youuu

also view all changelogs page did not lead to the changelog page 😅

[placid sluice]

whoops...didn't realize Flow App was only on Android, almost signed up to be a tester 🙁

[main saddle]

TestFlight coming soon!

[rigid nest]

Wow, I do not have this yet.

But useful addition. I had made a similar comment few weeks ago

[ionic halo]

does anyone know why editing mode is not available for a video? I have not seen this before and I am stuck

[round fern]

Hello, can i DM you?

[pulsar coral]

I'm alive yes

[pulsar coral]

Dude's also telling me he's still hoping some of Whisk's features alike Subject/style/scene/ will be added to Flow soon, since some staff at the old Whisk thread hinted at related updates. I rather still just have NB1 back tbh.

[rancid silo]

What happened to delete button i cant delete images any more...

Even the one on the side is gone

[plucky vector]

it was mere moments ago that I decided to click on the "batch" view option, and realized it perfectly addresses several of my gripes about the Flow interface. It groups them up nicely, shows the prompt on the side, and lets you bulk download them. Very cool!

[violet smelt]

I'm in Australia. I'd like to know if the first 3-month ultra plan promotion is still available? Thank you.

[pulsar coral]

... people are messaging me about how to run the Whisk-like Gemini canvas since I think they might be reluctant to ask around at the Flow thread now since might not be so much on topic, but yeah I'll check and answer when I'm around.

Yaska did also confirm just in time that a canvas wasn't gonna get better than running NB1 anyway. No Whisk user's still wanting to deal with Flow's NB2/Pro seems, go figure.

[pulsar coral]

If it's not availaible for you here anymore more likely isn't. Tho maybe it'd still be there from a different account. It works weird.

https://one.google.com/ai?utm_source=gemini

[pulsar coral]

You can send them to the archive and over there they do have the delete button. Archive works more like a recycle bin than an archive, it was necessary like that because people kept deleting images by mistake. You'd rather not lose anything tbh.

Man I can almost see the tumbleweeds rolling around here.

[distant ferry]

Hey guys, is it just me or is Veo3 acting up today? I can’t get a single decent render, no matter if I use old or new prompts. The output is totally ignoring my instructions—the action is just way off. Is anyone else running into this?

[stone garden]

u_u

[kind dirge]

Is anyone missing a delete button to delete the image you generated. I don't see the trash can icon

[unreal solstice]

Does anyone know how to add an accent for the characters?

[frosty niche]

Try adding it specifically into your prompt description before you generate.

[wet flame]

Same here

[wet flame]

It's the "Archive" now, you can archive your generative content, then delete them later

[inland sorrel]

WHAT HAPPEND TO WHISK CHANNEL

[kind dirge]

THAT DUMB AND STUPID. I rather have delete button than archive button.

[wet flame]

There's no return after deleting so I'd say that's an upgrade, incase you accidentally deleted something

[wet flame]

Gone, they shut it downed

And Flow was okay a week ago (I was away for a week), and now there's some unusual problems again

[strange forum]

When I put a face reference image to make an image .. I’m getting a random image or it won’t make…without the reference it makes the image that is prompted

[wet flame]

What's your prompt?

[strange forum]

It’s a long json

[wet flame]

🤨

How long are we talking about

[strange forum]

I can dm you it

[wet flame]

Sure

[rigid nest]

Archieve = Delete which allows you to undo accidental deletions. Delete = Permanently gone

[frosty niche]

Sounds like an infrastructure approach/waterfall tbh. I'm sure they'll add more to it later.

[rigid nest]

It's a tool in constant development. I like the idea of achieve, but I hate that we can not select multiple files to achieve at once.

Also it makes no sense to me to have project specific archieve insted of a common bin.

[frosty niche]

That sounds like exe talk here. But I see what you mean, I'm sure there will be more updates coming soon.

[rigid nest]

Ofcourse there will be more updates. Waiting for Nano Banana 5 🙂

[frosty niche]

My mans is thinking future, future! Nice. Yeah i can only imagine what that model will be capable of, it'll probably come with a small mobile 3D printer of some sort, 😂 😅!

[rigid nest]

It will come with Super intelligence. So it will read my mind and generate images and videos while I am sleeping

It will be nice to have the ability to train GEMS which can be specialised in specific image or video types.
Sort of like LORA

[frosty niche]

Sounds pretty efficient or a replicated version of ourselves, which I believe Meta was doing over on FB awhile back.

Interesting, ehh maybe never know what the future will bring.

[rigid nest]

You mean the virtual reality and augmented reality projects?

I think Google is positioned really well in AI race.
Not just Gemini, Agents, but also in terms of hardware and quantum computing.

The other day Warren Buffet invested in Google after seeking my advise 😂

[frosty niche]

Hahahah don't get me started on that distraction project 😆!

Quantum is so interesting much needed in the race just dont care for the name they went with to be honest, but cool stuff either way.

Woahhh the buffet man, that boi is stacked!

[rigid nest]

What's there in the name.

Quantum and An Alias Name can co-exist, just like Qbits 😂

[frosty niche]

I think it was the name of a tree, should of went with a Redwood or Sequioa if they wanted to go the tree approach for the naming convention, but hey what can you do.

[rigid nest]

I think it was the name of a tree,

[tepid wigeon]

Anyone else unable to scrub to the last frame of a paused video in the Flow web site UI? Every time I do, the video automatically loops to the beginning and starts playing. Makes it impossible to select the last frame to export as an image.

[fierce nimbus]

you can select as many as you want to delete.

[hybrid pawn]

I think you guys need to get rid of lite and keep fast. Lite is terrible!

[merry mango]

I'm having an issue with Flow: when I create an image, it gets saved outside my collection, directly in the app's root directory—which is really annoying. Also, what does “set project cover” mean on an image? I get the impression that’s where the bug is coming from. Also, why don’t you let us know which collection an image belongs to when we’re sorting images? Why can’t we quickly assign an image to the collection we want?

[rigid nest]

Thanks @fierce nimbus. It was not a feature when I tried last.
Glad we can do that now. 🙂

[olive bear]

had a question, anyway to make the prompt not go away after u click the arrow to generate?

kinda annoying to keep copy n pasting n maybe a change everytime i want to generate sum XD

[sweet scarab]

How can I turn image to videos

[crisp basalt]

i think you guys mistakenly... disabled something for r2v model?

i cant even use any asset/reference image to generate video using r2v

[olive bear]

tyty

[lunar root]

Anyone else noticing ingredients is suddently locked/disabled on all versions? Or is this just me? 🤔

[tropic flume]

Ingredients is not working for me either.

[crisp basalt]

You guys probably need to check the new line I saw, which is Requirement_Character 👁️ for the video model requirement. @/google staff

[tropic flume]

It seems to be working for me now.

[lunar root]

Nice....yeah, it's back on. I freaked out when i saw it was gone across the board. lol

[azure peak]

anyone use dreamina before?
their prompt guide seem to be outdated

[glad yacht]

Their original version is Chinese, several versions ahead of global version 🤔

[tulip lake]

Good day, I started generating images using a celebrity's name on Flow without triggering the prominent person filter, and I was thinking it is still a policy violation.

[lusty sinew]

Hi, I'm having an error where I can only generate 4 images at a time, otherwise I get a "you're generating images too quickly" error.

[pulsar coral]

Don't see much changes on it tbh, probably that one character is not considered "Prominent" enough.

To generate character images, Flow's “Nano Banana 2/Pro” seems to rely more on Gemini asking Imagen 4 to recreate or trace faces from gathered training datasets in order to make characters look as accurate as possible, rather than relying on more organic artistry like the older Nano Banana (Gemini 2.5 Flash Image) used to do, which honestly felt better in my opinion.

From what I’ve observed, this appears to work heavily from images found on wiki gallery pages, because I’ve caught NB2 generating fictional characters using faces and appearances that look almost identical to images commonly found in wiki galleries multiple times.

Point is, For NB2/Pro American actors and celebrities seem to be treated as far more protected or restricted than most fictional characters or media from other countries, since those and fictional characters are often generated much more easily by comparison. That might be why your character is still passing through the filters.

It honestly feels like a biased standard sometimes, because not everything is treated equally as “copyrighted” or “prominent,” even though NB2 clearly seems capable of recognizing the difference.

I imagine that if Flow's NB2/Pro applied those restrictions equally to everything, it would probably end up generating even fewer images than it already does, which is already a major limitation in its current state, cause unlike Imagen models and Nano Banana (Gemini 2.5 flash image) which is what Whisk used to have, it mostly can't create art by itself and just traces images.

[simple stone]

i found there is no DELETE anymore in three dots menu. why?

[lusty sinew]

Can someone explain this bug?

I can't generate more than 4 images at a time anymore.

[pulsar coral]

The archive works like a recycle bin despite it's name, send anything you want to delete there and it will have a delete button over there.

[frosty niche]

Wait acouple of hours and it'll load outputs correctly. Its usually triggered when you generate 1 or more generations at a time.

[pulsar coral]

I think that's the daily limit tho.

[lusty sinew]

I've hardly touched flow today.

I haven't even started using it since about an hour ago.

[pulsar coral]

Try what Gaston says anyway because you never know with Flow.

Gonna paste my standard stuff tho.

[pulsar coral]

If it was working fine some hours ago maybe your Flow account just hit a daily quota for images, Flow sends the wrong error message for that, 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, they don't exhaust at the same time, but if both are dry Imagen 4 won't work either so don't bother with it either since NB models use it as base. Your Flow account actually runs out of Pro images much faster cause they cost more.

Sometimes clearing cookies fixes it too but more often if it keeps happening after like 5 mins after that your account is just dry for the day.

Your daily limit wont reset after 24 hours exactly, but more around a little later after it was working properly the day before.

At free accounts the "You're going too fast" message does also get fixed sometimes if you clear cookies because it's also used as a limiter, you'll only know it's dry after you do that and finally does stop working for the day.

But Plus and above do get a litle more priority against errors, so more likely it should be dry already.

[frosty niche]

Also, try and swap your browser if you can, it may work out a bit. Just a heads up if you generate multiple gens this happens alot even if you just logged on.

[pulsar coral]

But yeah it's not exactly 24 hours for the reset, more like a little later after it was working the day before, one thing or the other will work.

[frosty niche]

Nice wrap sheet here @pulsar coral

[lusty sinew]

Thanks both for the help. This is the first time this has really happened where it just refuses me outright after I haven't even touched it for the whole day.

[pulsar coral]

Yeah but it wont have any mercy with you if you were there just like 5-12 hours ago. Needs the whole day cycle at the least.

That if clearing cookies didn't work

[lusty sinew]

No, I think It's been about 24 hours since I touched flow. I haven't been on all day.

[pulsar coral]

It probably just needs a little longer then

[frosty niche]

You know they should have a daily timer for the next daily reset, thoughts?

[lusty sinew]

I think so, since Google Gemini has one.

It would help us determine how long to wait before we could know the reset time.

[pulsar coral]

Oh right I totally forgot it does have one, and it's been there all the time. It should be possible to prompt it at Flow then since both websites are probably generated with the same thing.

[frosty niche]

Wait google gemini has one, for real?

[pulsar coral]

Yeah once you run out of images it tells at which time of your own time zone you can come back for more.

They should bring whatever guy prompted that over there to do it here.. maybe the canvas mode can generate a page with it too

The message comes right under you where you tried to make an image and failed

[lusty sinew]

It's a wonder why they don't bring it over to flow, especially for pro or ultra users. I have a google pro account and it would be nice to have.

[lusty sinew]

I would like if they had it as a "reasoning" why the image generation failed, actually.

[pulsar coral]

Well you can still mark it around the last time it was working yesterday and come back a little later, I do that every day tbh

[frosty niche]

Theres not a sole feedback channel just the #1238621511401734254?

[frosty niche]

Some prompt description failed logs for a generation, aside from the error message I'm assuming, interesting.

[lusty sinew]

I would rather be told "hey you're out of generations for now" instead of "something went wrong" or "you're out" when those aren't the reasons.

[frosty niche]

Dont forget about the timer!

[lusty sinew]

I could even accept the timer in place of an error, like show that to me instead of a "something went wrong"

[frosty niche]

I sent a feedback report directly in to flow you can do it as well if you want heres what I added.

"Can we add a timer for when we run out of daily generations for future updates?

That way for consistent users it can help arrange a bit better when interacting with the site/app.."

[lusty sinew]

Thank you, I appreciate it!

[simple stone]

thanx. found it 😄
but instead it takes several steps just to delete something :))

[frosty niche]

No worries.

[lusty sinew]

Do you know, by chance, why they opted to add an archive and not just keep the delete option?

[pulsar coral]

People kept deleting things by mistake when the pages failed to load and they clicked buttons by accident, and it was much worse like that tbh. You wouldn't like to lose something.

I did ask for the feature a while ago but labeled it "Recycle bin". Name seemed more logical.

[gleaming lynx]

why was the delete button removed?

also, after the delete button was removed I could drag select images and delete them

but now that archives them too

I can delete them from the archive but, if you are letting me delete the images after all this BS why not just give the delete button back?

[full blade]

Can it please stop showing the "Unusual Activity Error"-message please?

[random river]

I’ve been testing Flow for a while now, and honestly, I really miss the magic that Whisk had. What made Whisk special wasn’t only the image quality, but the whole creative experience around it. The ability to combine multiple image references (subject, scene, style) with almost no prompting felt incredibly intuitive and inspiring. It allowed ideas to emerge naturally instead of forcing the user to over-direct everything.

With Flow, the generations often feel more rigid and less spontaneous. Even when the output is technically strong, it lacks that cinematic unpredictability and “happy accident” feeling that made Whisk so addictive creatively. The UI in Whisk was also much cleaner and easier to experiment with quickly.

@random riverHI

Would it be possible to reintegrate some of the core Whisk features into Flow? Especially:

Multi-image reference mixing (subject + environment + style)
Minimal prompting workflows
A more lightweight and intuitive creative UI
More emphasis on experimentation and visual mood rather than strict prompt interpretation

I genuinely believe Whisk had something unique that many creators still miss today. Flow has potential, but right now it feels like some of the creative soul that made Whisk special got lost in the transition.

[lone seal]

So is there any way to change the thumbnail of each project in Flow? I tend to start with a reference image so I have like half a dozen with the exact same thumbnail and its annoying

[paper ermine]

What's wrong w VEO lately? Start frame - Endframe don't work anymore. The model is too lazy to think or sum. All i wanted is the house transform into a different house. It started with create a complete different house then fade transition into End frame. Wth?

[stone garden]

Has there been changes again? Whenever I try to generate a bunch of more than 8, I keep getting errors about how I 'ask for too many generations' - It never did that before.

[lone seal]

I'm getting that too. very annoying

[stone garden]

At least, it's not just me. Still, sorry to read it's not just me u_u"
Really annoying indeed

[blissful abyss]

The flow app if nice

[shrewd raven]

Flow in general is pretty useless unless you are paying the $250. The workflow crippling leaves the best features behind a paywall that individuals won't pay. This is an enterprise product and no app, that isn't very good, is going to sell flow. It needs to be unlocked for any type of honest assessment.
As a Pro tier customer i would love to try out the best features and maybe even pay a little more but it is priced out of people's range.

[lone seal]

I don't know, do the "too many images too quickly" and "this looks like a prominent person (when it's not)" messages go away if you pay them $250?

[candid bay]

Nah, the filters are overly or badly tuned

[tropic laurel]

will this ever be fixed?

[lone seal]

I got that too just now, but was able to generate again immediately after

[brittle bloom]

For me use reprompt

[wintry pecan]

I'm an Ultra subscriber, but I've only created a few videos and I'm already seeing this limit message. How can I fix this

[frosty niche]

Unsure on the official technicalities, but if your on that subscription level thats a bummer, try to change your browser or restart your device maybe it'll help with a quicker wait time.

[rigid nest]

It takes up to 12 hours. Might be faster for Ultra subscription.

I had requested to improve the error message and have a timer, to remove the guess work.

[surreal shard]

Hi, any news on when Whisk is coming back? Flow is an unusable downgrade and many people are months/years deep into projects that suddenly got the brakes slammed on them.

[rigid nest]

As fas as I know, Whisk is not making a come back

[exotic shell]

last few days of veo 3.1 fast (low prio)
enjoy before its gone

[rigid nest]

Hello Google Team,

Can you please clarify what is the main purpose of Flow at this point?

This forum is filled with reports of a steep decline in user experience with recent changes to Flow and the overwhelming bugs.
Flow was supposed to be a better platform than whisk, but so far it feels like a half baked product.

We have no way of tracking bugs and progress made since reporting an issue.
How can we contribute more effectively to improve Flow?

[tropic laurel]

it's so frustrating, i can't generate more than 2 videos without refreshing the page

[lusty sinew]

It’s happening to every flow user that I talk to, it’s frustrating they won’t address this at all. This happens way before Gen limit now despite what some people have said as a solution.

[gloomy verge]

Hello. Can I ask, why when I create video, always is in slow motion? Can I create video in normal speed too? If yes, how I must write prompt, if I want result video in normal speed?

[lusty sinew]

This also isn’t fair since the “you’re requesting too quickly” eats up actual generations

[vagrant narwhal]

I'm moving on to Seedance, i'm not dealing with their "prominent people" bug anymore lol

[exotic shell]

anybody has a clue what could the reason that my ultra subscription produces videos with a veo logo? is there any trick to get rid of it?

[last furnace]

It happen to me, and with images some times when I try download the file is not what shows in the screen

[unreal viper]

hi, please use #1238621511401734254 to submit this bug! these threads are actively monitored by the team and so may be seen faster!

these bugs wont be fixed if the team doesnt know how many people its effecting!

[crisp basalt]

Has anyone managed to find a solution for this? I’ve tried all kinds of prompts and tested Veo 3.1 Fast/Lite/Quality, but no matter what, it still won’t generate a seamless start-to-end frame. It always hallucinates random elements that were never even part of the original image

Honestly, I suddenly miss the old Veo 2 start-end frame model. I never had these inconsistency issues before no matter how I prompted it, it could still maintain a proper seamless transition. Now with Veo 3.1, it keeps hallucinating random stuff that wasn’t even in the original image/frame. 😅

https://labs.google/fx/tools/flow/shared/video/34e7e7c8-b10b-480c-8787-16e6445777cd veo why 😭

trying to make sense of whatever this is like
ok so this is the end frame of the house

veo gave me :

https://tenor.com/view/frustration-hands-through-hair-gif-21162137

I DIDNT EVEN PROMPT "ROBOT"

if anyone want to give it a try or at least try to find solution for this : https://labs.google/fx/tools/flow/shared/video/34e7e7c8-b10b-480c-8787-16e6445777cd

[astral yacht]

Good morning Discord! I'm excited to announce that we have expanded our Android Beta program to 10k spots! Additionally soon you should see the following:

1. no more false errors on video generations (and then they show up on desktop)
2. crash fixes
3. a pretty new icon 🌈

We will be fast following with more updates and features shortly!

Click here to join from your phone, or click here to join from web
You'll need Android OS 7.0+. For the best device experience use a Pixel 6 or higher.

I promise iOS is coming so soon!

[crisp basalt]

https://tenor.com/view/sad-gif-16315795042465281757

[lusty sinew]

https://tenor.com/view/disappointed-sad-fridge-no-food-gif-5296071

[severe lily]

The continuous failed imagery due to "unusual activity" is killing me. There's nothing controversial, it's worked before and it's randomly 10+ shots that get flagged.

[crisp basalt]

that error means its not your prompt violating rules or anything, its just server issues lol