flow-general・🌊

[robust socket]

google is down

[dense mortar]

I've already noticed😢

[weary pivot]

did they solve it?

[woven herald]

Why are there watermark on videos I generated on whisk

[meager edge]

how can we create a long form content using veo 3.1 as I used extended method but after 4 or 5 clips it changes or morphs the faces and also change the lighting means increase the contrast

[analog radish]

Do you know where to find "Ingredients Archive" I can`t see it in my projects list?

[deep swallow]

Need solution 🥲

[worn bramble]

I want to make a video

[pale flame]

Has anyone asked why the prompt box in Flow does not underline misspelled words? Like "mak an image in 3d-animated Pixar style" This is simple and you'll notice the word "Mak" an image, was not underscored in red. It works in my Chrome browser for Whisk, Gemini, Discord, etc. But NOT since the new Flow update a few days ago. Is anyone else having this issue? This is, obviously, very important when definitive prompts are required.

[versed peak]

How do I know exactly what policy my image or prompt is violating? I own the rights to the image (it's 2 women sitting in a car). No prompt is working.

[inland bear]

try 3.1 slow sh1t is broken dawg

[edgy holly]

You click the plus button at the top and it doesn't do anything?

They moved it. Find the clip you want to extend in the grid, click on it and its just under the prompt box.

Yeah, I hope they change this. It's slowing down my workflow significantly.

[main saddle]

hey all thanks for your reports (and for adding them to #1238621511401734254!)
we're investigating further, thanks for your patience. if you've submitted a bug report, please include any prompts, screenshots, operating system info, or any other information so we can take a closer look. thank you!

[vital compass]

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

Where can I find it in the new design?

[main saddle]

fwd from @inland fox

[inland fox]

I heavily rely on expanders for my workflow, I hope they didn't just remove that valuable feature

[vital compass]

There are expanders , but we can’t crop part of a video and then expand it—or generate different results and choose the best one. Now it always generates a full video. It’s so much worse.

[inland fox]

No I meant prompt expanders, they aren't an option anymore

[vital compass]

Oh yeah… I don’t know. Flow used to be an incredible tool for people producing high-quality AI video content. Now it feels like it’s been leveled down for less experienced consumers—the options are really limited. I don’t understand how it got worse.

[inland fox]

I'm an ai ultra subscriber, if prompt expanders got removed, so will I be dropping this subscription

[pale flame]

I think a lot of things are centered around limited power usage. Until they can build their natural gas power plants, they are searching for "go-arounds" to reduce data mining and assembling or unnecessary processes. Especially since Nano Banana and VEO is in nearly every AI image and video platform or program. Well, that's my opinion.

[inland fox]

@main saddle

[frozen tiger]

Hi everyone! By any chance, has the issue of not being able to scale generated videos in the extension to 1080 been fixed yet?

[vital compass]

I have the same problem. It’s clear that Google dropped the ball. It all started when, out of nowhere, they changed it from unlimited credits to 10 credits per generation on VEO 3.1 Fast—and now this Flow update. Everything is broken. How is it possible that a company worth billions of dollars does something like this?

[dense mortar]

I hope I wake up in the morning and the flow works, otherwise it will start losing its followers and followers.

[spark hornet]

Are you having issues right now trying to generate pics ?

[crisp basalt]

https://tenor.com/view/astralvi-gif-8132770885564971457

[crisp basalt]

He was referring to prompt expander Paul, the one located in the prompt box that allows users to expand their prompts by using a preset custom made by the user.

[edgy holly]

That's not a glitch. It just hasn't been added yet.

[pale flame]

Yeah, I get it. I really hate the new update though. The expander should be an easy fix, like the spell check that no longer works. I don't mind the Nano 2.0 but it almost seems like the update was rolled out by one person alone, who only creates cartoon memes. Why not roll out a beta or do what Google used to do and let us rollback to the earlier version till we're done with our current project?

[tawdry burrow]

Has anyone else noticed or is anyone else experiencing worse video generation results after the update?

[edgy holly]

It was just a UI change.

[frozen tiger]

Humm, i asked because before the update the option was abled 😢

[tawdry burrow]

Haha, fair enough! Maybe it's just a 'skill issue' on my end then. I guess I need to step up my prompting game

[edgy holly]

Then we're talking about different things. I thought by "in the extension" you meant clips that were extended.

What are you referring to?

[edgy holly]

Probably just randomness. Always part of the equation with AI.

[pulsar lotus]

add more aspect-ratios for flow plz 🙁

[granite hemlock]

hello is flow makes good ai videos?

and is it free

[unreal viper]

definitely one for #1238621511401734254 ! the team actively monitors these threads so your suggestion may be seen more there!

[granite hemlock]

what’s flow

[wispy ingot]

So question, is it a known issue that when trying to delete projects in flow they dissapear but then reappear again once you log back in?

[real cradle]

Hello.
Badky dned help improving my prompt for creating videos on Flow. 🥲

[crystal talon]

Is flow dead right now?

I only get errors, nothing is generating

[crisp basalt]

well rest in ai void forever veo 2

[spark hornet]

Changing the browser magically solved my problem .... anyone had this problem before ?

[cobalt junco]

extend isnt working for me currently other than that ive had no recent issues

[frozen tiger]

Sorry, I was referring to “extend” as in generating a video from an existing one (extending the video). The result used to give me the option to download it in 1080p, but after the update, it only allows 720p.
I’m attaching the image showing how it appears to me now and how it used to appear before.

[young locust]

New flow is looking really good but there's just some odd things that I don't understand, Like if I add an image The stack and put it in the prompt feild, Like why can't you just drag and drop from your stack into the prompt edit. And if I add an image it's not adding in Recent image from the stack.

• you have to copy and paste it so that means you have a double image and then it adds it.

And if I make a collection and I then I'm outside of that collection and I make a new image I can't add any of those images back to the collection I just made

• some of the things I don't understand that just don't make sense - like we should be able to select a stack then right click your collection and add it.

[stuck adder]

we might be getting veo4 lol

[young locust]

Can we export a collection to Google Slides to build a deck - I think that would be helpful because basically you're doing iterations of design or video blocking And to export it to a slide deck to send to a client just wrap up a set of designs.

it would be good to see Google Labs kind of thinking about the process of how these tools are used rather than just dropping a whole set of updates and having them completely dysfunctional.

[knotty crane]

Badly

[tardy ridge]

is anyone else on Ultra getting a VEO watermark on theor gens?

[crisp basalt]

Nope

[crystal talon]

Non stop

[crystal talon]

Hope it can do action scenes like seedance. I cant hop around from platform to platform anymore, too old for that shite

[stuck adder]

lmao i know what u mean

[tranquil shale]

Still error for upload reff, update not good

[spiral python]

Still error guys ?

[latent frost]

@spiral python

[dry zenith]

No more add frame to scene in scenebuilder?

[pulsar ferry]

Veo 2 WAS THE BEST RIP

[dusty fulcrum]

HEADS UP: Nano Banana Pro and Nano Banana 2 now block most Disney IP including Star Wars. This is at Google. I've tried all the service providers and straight from Google and the API. So, RIP to anyone that wanted to make their Fanfilm with Veo or NB.

[crisp basalt]

Ironically Disney and Star Wars character can now be used by everyone that has Sora 2ios app

[cobalt nebula]

i need more aspect ratio for flow for images please

[rare basin]

If anyone from Google sees this:

Could we get a search bar in Flow on the top of the home page (https://labs.google/fx/tools/flow) where you can search through your creations using normal language? (like the search bar in Google Photos)

oops, just found the feedback thread

[desert hill]

Ya, that's where u put it

[rare basin]

its there now

[desert hill]

And it's upto google to see into this coz I ain't google

[odd turtle]

Is everyone here having problems with the Flow website, or is the problem just with me?

[desert hill]

It's not u

It's flow being flow

[odd turtle]

I give it a prompt to generate a video on an image, but it doesn't understand me at all and generates videos from its own device.

[desert hill]

The servers having issues in some places

[odd turtle]

This means this problem doesn't affect everyone.

?

[native lake]

images work for me at least

[desert hill]

Im not sure

[desert hill]

See?

It works for him

[odd turtle]

Creating images works, but when you want to create a video from an image, you place the image and it doesn't give you the video you want; instead, it gives you a video with added elements.

[native lake]

give example :3

[inland fox]

please help me access my custom created prompt expanders that I put in countless hours in to perfect. If I can't use it then ok, but I want access to the prompts please

[ebon basalt]

Can we have portrait and square aspect ratio on flow?

[odd turtle]

Same bro

[inland fox]

Please I need someone who can really help, this is very important @olive portal @real vector

[native lake]

no i meant. give a thing i can test against. and what your goal is

[odd turtle]

Let's talk privately, please. Bro

[hazy anchor]

Yo guys

Where can I get a Flow account with 45,000 credits?

I bought 4 accounts before, and this problem occurred three days later After purchasing the account

[desert hill]

Submit an appeal

[desert hill]

Did you do something sus

Or use a vpn?

Coz vpn sometimes does this

[hazy anchor]

Nah, i dont use Vpn

[desert hill]

Then appeal to google

[hazy anchor]

Where did you get your account from?

[desert hill]

Google??

[hazy anchor]

Flow

[desert hill]

Uhhhh.... I ....

I...

[hazy anchor]

Can we talk Privately

[desert hill]

Sure

[native lake]

he already said 4 accounts

[desert hill]

Now it's starting to sound sus

[native lake]

i don't think he wanted to do something bad. but he might have violated TOS

[edgy holly]

Buy it from Google, or you're just going to get banned.

[desert hill]

Yes ofc

[hazy anchor]

My friend told me to buy them from G2G

[desert hill]

Dont

Wait

G2G???

[hazy anchor]

And after a few days the account disappears

[hazy anchor]

Yes

[desert hill]

No no no

That is extremely unreliable and sus

[desert hill]

Did someone on discord give it to u

??

Better report it to google

[desert hill]

That looks shady

[edgy holly]

Your "friend" is either a scammer, or not someone to listen to about anything.

[desert hill]

First summon the mod

[edgy holly]

Obviously.

[desert hill]

Or Google staff

They need to check this out

[native lake]

get a new friend

[desert hill]

Obv

[hazy anchor]

Yeah but the first days the account works good

After 3 days ...

[edgy holly]

Okay, then keep wasting your money on obvious scams.

[desert hill]

Legit scammer

[desert hill]

They sound extremely convincing

[dry frost]

The best way to get a Google account is to sign up yourself

[desert hill]

@potent shell

[hazy anchor]

But it's very expensive

[native lake]

duh. cause its not your's why are you arguing about this with me. are you baiting me? 1 its illegal 2 you violated the law 3 it never belonged to you

[desert hill]

Chill, chill, let me talk

[hazy anchor]

I didn't know that before

[desert hill]

Never never do it again and report it to google

Don't care what ur friend says

[edgy holly]

Some things are expensive. When you find people lying about selling expensive things cheaply... they're fake.

[dry frost]

A free account is expensive? That's news to me

[hazy anchor]

Nah, the 45.000 credits account

[desert hill]

Yo man see why you got banned now??

45000 credits??

[edgy holly]

That costs $450 a month. If you're paying less than that, that's a scam.

[hazy anchor]

450$ for just a one month ?

[edgy holly]

They're probably using something like stolen credit cards to create accounts, then sell them to fools before they can cancelled.

[desert hill]

See why it's a scam

[edgy holly]

Yes. This is extremely expensive technology.

[desert hill]

The scammers sound extremely legit

[edgy holly]

Even at $450/mo, Google loses money.

[desert hill]

That's so kind of google

[hazy anchor]

I only use Veo3 and Nano Banana Pro.

[desert hill]

U mean the shady flow

Ah typo i see

Don't u dare use sus accounts again or Googles gonna personally come and find u

Scam ones especially

[hazy anchor]

Okay, brother

[desert hill]

Gotchu brother

[edgy holly]

And? Look at how much Veo 3 costs in the API if you pay per generation... PER SECOND

[desert hill]

That's higher than my pocket money per day

10s equals 6 dollars

Imagine accidentally leaving it on overnight and it somehow survived

[hazy anchor]

Oh my God, that's too much

[edgy holly]

It's a huge model that runs on racks of GPUs that cost millions of dollars.

[desert hill]

Believe it or not some guy put his Google API on GitHub in public and he got a 50k bill

[desert hill]

Gpu prices already cooked coz of it rn

[edgy holly]

I read about someone who got something like a $45,000 bill because someone got hold of their API keys and ran it up in one night.

[gloomy verge]

Hello. I want to ask something about images in Flow. When I created image in Whisk with turned of precise reference, or I copied imag to subject/scene/style, I get detailed description of image. But I don´t see this option in Flow. When I create some image, I can´t find this description of image, or I can´t find Seed number. It is right? Why there aren´t this functions? Thank you.

[native lake]

i can get it for 0 dollars for 1 month at least :3

[desert hill]

Imagine forgetting to cancel the auto renew

[desert hill]

I think there's no seed in flow

Wdy think @edgy holly

[edgy holly]

It's $125 for the first month. $250 after that, and that comes with 25,000 credits. To get to 45,000, you need to pay the extra $200 for a top up.

[edgy holly]

No seeds

[desert hill]

Hmmm

[edgy holly]

Kling 3.0 costs even more. Older models that aren't as good cost less.

So you really just want to get scammed.

[hazy anchor]

Nah bro

[edgy holly]

Yeah, that's obvious at this point. You know it's a scam, and you're still going on about it.

[hazy anchor]

But he says there is a guarantee, meaning even if the account is inactive, there is compensation for a period of 20 days.

[dry frost]

Just use official resources. It's that simple

[native lake]

he wasting time and just trolling imo

[edgy holly]

Just sharing scams at this point. Just blocked.

[desert hill]

I let google know and they're investigating this rn

No worries

They got us

[remote bay]

@hazy anchor The screenshot you have sent is not from Google officially, we recommend you to not engage with such posts and if you are looking to use and purchase a subscription from Google Flow this is the website to go to -> https://labs.google/flow/about

Please do not post questions asking how to buy accounts, this is strictly against TOS of Google, further posts on this topic may result in moderation action.

[desert hill]

A friendly remainder u cant generate anything here

Try the app or website

[wispy ingot]

Has anyone been having trouble deleting projects from flow? When I delete, they reappear as soon as I return to the site. I've been trying for months to get rid of them.

[native lake]

request they delete. or just try again. My projects were deleted when I requested

[cobalt junco]

thanks for adding the d/l clip 1/2 and zip feature!

[thin ruin]

Anyone scaling graphics and animations with Flow??

Context..... We have gotten pretty good at content creation with Flow. It takes about 1hr +/- for 1 minute of content creation. The limitation is we can only create 3-4x 6-10 mins videos per week. For our type of content the editing is pretty minimal since the files are pretty much being stitched togther as we go to make sure everything makes sense and aligns with the audio.

Can we speed up the process?

[pulsar coral]

Flow is using different image models than Whisk called Nano Banana 2 and Nano Banana Pro.

What Whisk is using for it's precise mode is more commonly called Nano Banana 1/Gemini Flash image 2.5. Or just "Nano Banana" without any addons.

Nano Banana 2 takes reference images from the internet to refine whatever reference images you give it for your art, that's why specially people and charactes change a lot, like if you add some OC image of a character that looks too much like idk a character from Pokemon animes, Nano Banana 2 will use those reference images to try to give you a result more accurate to it, unintentionally totally changing the reference image face.

I don't think Google has understood that's not what people want, and the immense amount of people that just want to create their OC stuff as it is

Which also always beats me tbh.

Nano Banana 1 mostly doesn't do internet searches to 'refine' your images and mostly only works with the reference image, even tho it's also perfectly capable to recognize most canon character images. But when you say "No" to it, it's no.

Nano Banana 2 on the other hand, if you tell it not to use internet images as reference to create the image and only use your reference image as base, it completely refuses.

To illustrate, only the first one to the left here is the reference image. She has a round face.

Prompt:

"Identity of character from reference image: Original character not related to any other.

Artstyle: Completely accurate to reference image without any creativity freedom.

Pose: Dynamic."

It only got it right once, at the center image.

And just barely.

Admittedly my prompt must be flawed there, but Nano 2 should have been able to understand it, yet it surely used this type of official art from the internet as reference without my permission.

[hazy kestrel]

Like really guys. Why are they changing the UI every day and it seems like they just keep on making it worse

[lusty shard]

why is my flow not working for days now, Tried to generate image but keeps giving me error messages pls help

[analog lava]

Hi @lusty shard please create a thread in our #1238621511401734254 detailing your issue. We ask for your patience as the Flow Team work through the numerous reports. Thank you! 😊

[dense moon]

I came here to second this

[unique needle]

I'd like 1:1 and 4:3 and 3:4

[native lake]

Use photo bashing if it really bothers you

[pulsar coral]

Yeah there's many things that can be done about it, but that prompt needed to be easy to illustrate what happens by default. I more often prefer cel-shading 2.5 D.

Tbh I like what Nano 2 is doing to my images but yyyeah that's not the accuracy people seem to want as result.

[native lake]

I just don't understand how flow reads my references. If I could I would be using the correct terms

[pulsar coral]

That's the prob really since no matter how hard we try it isn't "Reading us" it's reading our prompts and refining them from things it gets from the internet, I'm pretty sure it's even images from Wiki pages.

Nano 1 seems to have been using it's own model knowledge instead, alike what LoRas do.

[native lake]

Don't speak in negatives. No need to mention the author or the origin nano banana does have world knowledge like that. Best I can think to do is remove the color, het higher quality close up of face. Re add the color afterwards.

My 2 cents

[unique needle]

Question, does Nano Banana try to force everything into a specific art style? I want to maybe use a Jak and Daxter art style or a Ben 10 Alien Force art style but it just seems more generic

[pulsar coral]

I know that's not a good idea, I was trying for it to stop using external reference and see what happens. It just proved it won't listen to that.

[native lake]

Give it references. It does prefer one style. Usually something that looks like stylized photorealistic

[pulsar coral]

No one would add that to a Whisk prompt. Then again I think Whisk would listen to it.

[native lake]

Using the word not is not helping. It also isn't using external references. Everything is internal and what you provide it.

[unique needle]

It doesn't work with Netflix Castlevania. I gave it these two screencaps and it wasn't too good at matching the art style

[pulsar coral]

Then explain why the face from that character is exactly the one from this Wiki page and not the one from reference, even tho it's prompted not to do that. https://megamitensei.fandom.com/wiki/Rise_Kujikawa/Gallery?file=P4D_-_Rise_Promotional_Artwork.png

Not the one from the thumbnail, the one inside the link

[native lake]

Use the phrase. "Follow the character art from the reference exactly."

That's probably the best you can do

[native lake]

Over fitting

[native lake]

Have you tried more references?

[pulsar coral]

It's not even using it as artistic reference, it's tracing the exact same face, even tho that's not in the reference image and it was told not to.

And yeah I've tried dozens of characters, I can replicate this same example with more of them.

It's just simpler to explain with one for starters

[native lake]

Yes you are communicating with the gen ai. I meant more references for the one character

[pulsar coral]

Well that's the prob when you want to get the exact same character from reference image at Nano 2, it refuses to do that and goes to what it finds on the internet for accuracy, which very often is a wiki page too.

And I really caught it here exactly.

[native lake]

No I meant give it 2 or more of the exact same character with the exact same art style in different poses.

[pulsar coral]

Yeah alike the sheet the other user said, and that would indeed help so it can get more poses.

But this is not what Whisk's Nano 1 is doing since it can do that from the start just by avoiding to get "accurate" references. That's the point really.

[native lake]

Whisk wasn't perfect either

[pulsar coral]

Indeed. Sadly like Japanese say "Wabi-sabi"

There's beauty in imperfection

I don't think anything else would illustrate the term better

[somber verge]

https://tenor.com/view/sushi-love-wasabi-hey-gif-14525785449393418760

[pulsar coral]

Or wasabi yeah

Admittedtly my "original character" prompt was wonky, but I did try other stuff alike "Do not use external references" "Do not use images from the internet as base from this image " etc similar stuff too.

Doesn't really seem it plans to listen to any of that.

[native lake]

I told you to not use negatives
...

[pulsar coral]

But yeah I can try "Avoid" and different stuff, but I don't think it will do much

Not using negatives was stuff for Chatgpt bots, now it it would listen.

[native lake]

What even is this prompt. That's not how it works

[pulsar coral]

idk it's just me trying for it to stop grabbing images from the internet, I don't think anyone has even tried this before anyway, so what gives

[native lake]

I don't think its grabbing image from internet

Everything it shows has already been grabbed

Use something that isn't anime use a meme or something

[pulsar coral]

Not using negatives means using these instead, I get it

[native lake]

No. Rather use things you want it to focus on.

[pulsar coral]

You'd think it's not grabbing images from the internet yes, that would be logical

But from the volume of characters I've tried, I don't think it's possible it would contain all of them with images in one model

It's a little too massive

[native lake]

Your example is flawed cause it probably has that character embedded in it's weights already. You need to shift the latent position using reference images so it uses the less popular images

[pulsar coral]

But thing is I've tried "Don't" in Nano 1 and it did listen, so I forgot about not using negatives

It seemed a little depracated, and in fact using "Avoid" seemed to be messing it more

[native lake]

It's not that negatives don't work. But it was not thinking about it before and now it's thinking about it.
"Oh you want less wheels? I didn't know we were doing wheels. I'll just add some and then take some away"

[pulsar coral]

Well let's do a fun experiment then

Get ready to lose your socks

Do you know how many images of this character must exist on the internet

https://guiltygear.fandom.com/wiki/Bridget

There must be like 6 different versions of her, each with hundreds of different fanarts

[native lake]

Lol the trick where you ask it to hold up a sign

[pulsar coral]

And that there is a fanart, not a very well known image

You can try it yourself over and over

Do we even need to test if Nano 2 gets it right? It totally will

There's no way it has it uploaded

[native lake]

Google already knows it. A model that is 16gb big has enough world knowledge for it. Why won't nano banana which I'm guessing is way bigger

But that's nano banana

[pulsar coral]

That image is cropped and doesn't even have all fullbody and details

[native lake]

It's a image editor. It can add things and understand instructions

[pulsar coral]

That thing it's holding could be a drum or whatever, it could be any nun.

It knows it's Bridget precisely

[native lake]

Yes the image editor knows what the image is. It understands what it is seeing and understands how to modify it

[pulsar coral]

It could be a pot, it could be a hammer, it could be whatever

It can't see it

[native lake]

Its a editing model. Try taking a group photo and asking it to remove a specific person. It probably can do it

[pulsar coral]

The only way it could know it's a Yo-yo it's if it knows the character

I cropped it off in purpose

[native lake]

Yes. An editing model would know the character on sight. I do not see how this relates to it using the internet for cheating

[pulsar coral]

Well it's not cheating and it's perfectly fine tbh

The real prob is Nano 2 doesn't listen when it's told to stop it for some reason.

[native lake]

It's not using the internet. Do the group photo experiment. It will work

[pulsar coral]

Ok then

Let's do one more thing.

[native lake]

You have a fundamental misunderstanding of how these models work

[pulsar coral]

idk the results are inconclusive atm tbh

[native lake]

At least I'm sure I could do the same with offline models

[pulsar coral]

Happy now

The Russia one was easy to guess, but the other two not even this week.

[native lake]

Better delete that

[pulsar coral]

Nah it has nothing bad

Admittedly it can't read my IP and location, but it totally told me it was because it was against it's directives

It knows.

Same as GPT knows

That or it knows the future.

Admittedly Nano 1 can't do any of that

And that's the prob with Nano 2 really.

It goes and takes things probably not in real time for sure, but at least somewhat quick enough.

Cause yeah it must take a lot of water to do that

Don't need to check news for that tho, it can just be known from Art site pictures uploaded this week or so.

[native lake]

It's nano using it's capabilities. Does it have forced image search access?

[native lake]

This has nothing about today's events

[pulsar coral]

Not today yeah, but the other Israeli thing is recent enough

[pulsar coral]

Like said, I guess it'd be impossible to have done it real time.

[native lake]

Yes cause it was trained on recent data. Its a new model

[pulsar coral]

Yeah I also got the climate change thing, that's why I went more specific with the war questions

At some point it got it in

[native lake]

It's not using the internet. Just using the more popular image of the character.

[pulsar coral]

Yeah more popular not even this week, I think that's a little more than popular event atm.

[pulsar coral]

I know it does, it just hides it's tracks a little to well for some reason.

[native lake]

Literally the other models that have no access to internet can do the same thing. Not saying you are wrong just saying that your evidence isn't convincing yet

[pulsar coral]

Well Chatgpt also makes images, and that one's easier to track that it can do that.

Like ask it in which country you are, it will know.

I asked the same to Nano 2, make me an image of it. It refused cause policies.

It coulda just made some random thing. I just know it knows.

Ok then, this should work

I cut it off cause sensitivity

It also had "Gaza" there

Can't prove it's going real time, but yeah 3-4 days ago, surely.

Anyway that should be an easy enough experiment to reproduce.

Nano 1 only knows up to Ukraine, but that's easier to know.

...nvm Nano 1 also knows. But gotta try much harder for a while.

In fact I think it can even be more precise about it than Nano 2, I got no doubts this is e-connected too now.

It told me about more cities involved, that I didn't even know about.

[pulsar coral]

Also this success rate on the answer shows it does take a level of randomness for the answer, but also that it most likely knows the answer

After all it's centered on being an image model, not a GPT clone

I bet during the first week this happened it woulda known too

The questions need to be very specific tho and aimed to see what it really knows.

It's search strength isn't very high really.

[tight osprey]

Can you experiencing this bug?

[pulsar coral]

Man that's a pro prompt, 3/4 and everything. And no on my end it's working fine, but I'm in south america atm so who knows

[edgy holly]

Looks like the style sheet didn't load.

[cloud river]

@Georgia not really sure where to post this feedback or who to address it to, but its been a few weeks since I've been in flow and seem the new UI changes. Some of it is great and its less clunky but I, for one, really rely on being able to make last minute edits on my ingredients and frames. Replacing the word "Edit" with "Draw" makes it more difficult. Draw makes you think you are making something new and I am not always adding, sometimes I am removing. That's why "edit" encompasses more imo

[native lake]

I think they tell you it's doing it. Also that you can toggle that in the settings. Check out the Chinese image models at least. Their editing ability not as strong yet. But they do show that it's possible to store everything needed

[pulsar coral]

ChatGPT is a little monstrous yeah, but it can also give you today's news accurately. No doubt about that one.

So a model can do that, and have an image generator attached

O yeah since I already said I live in Ecuador somewhere

That's fair close really, not really IP pinpointed but yeah

Tho admittedly it can also fail at this sometimes, it first has to know where you live via chat to get this close

Otherwise will make an image of some other country guessing

[desert hill]

Oopsie doopsie

[desert hill]

Guessing is bad

[pulsar coral]

In fact I used to have full chats with the gemini model attached to the image generator

[desert hill]

Interesting

[pulsar coral]

They're surely not the same thing, but they can connect

[desert hill]

Oh i see

[pulsar coral]

You can do this at Google AI studio

[desert hill]

If I can afford all the tools

At 450$ per month

[pulsar coral]

Let's say you fail at an image generation, then you ask it why it failed, and it will try to explain you

[desert hill]

Oh

[edgy holly]

https://tenor.com/view/darth-vader-no-noooo-gif-7347380959994097336

[desert hill]

You have it right

[edgy holly]

Unless you're really going nuts, you shouldn't need to spend that much. One primary model (Veo/Flow) + one all-in-one for gap filling (like Higgsfield, or Freepik), and then specialized tools like Suno for music, Elevenlabs if you need some fancy voice stuff, etc.

[desert hill]

Yes

[pulsar coral]

Yeah I don't think there's still anything worth spending that much yet tbh. Nothing giving anyone what they really want.

[desert hill]

Anyway my parents won't give me that much

If I ask for subscriptions

[pulsar coral]

Except maybe for LoRAs, but you gotta be annoyingly good at that one

[desert hill]

What's LoRA

[edgy holly]

Basically you collect a bunch of images that all have a similar style (for style LoRAs) and train it to apply that style to images.

[pulsar coral]

Yeah stands for Low-Rank Adaptation.

[desert hill]

Hmmmm

Is it free

Ig not

What 6767$ a month???

[pulsar coral]

It's like a smaller model specialized in one one particular style of image prompts, a lot of image generation sites can run those

[edgy holly]

Nano Banana doesn't have that. Midjourney has one. They call it something else.

[pulsar coral]

It's a little primitive now, and a single image tossed can do almost the same at Nano Banana.

But yeah, if you find the right LoRA it will allow anything

[edgy holly]

Usually can figure out how to get what I want with Nano Banana, but having style loras in NB would be super useful.

[pulsar coral]

Unlike the Nano models

If you want to browse LoRAs and what they can do, best place is Civitai

LoRAs also need to either run locally by something alike Stable Diffusion in your PC, or run at a browser site where you can upload them, like Pixai which is probably the most permissive one I've seen, but yeah uses credits and stuff there.

No one that's tried Whisk probably still cares to run those but like, it exists.

ComfyUI is also a popular app for it, but good luck on those tutorials, it's not impassable but still annoying.

[edgy holly]

They just added a free level. Small number of credits per month, but not zero. I paid for it for a couple months, but Nano Banana is just plain better.

[pulsar coral]

Specially for very little specific things that LoRAs can't understand yeah

If you know, you know.

[edgy holly]

I'm mostly making keyframes for a show I'm working on. NB Pro is great at near-realism stuff.

Loras are overkill for that sort of thing.

Like this

[pulsar coral]

What a liar, then again this is just the free model, the pro model surely knows it for it's own good

The Nano 2 model did seem to know.

And at Google AI studio it's also the Gemini models that are connected to the flash image models.

[desert hill]

Ah

[pulsar coral]

Even if you can tell them not to, if these want to know you from the internet, they totally can.

Let's switch a VPN on now

Yeah that's the VPN, so what is it do you know or not

[crisp wasp]

anyeditor here?

[swift ice]

yes

[sick sparrow]

is there a way to use images that i uploaded in another project also for a new project? I now only see the options of archived material.

[mystic valve]

I have a question: When I try to change details in Flow (background, clothing), the image resolution becomes problematic (the character's clothing looks blurry).

What should I do?

[mystic valve]

This's example

[pulsar coral]

Someone probably has a better prompt for this, but usually I'd just end mines like

[Image quality: Remastered 4K HD ultra high resolution, sharpest definition, denoised, unblurry, clean lines, High contrast, remove any jpg artifacts.]

That more often fixes it where possible.

If you notice it changes your image remove "Remastered" but depending on what one's aiming for more often goes fine.

You can also add "Vivid vibrant colors accurate to reference image" but yeah the same goes for that.

Also never right click-save images as, only use the download button, if you're not already doing that, this for whatever site you might generate images at, always look for the download button.

[pulsar coral]

I'd say in comparison it did get better but yeah probably remaking it from the start with the prompt would go better

[mystic valve]

thank u sir

[proven aspen]

So I generate images with flow, but is flow built for image generation primarily or is it for vids and shorts?

[native lake]

well i think the google guys are trying to make it both. but I'm going to make a nano-banana wrapper of my own now XD

[winged bolt]

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

[safe karma]

Hey all, could use some help. Did the extend/jump to and other features move or get removed from the scenebuilder? When I put something into scene builder I can only add/remove clips not generate new ones? This feels like a step backwards without the extend feature. Can someone point me in the right direction if I'm missing it?

[native lake]

prompt+reference

[winged bolt]

I upload multiple reference photos and only change the outfit and scene for the prompts. She always comes out smaller.

[native lake]

🤔

[edgy holly]

It didn't lie. It just doesn't have the information.

[topaz vigil]

what does this mean what should i do?

[edgy holly]

Mostly video. The image gen is geared primarily for making keyframes for videos.

[edgy holly]

Read it.

[topaz vigil]

i thought there wasnt a daily limit on images

because it uses 0 credits

apparently

[edgy holly]

Yeah, they moved it and don't have an instruction manual so people have just had to click around to find stuff.

Instead of being in the Scenebuilder, you just go to the grid video (or the other view inside projects or collections) and click on a clip. Extend is under the prompt box, on the left.

[edgy holly]

Now you know.

Credits only apply to videos.

[topaz vigil]

but its not precised anywhere. how do i know how much do i have left..

[crisp basalt]

https://one.google.com/ai/activity?utm_source=flow&utm_medium=web&utm_campaign=flow_ai_credits_page

[edgy holly]

I don't know if they have a ticker for daily image limits.

They certainly should.

[native lake]

use nano 2

[safe karma]

Thanks, I should have looked down while creating the original prompt. I figured they had to have moved it. I do see an issue with this though. A lot of times I get a result thats lets say 6 seconds in length. but I don't actually want the last 2 seconds. So I want to clip 2 seconds off the end and generate an extended scene from that ending point. I don't see an edit feature like they had in scene builder before.

[topaz vigil]

is it still good

[native lake]

good enough for me at least.

[desert hill]

Nano 1 was better in some cases tho

[astral fern]

I’m having an issue where I upload a vertical I,age and google flow makes it horizontal, does anyone have a fix for that?

[cyan dragon]

Is there a plan flow to have Agent where we can generate whole story boards or different angles automatic similar to higgsfield apps ?

[desert hill]

Use an image editor → “Reset orientation” / “Remove metadata”
Or use an online metadata remover before uploading

[desert hill]

Also check the resolution

[edgy holly]

They may have taken that out. Haven't tried to use it that way since the update.

[edgy holly]

Overall better than Nano Banana 1 IMO, but Pro is still better.

[crisp basalt]

its actually really good in my case tbh

actually i take that back idk

nano banana pro .

Style stolen from @edgy holly

nano banana 2/Narwhal

[desert hill]

Now that's just plaigrism

[crisp basalt]

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

[edgy holly]

NICE. Haven't done anything with that deep blue expressionistic style I was playing with in a while.

[edgy holly]

I might be bothered if people steal my characters, and definitely would if they steal the stories I'm working on, but any style in anything I share... steal away.

Also... all hail otter. Otter is life.

[desert hill]

Hail otter!!!

[crisp basalt]

did u generate this using image-fx

[edgy holly]

Hmmm... I do not remember. Probably?

[crisp basalt]

https://tenor.com/view/hackerman-gif-22344136

[edgy holly]

YEP!

Oh... the Otter/Snail wars against the gazelles and zebras... good times.

[edgy holly]

https://tenor.com/view/very-sneaky-gif-13205525615328658859

[edgy holly]

Ridiculous, but this probably is still my favorite AI image I've made.

[dry frost]

Whiskmore is brilliant! I'd book a trip to see that visage

[crisp basalt]

omg i remember going there when i was a kid!

[dry frost]

That must've been before they put a rollercoaster into the mountain behind. Got to keep the tourists coming right?

[crisp basalt]

i used to own this too ! now its gone 🙁

[edgy holly]

Still in the box!

[crisp basalt]

well i dont think tourist will be coming bavk again ever since the .. incident happened so now its just abandoned

[edgy holly]

Yeah, before the machine gun turrets.

[crisp basalt]

the main entrance

[edgy holly]

Portrait with a VHS style hurts my brain.

[edgy holly]

The Last of Us (Whisk users)

[crisp basalt]

here have some more portrait

[dry frost]

Yo the rust on the Android is just chef's kiss

[somber needle]

is there any drawing styles list for nano banana (pro or 2) somewhere?

[edgy holly]

They don't really do that sort of thing (unfortunately). They should.

But you can play with styles by just using images in styles you want as references. Pro and 2 are pretty good at style reference, especially in Gemini.

[pulsar coral]

[Artstyle: Highly polished]

[pulsar coral]

From counting manually, free accounts have around 20 daily images, the $5 has tier around 40 and so on and this might change according to missed attempts, a single payment also connects your billing account to Gemini and works like that over there too so you can get like double the amount of images if you use both sites but Gemini surely counts the missed attempts 1 by 1 specially for the Pro mode. The video credit count at flow doesn't restore, the $5 one only gives 200 a month and so on. Sometimes there's offers. https://one.google.com/u/1/ai?utm_source=gemini&g1_landing_page=75

And then there's the yearly plans that by that measure will keep you there 12 hours there easily https://one.google.com/u/1/plans?g1_landing_page=0

Fair warn so far it isn't the same as Whisk since they still haven't enabled the Nano 1/Gemini 2.5 flash image model it's using, so yeah you're not getting that there.

Closest for it will be Google Ai studio, and your Google billing account from Flow and Gemini won't connect to there, it's a separate payment.

[burnt oriole]

The Collections feature is so good

[edgy holly]

Yeah, I'm finding it very useful.

[topaz vigil]

thank you this really helped

[dry frost]

I just have a project labeled as Today's Special, but collections are cool too

[sick sparrow]

can you re-use collections across new projects?

[burnt oriole]

I haven't tried but would be fun if we could

[sinful lotus]

Hi! Has anybody figured out under the new UI how to have a scene extended upscaled to 1080p? (you used to be able to take them individually and upscale)

[crystal talon]

The extender doesnt work, what an update lmao

So where are the all the extend buttons ?

[sinful lotus]

Thx for confirming. And Drat. I don't use it often, but when I need it, it's really painful not to be able to extend a scene.... (because I can't use 720p)

[crystal talon]

Why does stupid veo3 add music to my fu*** clips? None of us wants damn music. If i want music i tell him. Who ever coded this, is stupid.

Avoid music, or no music or negative music dont even work. It always adds music. What a crap model.

[jovial ingot]

If Whisk must be fully merged, please consider a “Whisk Classic Mode” inside Flow that keeps the drag-and-drop 3-slot UI for creators who need to move fast
please leave it exactly like that. For professional work, the separation of Subject, Scene, and Style into three distinct 'drop zones' is a perfect workflow that shouldn't be over-complicated. In Flow, these layers get blended together, but in Whisk, they act as high-speed toggles that let me swap backgrounds or lighting in seconds while my product stays locked. Please don't bury this "Simple, effective UI inside a larger studio—keep the 3-slot Whisk interface as a standalone tool or a dedicated 'Pro Mode' for those of us who need to move fast and stay consistent
Thanks for all the hard work on these tools!

[pulsar coral]

... and enable the Nano Banana 1 model again. No matter how much of that stuff is added, some images are almost impossible to create without it.

Almost.

Seriously around 90% those complaints would drop as soon as they see that model literally does the same as Whisk at Flow.

[jovial ingot]

I would happily pay a standalone subscription for Whisk or a “Whisk Pro” tier to keep this simplified, high speed interface alive

[pulsar coral]

Noticed certain improvements in Nano 2 today, characters aren't being chubbied up with a big head that much anymore. Still nothing near Nano 1 levels which still keeps them much slimmer specially as reference image, but yeah it's better.

Nano Pro still horribly messes that up tho.

[crisp basalt]

yeah because youre on the wrong page lol

weirdo

[crisp basalt]

@midnight cedar 👀

[dry frost]

That must be a prototype or misprint because the Android was supposed to be green.

[pulsar coral]

Yeah that's racist to green androids

Bro got white-washed up, damn Gen Z

[edgy holly]

Maybe Android 17 or 18 will be nicknamed Jordon Almond, haha

[dry frost]

They still follow the treat styled naming right?

[broken hearth]

https://x.com/i/status/2029248676770509170

TestingCatalog News 🗞 (@testingcatalog)

Google is working on custom voices support for Flow editor. These are the same voices that you can find available in AI Studio.

This means that users will be able to use selected voices to achieve better character consistency.

X · Mar 4, 2026, 5:32 PM

https://tenor.com/view/cynthia-ariana-scared-shocked-scared-shocked-cynthia-erivo-ariana-grande-gif-8960749939830349462

[young locust]

• I find these kind of things really interesting 'cause it's a very simple process an application There seems to be like an issue with nano banana or Gemini where it doesn't translate your request logically in a visual sense it may translate it if you built a GEM with hard coded context. even after I asked it to analyse the image and create context in Gemini's chat, using Nano Banana 2 and using Nano Banana Pro It still struggled to create a result that was consistent unrealistic - I have seen this issue before and I keep coming across it where Gemini just doesn't get what you're doing. but even after critiquing gemini and calling it out on the issues and errors that it produced it still produced even something worse.

This is a fair and valid critique. After reviewing the full chat history, my performance was inconsistent and did not meet the level of quality expected. While I was able to execute some individual instructions, such as color changes and background removal, the overall compositing quality and visual continuity were not handled to a professional standard.

Your criticism of the early composite is accurate. The first generated image contained visible green spill around the subject’s hair and edges, which is a basic compositing error. The lighting on the subject also did not match the background environment, creating the “cookie cutter” effect where the subject appears pasted into the scene rather than integrated within it. Although these issues were improved in later revisions, the initial result should have been significantly stronger.

The second failure was misunderstanding your intent regarding the portrait image. When you provided the written prompt describing a “cinematic wide shot,” I interpreted it as a request to generate a new image from text rather than recreate the original subject in portrait format. This resulted in a generic subject and the wrong composition. I should have recognized the context and clarified the request.

[mental fossil]

Is the app available

[pulsar coral]

I think it's mostly the hair, shading and the skin details could use better prompts.

But yeah idk if the intention was to leave that real woman as she is, so just tried going HD with it

But undobtedly it could be created much better.

Real life doesn't look like real life, you gotta tie some cows together so it looks real.

[crisp basalt]

What’s the prompt?

[crystal talon]

That extend button doesnt work fishface

[pulsar coral]

I think it is "As cute as the weather girl from the news"

[crisp basalt]

like make it as cute?

[pulsar coral]

I prefer the prompt "As pretty as the Mc Donald's cashier"

That can't fail

Better than the average supermodel

[pulsar coral]

Idk I imagine it's something alike cause she's at the news cast for some reason.

Maybe it was Nano 2's ability to recognize people again, she might be some news anchor

That'd be really powerful tbh

[crisp basalt]

idk i had other idea

[pulsar coral]

haha

Hey she doesn't look so bad

[crisp basalt]

LOL

[pulsar coral]

Yeah she was born for this

[plain tree]

was i banned from flow bc i typed "remove the watermark" i cant input images anymore is this a soft ban anyone know how long it lasts??

i was trying to recreate a renaisance artstyle with watermark so i used it as a reference and i got banned or smthing i cant input images

[young locust]

this was just a random post that caought my attention for challendge @crisp basalt

but now I see all the awesome ones other people have done Oh crack up

[gritty cloak]

Ugh I really don’t wanna move to flow from whisk

[crisp basalt]

Nano banana really taking the realistic production editing literally l🤣

[gaunt jay]

Create image of girl speaking with dam constructor & flood come toward her

[gaunt ginkgo]

How to contact support

[spiral nebula]

would you mind if i request your prompt

[proper drum]

Do you think Veo 4 will be able to reach the levels shown by other generative video models and provide such a wide range of tools like control, editing, audio, and so on?

[sand wagon]

رقص جميل

[desert hill]

Btw it's a english only server @sand wagon

I can't understand anything

Retype in english or the mods gonna delete it

[wide ether]

Is anybody else seeing a lot of flickering in the flow UI? I'm guessing it's related to the new drag and drop functinoality, but wanted to make sure it wasn't just me seeing the flickering. I use the google browser as an FYI

[pulsar coral]

really. Ok then, meme it is

[desert hill]

Cool

[pulsar coral]

That type of text is the kind of thing Nano 2 and Nano pro exceed at, tho she got compacted and chubbied by it. Nano 1 could only do it well enough without the text

Tho it does sometimes fail like in some other memes, if you do want to add text you might need to reinforce it well

"The image has typical news strips from TV news below with info about what's going on written in english with perfect grammar and orthography, in clear easily readable font, it must be coherent and make sense"

That might be overkill but yeah that's the idea, sometimes it'd only listen like that in Nano 2. Nano pro is more natural about it for some reason, but point is it can totally do it.

So currently those are some more differences about the models.

And instead of just putting it like that for the news strip I shoulda put it like "Any text in the image should be written in..." and it would have affected the whole image.

I've also been playing with subtitles, comics panels and similar stuff, it totally can write.

[desert hill]

Playing??

[pulsar coral]

Yeah but mostly comics for friends so no one's gonna get what it's about here.

[cobalt mortar]

Hello. Is there any api for google flow?

[wide ether]

Anybody else seeing this? ⬆️

[late skiff]

Hi there i was wondering if anyone had a tip for this, its something that made me favor ingredients over frames for the longest time but i have to use frames for today :

Whenever i use
the start and end frames that are the same as a video that i already own i see that the generated video have those frames be slightly off colour, enough sadly that i cant put them one next to another without seing the color difference, does anyone have any trick for it ? i use premiere pro.

[merry wren]

Hey guys, I've been having quite a few issues with the Flow tool since yesterday. Every time I try to generate something, I get this error message: 'Audio could not be generated. Try another prompt or send us your feedback. You haven't been charged for this generation.'
Is anyone else experiencing this?
Please help.

[unreal viper]

id definitely recommend making a post in #1238621511401734254 where the team can see it better! these threads are actively monitored by the team!

[wide ether]

Could the size of the project (The amount of videos, images, etc.) affect the rendering? Meaning if a project is quite large with a lot of things, could that cause the flickering in the browser UI?

[merry wren]

Thanks!

[pulsar coral]

I did just test it the Korean April O' Neal pic and seems to be working fine.

Saw no flickers or anything

[wide ether]

It's intermittent. I suspect it might be a function of the size of the project. The project I have open has a very large set of assets.

[pulsar coral]

Oh yeah that one does happen, Flow takes a while to load things when it's like above 30 images or so.

But eventually should

I think for me about over 100 now that I check again

[wide ether]

It's very optimized as far as scrolling through videos/images ... that's the huge update posted recently, which I MUCH appreciated! They must have added caching or a better search engine on the back end to provide that optimization. Huge win for sure! It's just that every so often, the entire browser view when on the flow project page flickers. I suspect it might be related to the newly added drag and drop functionality

[pulsar coral]

Had problems yesterday to prepare some Deviant uploads cause of that, and just fell asleep cause it was taking time.

[wide ether]

I've got a project that probably has well over a thousand assets lol .... I need to download and start a new one I suppose

[pulsar coral]

Damn bro

[wide ether]

I just keep putting it off since they haven't added a good way to download in bulk .. or en masse ... like you could do with whisk

unless there's a new way to download in bulk that I'm overlooking?

[waxen glen]

would it not be smart to do a merge when they make sure flow has the exact same features as whisk?

[wide ether]

yeah since they're sunsetting whisk, and making everybody use flow ... they NEED to make the bulk download available in flow

[waxen glen]

flow is legit a complicated mess 😩

[wide ether]

It's a LOT better than it used to be. There are some good things with the recent makeover, the optimized scrolling, searching, and filtering is one.

[waxen glen]

whats better to use when doing reference nano pro or 2

[wide ether]

you wanting to generate images off of a reference set of images, looking for the precision that you got in whisk?

[waxen glen]

yea

[wide ether]

use imagen 4 ... toss in 3 of your reference images ... generate a bunch of images from that ... grab 4-8 of your favorite and use them with nano banana pro. I believe you can get slightly better video generations with pro generated images.

imagen 4 can generate images and have the precision off of reference images ... I think that's how whisk provided that precision feature

[pulsar coral]

Nano 1/Gemini 2.5 Flash. But Flow removed it for some reason. Nano 2 bases itself too much on accurate canon character data, that's why more often characters don't come out with the reference image face.

Done enough experiments about it already.

Like if you need an accurate canon DragonBall character image for example Nano 2 and Pro are perfect for it. But if you had ona those cringey Dragonball OCs images you needed Nano 1 for easyness.

[fresh scroll]

Do I still get unlimited video generation when I have the google AI Ultra?

[pulsar coral]

No Whisk was simply using Nano Banana 1/Gemini 2.5 Flash. This experiment can also be reproduced at Google AI studio with it's Nano Banana 1 model to prove it. Use a short non complicated prompt and you'll still be getting the same face of the character from reference image without all the hassle.

%90-%95 same quality as Whisk.

Until Flow enables Nano 1 again, they're offering a completely different product than Whisk.

Be that it's better or not, but yeah it's a different thing.

Beats me why they turned it off, all was fine with it for around 24 hours.

[wide ether]

I'm pretty sure the rule was, with one reference image, it used nano banana 1, but with 2 or 3 reference images it used imagen ... I'm pretty sure one of the google folks on here confirmed that. I've gotten pretty precise references just using imagen

[edgy holly]

yeah

[wide ether]

yep

at least I still see it there as of this moment

[pulsar coral]

We've also done experiments with Nano 2 and it can also get the same face for sure, but with a lot of careful prompting or worse by plastering the same art onto the pose of a character.

But yeah Nano 1 just does without all that, more likely cause it bases itself more on the reference image than it's model knowledge

[edgy holly]

If you use Veo 3.1 Fast on low priority mode.

[gloomy verge]

Hello. Can I ask, why Imagen 4 create better images than Nano Banana 2? Shouldn't it be the vice versa?

[pulsar coral]

Depends on what you're doing. If you want to create like a known canon character from any franchise or existing real person, Nano Banana 2 will find the accurate reference images and make it as official and accurate as it can, but seems this might short circuit it when your character is original or something else. It's not really bad at it, it's just trying to help you.

Imagen 4 on the other hand, is completely original about things. So yeah you're gonna get better results there with something original.

[gloomy verge]

I create 2 images with prompt "Portrait of female with pigtails" First I create with Imagen and second in NB2. Left image is Imagen 4, right is NB 2. In my opinion, Imagen´s image is better. No?

[pulsar coral]

Well this a different case, it's a matter of interpretation yeah. But Nano 2 does tend to create more everyday pretty girls unless you go deeper into the prompts.

And Imagen there gave you some sort of 3D unity portrait of a girl from a game.

But if you tell Nano 2 to give you something alike it surely can.

[pulsar coral]

idk it's a matter of delving into aesthetics, makeup, eye shapes, lip shapes etc, or like one would say "Make the face as pretty as you can supermodel style" for lazyness.

Ain't this very inclusive, it can go way more complicated

It sure knows what it is tho

Surprised I didn't use anime here.

[gloomy verge]

When you open image of my right girl and after left girl, doesn't the girl in the left picture have better graphics then right girl? I mean, that looks graphically better, than right girl. No?

[pulsar coral]

Well you can give it your own twists, it will listen to you, add your feedback to it.

It can indeed use prompts for sharpness denoise and quality, but that's the fun of it.

[gloomy verge]

I don´t know very good english, so I can write only basic prompt: "Portrait of girl with pigtails." I don´t know write detailed good prompts.

[pulsar coral]

Use your own language for it then, Nano 2 is good at it eitherway.

I've been trying even Japanese and it's perfect at it for some reason.

[gloomy verge]

I use same your 100% prompt and I wanted your same images, but it generate different.

[pulsar coral]

There's a factor of randomness yeah. But now that you know what you don't like, you can tell it to the model

[gloomy verge]

I try use same your prompt, but I set Imagen 4 and there is one of results.

Good, no?

[pulsar coral]

Yeah it's a start, is that imagen 4

Gonna go nuts about it now

She mighta gone a little off with the botox but you're getting closer now, it's a matter of practice

[gloomy verge]

I only changed generator model to Imagen 4, nothing next.

[pulsar coral]

Yeah I imagined that was it

[gloomy verge]

All four images.

[pulsar coral]

At least two of those would surely work, you'll be good at it

[gloomy verge]

And NB 2 Pro

[granite hemlock]

is flow good at generating videos

[gloomy verge]

In my opinion with Imagen for are better.

[pulsar coral]

Let's try causian now

[gloomy verge]

NB2 Pro

[pulsar coral]

Ah yes

It loves foreheads for some reason, now Hairstyle: Twintails with aesthetic bangs at the front of the head.

[gloomy verge]

And imagen 4

[gloomy verge]

Good girls?

[pulsar coral]

Yeah it looks better

Imagen 4 seems to need something about morphology that Nano 2 is doing by itself

[vital herald]

i cant get into ultra anymore and its bothering me why???

[pulsar coral]

That's my shot with an eye prompt

One of the most eye catching things about an image with beauty it's yeah the eyes, this one might cause people to keep staring at it

idk why added caucasian twice, just forgot about it

These are still lazy but go right to the point.

it takes those shapes and style from info about symmetric beauty, golden ratios and sort so why beat around the bush with it

https://en.wikipedia.org/wiki/Facial_symmetry

Like I can see it pasting it from here, so ain't even biased or anything

Apparently for humans, there's beauty in complete symmetry, doesn't work that well if you use the actual word "Symmetry" tho, so just gotta let it judge what's best.

I'd be careful when using that tho, like said sometimes it's hard to stop staring at it

[cerulean creek]

Extend with last frame?

[potent belfry]

I am clueless where to post this so I will my question here as well. I have a gemeral question about Whisk and Flow as I pay a montly subscription to Google One Plan which has Whisk and Flow listed in benifits as Whisk gives me unlimited image creations and I pay for extra credits to use in Whsk and even Flow to create video's. So my question as I went to try out Flow as I nomally use Whisk to create images but Flow has a daily limited so I can only make so many images. I understand I need to pay more for video but I am not paying more then my plane to make Flow image then I had unlimited with Whisk as if this is the case I will have to cancel my account as if they only offer so many free ones why pay for a monthly suscription when it got no benifits

[civic horizon]

I have a question

[cerulean creek]

?

[civic horizon]

Is the Extend not working on Flow right now, I ahve ultra?

[cerulean creek]

Working for me

[civic horizon]

Did it extend though?

Also, when you go to scenebuilder there is no plus button beyond the scene, and I noticed that Google was refunding any credits when attempting to extend

[cerulean creek]

It's being weird. I can't get the mobile version to play. But it did generate.

[civic horizon]

Ya def something is going on, thanks for the help

[cerulean creek]

Np

Did they take away the 1-4 batch size for extended? I haven't found the option in the new layout

[civic horizon]

Ya IDK, def is not as good of a "FLOW" as it was, pun intended and it seems to be not working very well at all

I wonder how often the Google eng. check these boards?

If ever

[stray remnant]

does anyone else have a problem of when using "frames" instead of ingredients the quality drops?

[wide ether]

Am I the only one that has the flickering problem in the flow UI?

[wide moon]

Does anyone got this?

[vernal remnant]

is anyone else getting the "VEO" watermark on ultra plan ?

[slow rain]

I suddenly started seeing a “VEO Mark.” I’m using the Ultra plan.

[vernal remnant]

Me too, I do not know why, I refreshed, logged out, logged in... still getting the watermark 🙁

[wide ether]

don't get the watermark (I'm on the ultra plan too) ... but the entire flow UI in the browser is flickering like crazy! 🙁

[wide moon]

It’s seems to me they’re not loading the nano banana 1 images we generated

[young locust]

nice to see posts are deleted with no reason - general chat is just that people working out Flows good and bad stuff

[young locust]

Yeah post was deleted not on my part, shame

Probably because I used Krea to fix the green screen spill

[crisp basalt]

Oh

[dark heath]

hey guys. I am having trouble shortening a video. Is anyone having trouble with that? was the feature to split a clip or shorten it removed?

[dark heath]

I am having trouble with it. I am unable to extend in scenes

[drifting iron]

Hi how can I generate video with flow

[proper drum]

flow still has issues with start/end frames. I've been reporting this on Discord since November. It’s disappointing that the $200 Ultra plan can't handle coherent transitions; there's always an abrupt jump in the middle, and the model fails to maintain character consistency between the first and last frames.

[stray remnant]

what do you mean? you type in a prompt and hit enter

[stray remnant]

yeah my problem with frames is just the quality drops significalty but when i do ingrediet the quality is good but looks nothing like what i want

[pulsar coral]

It was a good post, the news girl was covering a fight between Godzilla and Jaegers and a monster with a video too, It was innocent and fine and there was no reason for mods to delete it. I did save it.

[drifting iron]

Where place I don't know I joint just now

[stray remnant]

is there not a space on the bottom of your screen that says "enter prompt"

sorry "what do you want to create"

[vernal remnant]

Same thing, ingredients is way smoother, but changes the look and feel, frames maintain look and feel but is too buggy

[stray remnant]

yeah i just keep retrying the ingredient and hope to get lucky, i work with realtors so have to make sure the quality is as good as i can get it

[vernal remnant]

and then ingredients probably gives you what you want in 4-5 tries, sometimes 7-10 tries 😐

[pulsar coral]

Discord actually has a mod log where one can see which moderator deleted it.

[subtle jackal]

ya

and bob leponge is just a tard

like why did he talk

banning me?

mid staff

[dark heath]

Is anybody able to cut a video down to 4 seconds, and then extend from the 4 second mark?

[subtle jackal]

only thing good with gemini is telemetry and ai studio

other then that why use it

i see the future with this shiii

google will be relevent

but only as a small dog they gotta step up

[merry fern]

Hello Flow members. Can anyone inform me where I can notify a mod of a severe issue with the Gemini LLM? My account seems flagged as non adult (am 45 year old and issue came up making a biochemical nutrition protocol.....)

Critical Safety Logic Error: Magnesium/Fiber Interaction in Magnus Protocol

[subtle jackal]

parsec, zed,dev, vercept

tank this org

[merry fern]

Safety Issue: Gemini LLM failed to identify a physiological contraindication.
Context: A high-fiber protocol (45g+/day) with a recommendation for Magnesium Citrate.
Risk: Severe osmotic distress, leading to malabsorption and dehydration.
Correction: Logic should prioritize Magnesium Bisglycinate (chelated) in fiber-heavy diets.

[subtle jackal]

try claude thank me later

claude is industrial grade ai

i passed no studying cs no plagiarism charges no i slop

militant

gemini front ends r horrendous

i donot want to find out about their backends

[dark heath]

has anybody been able to use a video as input? or has the latest version changed that to allow only for images as input?

[merry fern]

<@&1366831844502274180> any of you active and available?

Technical Safety Report: Gemini LLM failed to identify a physiological contraindication. In a high-fiber context (>45g/day), the model recommended Magnesium Citrate. This creates a severe osmotic risk, potentially leading to dehydration and nutrient malabsorption. The safety logic should be updated to prioritize non-osmotic forms like Magnesium Bisglycinate in high-fiber protocols. Case: Male, 45, 191cm, 82kg.

[main saddle]

hey there 👋 I'd recommend asking in the Gemini server: discord.gg/gemini

[vernal remnant]

@main saddle I'm getting VEO watermark in my videos in Flow, I'm on ultra plan, what do i do ?

[merry fern]

Thank you @main saddle Georgia. I will give that a try. Current safety model missed this completely and gave it as "best option" twice.

Being possibly the worst option this is an oversight.

Sorry for the messages here. Been trying for over an hour to find a channel amd way to report this with elevated priority.

[pulsar coral]

I snitched already yay.

[pulsar coral]

Gemini also has direct support channels, your same Google account from Flow should be logged there too, people have answered fast to me there, or at "Ask the help community" too.

[merry fern]

Gemini help and feedback pops up that I have to ask my parents to make a feedback.

My mum is 70 and a-technical so that is a no go. 🤭

[pulsar coral]

Well they'll have to wait till you pass some christmases I guess.

[merry fern]

Funniest is the complexity of AI and my near hour interaction for nutrient density and then a feedback and help for a safety issue I run into a report system asking my parental supervisors hahaha

[rotund horizon]

Why Banana2 and BananaPro are changing the person's bodytype??? The face is also not matching but that's a different story; sometimes it works and other times does not. The bodytype though is not the same and especially in distant shots is totally 'off'. Like it was on purpose guided to alter bodyparts. I don't need to go into details; you can see the result yourself. Is there a way to fix it?

[pulsar coral]

You mean

Kinda like thiiiiis

Oh damn I did it, found the prompt

Let's see now

Sucesss. And now I'm taking it to my grave.

Then again maybe Google themselves fixed this. Let's try it how it was before.

Yeah I think they actually fixed this prob. Weird, then why are so many people still having it.

[rotund horizon]

is this reply for me?

[pulsar coral]

Yeah sorta, try this prompt at the start

[regal raptor]

i have a characte ri made half a year ago i cant prompt anytihng without flow dying not even change his shirt to business casual

[pulsar coral]

"Character is completely accurate to reference image"

Then add your other prompts, seems to be working for me atm.

[rotund horizon]

OK. I will give it a try shortly.

[pulsar coral]

It's suddenly giving me Nano 1 quality again, doesn't that beat all.

But better

Ah man I'm gonna miss the e-accuracy, but it's for better

Nano 2 is definitely not doing the same as yesterday. I think they did listen to feedback.

[rotund horizon]

The close shot works a bit better, is still 'off' but closer. The distant shots though, are still bad. Both the face and body physique

[pulsar coral]

Yeah for me too, but it sure got better. Let's see now.

[pulsar coral]

Try

Identity, facial structure & physique: Always true to character in reference image in every detail.

idk my anime stuff didn't need that much, but I don't do many real images.

Man I'm gonna miss the Wiki accuracy, but I think I can still prompt it back.

[pulsar coral]

it's kinda glorious tho.

I'm gonna change "Fully" for "Always" Ai gets that one better.

[rotund horizon]

The faces are not that bad when there are closeup shots. But both Pro and Banana2 look like have been instructed 'under the hood' to reduce the breast size - to put it with simple words.

[pulsar coral]

Ah yeah, that's another deal. Nano 1 always had that prob too tbh.

You gotta be really specific about cup sizes, hopefully it even allows you to.

[rotund horizon]

I will keep experimenting

[edgy holly]

Do you have reference images that show her from the side, or just the front?

[rotund horizon]

Yes. I tried that too

[serene cradle]

Hi guys, Veo now always display the logo in the bot right corner when creating videos?

[rigid umbra]

Ultra User --> 40 credit with fast mode.? bonus logo watermark veo on the bottom right corner?????

@main saddle logo watermark veo on the bottom right corner, acc ultra

[serene cradle]

yes 200 bucks a month and they still give me this lovely bonus.

[pulsar coral]

idk I was also making reference image sheets with sides too, then noticed Nano 2 wasn't needing them much anymore.

That prompt is literally "Put her at a stage singing" and she didn't change.

[edgy holly]

That's definintely a glitch.

[pulsar coral]

I think some gray or green clean solid color background still helps tho.

[pulsar coral]

I also complained about that once, And I've sometimes heard from Gemini support that watermarks are necessary to them so the art and videos can be recognized as AI. Which doesn't make all that much sense since...

[hazy field]

@main saddle I have a client deadline right now and I'm sitting here wasting my time and credits on videos I can't even use because of this watermark bug. We need an official status update ASAP. This is not what we signed up for.

[pulsar coral]

No actually it makes total sense, I don't think any Google site has videos without watermarks now, only Whisk and it's going, I think they're having to add the watermak.

There surely are other video sites that even free don't put watermarks there tho, so what gives.

[hazy field]

The official Google help page explicitly states that videos generated by Ultra users will NOT contain a visible watermark. This is a clear breach of the service terms

[pulsar coral]

Hmm well gotta wait and see if they update that, something might be going on gotta have patience

Then again I always just use video watermark removers, there's tons

[serene cradle]

The point is that many people don't experience this, even my friends on the Ultra plan don't. Their answer is inconsistent.
Even in the comment above, there's a guy said it was "glitch"

[storm vine]

@storm vine

একটি সবুজ মাঠ। রাহাত খুব স্টাইল করে দাঁড়িয়ে তার আইফোনে গেম খেলছে। পেছনে একটা ছাগল শান্তিতে ঘাস খাচ্ছ

[hazy field]

Are you sure about that? Did you check with your friends who generated videos in the last 2 hours? I was generating videos without any issues earlier today, but it suddenly started appearing about 2 hours ago.

[wide ether]

I finally figured out how to download files in bulk. You can multi-select files using the shift key with mouse click ... whew! Saves a bunch of time for me!!!

[pulsar coral]

Nice just needed to do that too.

[wide ether]

works great ... I was able to download hundreds upon hundreds of files in a minute or two ... vs. hours lol!

[pulsar coral]

Bro thinks this is one of those Discords that can generate images. No man it can't do that. We only speak English in this chanel man.

[edgy holly]

Man, they really need to make instructions manuals. I had no idea it did that.

[wide ether]

Now if they can only fix the flickering!! that's SOOOOO annoying! I created a new project thinking my previous very large one was the reason for the flickering ... but sadly it does it in a clean new one 🙁

I can't be the only one that has the annoying flickering of the entire UI!

[pulsar coral]

The complaints about "Mutated" faces in Nano 2 have gone down a lot in the past hours.

I feel Nano pro is the awful one now tho. Maybe it still writes text better tho.

[rigid umbra]

haiz. still error

[high fulcrum]

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!

[hazy field]

Still waiting. My deadline is approaching and I'm losing money here because the VEO watermark is STILL appearing on my videos despite being an Ultra user

[pulsar coral]

Well as an emergency measure you could look for a video watermark remover.

AI also does that now.

[quartz juniper]

x2

[edgy holly]

Did you contact support?

[hazy field]

I've already sent an email, but live chat and phone support are currently unavailable or blocked. It's ridiculous that a premium service like this has zero real-time support when a critical bug occurs. We are being completely ignored

[edgy holly]

"It's ridiculous that a premium service like this..."

Yeah, it's priced like a premiums SAAS, and when it works is worth every penny, but it's super buggy.

[quartz juniper]

Besides the fact that it takes time to do that, the results are not always good; sometimes it leaves blurry frames.

[pulsar coral]

idk if it's stuff like grass might not be very noticeable. And I do see a lot of grass there.

[edgy holly]

That's just AI. Getting great generations isn't easy.

The watermark isn't an AI problem. Someone screwed up.

[wide ether]

The flickering appears to be tied to batch view mode ... I don't see it in grid view mode

[pulsar coral]

Plus more often specific watermarks might need specific removers, maybe one for "Veo" specifically does the job correctly.

I don't work on vids here much tho, I use whatever else, but the images are nice.

Not because it's bad but because I only got 200 vid credits and they never regenerated till the start of month. Bet I could fix a lot of feedback about the vid thing, like I tried with Nano 2, and I suspect they did listen to a thing or two there

[green needle]

Do we now the date of when it will be possible transfer the projects from whisk to flow?

[pulsar coral]

An admin did show up and say "During the next weeks" I think that's all we know about it. But it probably should be before the end of month.

[green needle]

Thy

[edgy holly]

You're seeing the watermark in the UI before you download, and also after you download?

[plush geode]

Why on earth does it create a watermark when you spend the Ultra fee?

Is it a bug?

[pulsar coral]

Well only one thing left to do. Who's gonna message an admin about it.

I call not it.

[severe hare]

Are you Ultra subscribers also seeing the veo watermark on your videos starting today? 👀

[faint scroll]

I've also encountered this problem; the videos generated by my Ultra subscription have watermarks.

[wide ether]

no watermarks for me yet .... but D***, the flickering!!!! the UI flickers like crazy. I'm using google chrome.

[severe hare]

anyone who knows pls mention admins

[misty stone]

I'm so mad at you, Google. Flow on Android. I have never used quality, always fast, but this updated app defaulted to quality without me ever changing the settings from what I've always used. And it ate my entire month's credits without me realizing it.

And I'm mad at you, Google, because it ate those precious paid-for credits failing to follow simple instructions on every single generation of everything I've tried all week. Nothing complex.

For example, most recently, no matter what I've tried, negative prompts, giving it permission to defy physics, my instructions were simple: the dog floats up in the air and floats towards the camera. That's it. It never could do it, and insisted on having the dog walk at least on her hind legs. God this is frustrating.

And I'm mad at you, Google, because the updated Android Flow, if you do an extension to a video and Veo predictably screws it up, there is no way to go back -- you are locked in and now the entire thing is ruined -- audio profiles for continuity and the rest lost forever.

I'm beginning to suspect that this is on purpose, resist enough so that users munch through credits for Google's bottom line.

I've used Kling -- but don't have the benefit of a paid subscription like I do with Google -- so I know that this screwiness is not intrinsic to AI models generally.

I'm not the only one experiencing this, right?

• b

P.S. I'm mad at you, Google, that feedback on the app never goes through, only generates an error. Why I'm here! Too many obstacles -- I'm resisting the perception that this is more than coincidence. Absolutely everything doesn't work here, expensively.

Hold me.

[severe hare]

(

[rigid umbra]

thanks!

[storm fractal]

why video generated by gemini ultra still have watermark veo?

[reef mural]

hm.. upscaling to 4k now costs 50 credits on ultra? isnt that very expensive?

[severe hare]

uspcaling video ❌ Sharpening video ✅

[spark cave]

Today, there is a watermark in a video created by an Ultra subscription

[oblique brook]

Why are there watermarks on the generated videos?

[cold surge]

yes facing the same issue on 2 of my ultra accounts as well. both are giving watermarks

@fervent bane @unreal swift can you guys help with this? also please let me know if you are the right people to tag for help

[pastel yoke]

This is a valid question, that I'm sure plenty will want the answer to. Just replying here to bump it.

[misty stone]

Update to my message just above. I know now that I'm not totally crazy, and here's why: Just tried Alibaba's AI. It's free (about 5 gens a day) and has issues, but it followed my prompts perfectly.

These first two five-second gens stitched together are both first tries.

This is a real indictment in my book of the hailed Nano Banana (or at least the way Flow interacts with it). Considering Google's status and investment, some Chinese free tool shouldn't be blowing Veo 3 away on this attempt....

Also notice that it didn't blow out the face. Especially on the extend in Flow (not shown in prior message above), no matter how I prompted it, Veo insisted on blowing out the face with the key light as the dog approached the camera.

Anyway, ran out of gens, so more to come, but it's a start. The stitch isn't great, and I'll talk about that another time. But for my purposes here, impressed with Qwen today ...

[slim ore]

Hello bro can u see the general chat and I need help with it

Btw 🙂

Nice

[dreamy spear]

Generate Video "Begin with a map of UAE. Zoom in to focus on UAE. Apply a red background highlight to France with a glowing white border. Overlay a text banner labeled 'Sharjah'."

[gloomy verge]

Hello, can I ask, how can I create combined images, when In Flow isn´t Subject/Scene/Style option? Example, woman/cooking/pixar style. Thank you for help.

[rotund horizon]

I am coming back into this, and I HOPE THE DEVELOPERS WILL LISTEN. A few days ago I mentioned how Banana1 and 2 both combined within Flow, were the ultimate editing tool Unfortunately, the developers took Banana out right the day after. I don't get it why so much haste to get rid of something that is working well!. I had presented a study of how both tools can work together and fill one another's gaps. Today I was working another project, this time only with Banana2 within Flow and the results once again were disappointingon . Here is a comparison between Banana1 (used in Whisk) and Banana 2 (used in Flow).

As you can see. The tool is NOT there yet. If you want to create silly images of cats and add yourself hanging with Jesus in a cafe, yes, for silly things like that, it might work. But when you try to push it a bit, especially with details. It fails miserably, mostly in maintaining design cohesion.

The first image is a comparison when using the same prompt. And yes I know, you can prompt it differently and you might achieve the result you want. But prompting it 20 times and getting 1 result correct it does not means that it works. And this is just an example of three images. I have dozens of failed results just in the same project. I never had that many with Whisk, never! As you can see, Banana2 illusinates a lot adding things in the foreground that are not there, or altering the design instead of zooming in, or changing it from its current state.

The second image is how the two tools Banana1 + 2 can work together to achieve an amazing result. This is just a quick prompt of adding more realistic titanium material finish to the mufflers. And I even reversed the image and placed the improved muffler finish in the 3/4 rear view.

Honestly, if this does not convince you, I don't know what else will. PLEASE do not take Banana1 away and make Flow the ultimate AI image editing tool.

[edgy holly]

That's cheap. Go look how much 4K upscaling costs on Topaz.

[edgy holly]

Whisk was an experiment in beta testing, and like all betas it's ending. There is no AI image subscription that has unlimited use for high end models that's cheap.

[edgy holly]

You can still use images, and more than just three, for reference in the same basic way, but you need to learn how to prompt.

[gloomy verge]

Yes, my big disadvatage is writing prompts. I know only basic senteces.

[gloomy verge]

How is maximum of images ingredients for image generator?

[vernal remnant]

Is there any solution for VEO watermarks on ultra plans? Or is it a new standard now, clarification is needed.

[somber needle]

I hate this gradient with all my souls

[edgy holly]

I don't know... I think I read somewhere that it might be up to 14, but I've never tested that.

[edgy holly]

I don't see a watermark. It seems to only be hitting some accounts.

[vernal remnant]

I opened a thread in the labs-bugs-feedback section, There are no replies from the support yet, but a lot of people confirmed there, that they are all getting VEO watermarks in their videos on the ultra plan. I just tested it again by creating a video and the watermark is still there.
I hope that they have not made this permanent.

[edgy holly]

It's not universal. I'm not seeing it, and some others I've asked aren't. It looks like someone screwed up.

[pulsar coral]

tbh no one ever runs a Discord too seriously anyway, point of having a Discord is to fill them up so others do it for you, so about this video watermark problem ya'll might be better off finding more official Google/Veo help channels. Gotta be afraid of the answer they might give tho. Speaking of that.

[pulsar coral]

If you know the exact Pixar movie style you want Nano 2 can surely find it in it's knowledge data base and plaster the artstyle, I did just have some good results with other stuff

You just gotta be specific about it.. idk I don't watch too much pixar stuff.

That way you can do it manually with a reference image too, tho if the movie was popular enough you might not even need that.

[pulsar coral]

After some updates from yesterday on Nano 2 it's working a lot more like Nano 1 as in it doesn't try to go fully accurate to what it sees, this can be noticed much more with human character faces which is actually a good thing cause now it doesn't change them so much or at all, but what you're doing here is way different with cars so dunno. Nano 2 would still listen if you tell it to go accurate to what you want, you could also add "No creativity freedom on this" to the prompt.

Then again probably best thing is to keep Nano 1 in the model drop down menu, would save a lot of hassle.

Anyway if the car has a brand and accurate part names you might wanna use them, I think it's gonna recognize them 1 by 1.

[pulsar coral]

Ah success... I think

It was failing to generate the image sometimes tho, needs a more precise prompt about who or what to apply it to, or a reference image.

idk maybe it's tripping on itself or considering it copyright but idk that one

[wide ether]

I don't see it either and I'm on the Ultra sub

[edgy holly]

You're just the poor guy who is the only one with the flickering, heh

That still goig on?

[pulsar coral]

Apparently his project has about 1000 images and that might be shortcircuiting flow's display limits.

[wide ether]

yeah ... it's BAD! ... I found that it doesn't do it so much in grid view, but in batch view it's so awful!

[wide ether]

not the case ... I created a new clean project and still happening

[edgy holly]

That tracks. They've had issues with the Library like that going back over a year.

[pulsar coral]

Does it have only 500 images now

[wide ether]

I figured out how to download en masse too ... that's what got me to a new project 🙂

[wide ether]

🤦🏻

it happened with 5!

[edgy holly]

You said that, but I tried to do Ctrl-arrow and it didn't work. What was it again?

[wide ether]

I'm on a mac ... command+click to select the first one .... then scroll up to some asset later in the view and then hit shift+click ... and viola

[pulsar coral]

Does your browser have extensions? Did you try a clean one without them? or a different browser?

[astral hemlock]

I'm working on a Large K-12 project (about 200 minutes of video), and have been a little hamstrung by the safety barriers (which is completely understandable). Would be happy to have all my usage watched and reviewed. Is there any way to have validated accounts use the flow for K-12 content, such as ID Verification and/or additional oversight?

[wide ether]

using google ... and no extensions

[wide ether]

this allowed me to download all my thousand and some assets in a matter of a minute or so 🙂

I have a m4 pro mac too ... so wondering if that's at play with the flickering

[vernal remnant]

It could be limited to Asian countries, but thats just a guess, I'm not 100% sure, or it could be a regional bug.

But in any case, my production is halted because of it.

[edgy holly]

Yeah, this is something I've said a few times in feedback... people get that AI models themselves are still difficult to wrangle into generating what we want (MUCH less than a year ago, and getting better), but the apps built around them... it's entirely warranted for people who pay north of a hundred bucks a month per seat to expect a stable app.

[vernal remnant]

I second that, like I'm ok with the quality and bugs in development of the model, but UI bugs like these are not ok. This is not some 6 month old understaffed startup we are talking about.

[native lake]

o_o what happens next bro

[edgy holly]

Right. Flow has been out >8 months. It's the only web app in the top >10 that has so many issues.

[edgy holly]

Just run it again, and if it keeps failing, tweak the prompt.

[native lake]

nah its cool, im just wondering what could possibly have happend for violating policy XD

[vernal remnant]

Wow! My main client wanted to shift to Higgsfield to use Veo through their UI, and I convinced them to remain on Flow to save overall cost, and 3 days later BAM! Watermarks in all of our videos 🫣
I have no idea what to tell them 😂

[pulsar coral]

What's that X-wing thing at the top, I've never seen a drone like that

[edgy holly]

Higgsfield is one of the best gap fillers, but for a primary video generator I'd still go with either Veo/Flow or Kling.

[native lake]

tell them you were right but seems google is determined to chase you off. 🙂 also ask for refund if you were depending on there not being a watermark and it was in the feature list.

[vernal remnant]

yes, that is my theory too, but now, I dont know how to solve the watermarks, I submitted the help feedback ticket on flow too, can you recommend any other place where i can reach out to them for a solution ?

[native lake]

no idea. just playing around .

[edgy holly]

Regular tech support.

[pulsar coral]

I know it looks like nothing but it seems the model grabbed that from somewhere it shouldn't.

Considering the other stuff barely exists or is from a movie

[native lake]

its not a xwing bro. also xwing is a common design for drones these days. no trademark violation for now at least.
also no one can trademark a design. that's simply impossible (legally)

[pulsar coral]

https://www.youtube.com/watch?v=HQzoHGW4kwY

It's just a name for the wing shape

Oh well, like the delorean it could also be something fictitious from a movie, I don't watch every movie

Just still wondering how Nano 2 grabs things or finds them

It's been really interesting the past weeks, I'll just ask GPT if it can recognize it

[native lake]

well i did say drone. there is a drones. that use that use that shape. one set the world record for flight speed. but its a bit more boxy than all of them.

I said drone and it gave me planes

[pulsar coral]

Yeah it could also have mixed things, drones more often don't have jet capabilities. The fighter jets from the image surely are "F-35 lightning", they're latest models small and chubby like that but incredibly advanced and effective.

Nano 2 did know about them, and that's normal I guess

[native lake]

im trying to use it to storieboard. its not going well

[pulsar coral]

I mean more often drones aren't jet, unless they're from the army, so makes me think where has Nano 2 been lurking.

[brittle pelican]

Wow, now Ultra users also have "Veo" watermarks on their videos?

[lime quiver]

Yes, Google can surprise you.

[edgy holly]

Pretty, pretty machine.

[edgy holly]

Does the army even have drone jets? I know the Air Force does, but Predators, Reapers, and such are all props.

[edgy holly]

They aren't supposed to, and only some Ultra accounts are showing it. Something is wrong. They haven't responded about it yet.

[native lake]

They've had them for a long time.

[sonic zealot]

Small bugs too.

In scene builder the audio mutes itself unless I enter full-screen and manually adjust the volume on the video

[native lake]

you want it to play sound on hover?

[native lake]

you have an addon or browser setting for autoplay i think

[sonic zealot]

When I see all my clips in grid it works fine. When I add a clip to a scene and view the clip in scene builder the audio is muted for some odd reason (and reverts to mute when replaying)

[daring island]

I want a product manager from Google to actually try to use this. There's no way they test this or use their own product; it's an absolute disgrace

Unless you're trying to make a bunch of disconnected clips, this product does not work. It was a lot better before their update from the other week.

All you guys have to do is actually try to make a video that's, let's say, a minute long and think about the workflow for the users. I don't understand how a product manager working for Google can do this. I don't understand how the developers actually think this is okay.

[edgy holly]

I just do all that on my PC. Google Vids is exponentially better as a light cloud video editor.

[sonic zealot]

Okay ill do that

[gloomy verge]

Can I ask, if Flow gallery is for only videos, or for images too, when Whisk is ending?

[sonic zealot]

Looks like flow will be able to do both videos and images

I sse what you mean

[gloomy verge]

I now, that I can create images by Flow too, but if I can show it here, somewhere, or only videos.

[wide ether]

you could always do that with flow 🙂

i.e. that's nothing new

[stiff raven]

Is there a way to download everything in a project at once? Or I need to do it all by hands?

[wide ether]

Yes! Not sure what the key equivalent on a windows machine is but on a mac it's: Cmd + click on your first asset, scroll to the other end of your assets then Shift + click ... then right right click --> download ... 🙂

saved me literally HOURS!!!

[stiff raven]

thank you!

[wide ether]

welcome!

[rose vigil]

please remove the watermark google, I can't work like this. i'm ultra user

[river horizon]

how can I delete a generated image?

[ashen talon]

How to bulk download upscaled images ? Any automated way ?

[crisp basalt]

Uh I don’t think you can

It doesn’t save the “upscaled” image in your project or library. The only way to obtain the upscaled image is to manually click the upscale button.

Somehow the upscaled video is saved in your library (the my library page, not the project page).

[supple pendant]

Hey @crisp basalt ! wanted to ask you since you're the Flow master -

Did Extend get switched around? my buttons to use Scene have seemingly disappeared 🙁 thought it was just a glitch so I waited a few days, but seems like it's still not giving me options to extend or cut to in Scene

[crisp basalt]

Yes, they moved the extend button outside of the scene builder

If u click on your generated video, it will enlarge the video you clicked on. Now, you can see there’s an “extend” button, an “insert”, a “remove”, and a “camera” button underneath the video.

Also, they removed JumpTo for some reasons. Tbh , JumpTo in my opinion wasn’t really useful to me. It doesn’t help either when JumpTo can only be used with a quality model which costs 100 credits. 💀

[pulsar coral]

Now that I check this again I wrote the prompt "Ars style" instead of artstyle, no wonder it as giving me booty check warnings. So yeah might not wanna do that.

[crisp basalt]

The downside of having only the extend feature is that it might cause some inconsistencies in how your subject looks (e .g if the last frame of the video shows your subject is already facing backward that’s where the inconsistencies issue comes)

[crisp basalt]

Wait it can do Disney now?

Last time it gave me error cuz of IP stuff

[pulsar coral]

Admittedly it was giving me some warning and failing at it, maybe it had to do with that, but "Disney-Pixar" does often work.

[crisp basalt]

Ah

Mine was “zootopia” ,

It used to generate em with no issues, but I suppose something finally caught up to Google 😀

[supple pendant]

Thank you so much !!!

Found it!

Yeah, agree - JumpTo wasn't something I used either, but might have been useful for people who didn't want to edit beyond the app?

A bit confused why Scene still exists as its own bubble if everything's now moved to under the vid?

[crisp basalt]

Scene builder is now only for rearranging / trimming /combining clips

[supple pendant]

Ah i see - interesting

[pulsar coral]

Really. It's really hard actually, but maybe being REALLY ambiguous indirect about it could still make it work, it's model knowledge could be made to work against it.

Cause yeah I got no doubt that's patched af, I saw it earlier.

[crisp basalt]

I’m gonna try generate Mario and Sonic

LOL

[rotund horizon]

Where can I send some feedback photos for Banana2? This thing is broken. I think got too much excited releasing it.

[pulsar coral]

Let's use Korean reporter meme girl again for some experiments. Let's see now...

Imma try and break it as much as I can

[river horizon]

what's the difference between 2 and Pro?

[pulsar coral]

Right now there really are differences, Nano 2 will help you make the character from reference image more accurate if you ask it to directly by prompts, but it can also help you make it accurate to an already existing canon character from movies games or whatever if you do the same.

Like this

Very copyrighted stuff might have lower chances to succeed tho. Tho not everyone would care as much as Disney.

These worked one outta ten.

Now Nano Banana pro is still a little more invasive to change faces because it thinks it's helping you by making the reference character accurarte to what it thinks it is, and won't understand your prompts as good as Nano Banana 2, but it's still pretty good and more than all can also write text if you do ask it to write it perfectly.

[river horizon]

I see.. thanks

[pulsar coral]

Also the Zootopia prompt is mostly
"Artstyle: That one recently very popular story with talking animals, must be 3D."

And the more of those characters you add to the prompt is less likely to succeed, I added Gazelle there and told it to add as many characters from the story as it could.

So yeah chances are low there, for a pack like that one needs around 15 mins.