flow-general・🌊

[severe lily]

The past week has had abysmal reliability.

[full blade]

Why is it now saying you're requesting things to quickly on Nano-Banana 2?

[violet ruin]

Please help me with this I can’t seem to get the camera behaviour for handout as I prompt to add in I tried our AI models and it’s slow comes out of static and I’m using VO3 as my model and I’m staying with it can anyone just help me with a prompt which I can give it

[severe lily]

Love it

[full blade]

Well, I don't love it at all, it's really annoying

[glad yacht]

A direct feedback of the Flow Mobile app:

[violet ruin]

Can anyone please help me how I can be able to get an output of a camera behaviour handheld shot because every time when I put it on VEO3.1 I always get a static shot as an output even though I give it tell it that it’s a handheld and also negative prompts to it

[pulsar coral]

Maybe you can post the video and we could see what's wrong and what to prompt, surely much easier than me alone cause other people might have better tricks

[violet ruin]

Say less

[pulsar coral]

Post video. We see how fix. Unga.

[violet ruin]

Give me a second

Hello everyone I’m just struggling to make this camera a few a lot more of a handout shake can someone please help me I’ve been constantly trying iterating and I’ve been wasting credits I basically wanted to be hand out there’s like shaking you know that feels natural it can’t be static like it’s a tripod

[pulsar coral]

I understand the prob now, think you can share the prompt, haven't ever done something like that but it's probably doable

[violet ruin]

Ok give me a second

[pulsar coral]

Cause sometimes it doesn't matter how hard one tries to make the video do something, the rest of the prompt could be interferring

[violet ruin]

🎬 VEO 3.1 (FAST) PROMPT — CLIP 1: HOOK (CAR VERSION)

SUBJECT: Sarah | AUDIENCE: Black women 25-45 | NICHE: Skincare / Dark Spot Correction

[VERTICAL 9:16, DURATION: 8 seconds]

---

🔹 IDENTITY ANCHOR (UNCHANGED)

Young Black woman, early-to-mid 30s. Medium brown skin (#A67C5B), symmetrical face, pronounced cheekbones, full lips, warm brown eyes (#5C4033), long wavy black hair (#1A1A1A) with natural flyaways. Navy sweater (#0A2342) w/ royal blue stripes (#4169E1) over white collared shirt (#F8F8FF). Delicate gold jewelry. Minimal natural makeup, visible pores, authentic hyperpigmentation.

---

🔹 SETTING

Driver seat of luxury vehicle — PARKED, NOT MOVING. Red leather seat right (#8B1A1A). Black interior trim (#0A0A0A). Driver window left: bright diffused daylight, green foliage (#2D5016) visible. STATIC background — no motion outside window.

---

🔹 CAMERA BEHAVIOR (NEW — ADAPTED FOR PARKED CAR)

Front-facing iPhone selfie video recorded by subject herself. PARKED, NOT MOVING vehicle. Authentic handheld smartphone behavior throughout entire clip.

Continuous subtle micro-shake caused by natural one-handed grip instability. Tiny unpredictable vertical micro-tremors from natural hand fatigue. Slight rotational sway and imperfect framing corrections as she casually holds phone while talking. Natural arm fatigue drift and soft body movement transferring into camera. Frame never perfectly locked. Minor handheld jitter persists through entire shot. Real smartphone motion inertia with tiny delayed movements during hand adjustments. Imperfect selfie framing with natural human instability.

NOT stabilized. NOT tripod mounted. NOT gimbal stabilized. NO cinematic floating camera. NO smooth tracking. NO perfectly centered framing. NO robotic steadiness. NO artificial camera glide. Motion should feel like a real person casually recording themselves in a PARKED car using iPhone front camera.

[pulsar coral]

There we go, something here could be interferring, this looks like a gemini prompt

Lemme read a bit and we'll see

[violet ruin]

---

🔹 SHOT AESTHETIC (UNCHANGED)

Candid, unedited iPhone look. Slight motion blur on gestures. No polish, no grading, no filters. "Shot 5 minutes ago" vibe.

---

🔹 LIGHTING (UNCHANGED)

Bright diffused daylight through driver window (~10:30 AM north-facing), cool-neutral (5200K). Crushed blacks in interior shadows. Authentic phone dynamic range.

---

🔹 DELIVERY PROFILE (SIMPLIFIED)

• Tone: Direct, calm, authoritative — like giving serious advice to a friend
• Voice: Lower pitch than normal conversation, resonant, no upward inflection
• Pacing: Slower than normal speech — pause a full beat after "melanin", after "tips", after "today"
• Facial expression: NEUTRAL throughout — no smiling, relaxed jaw, neutral brows, direct eye contact
• Key emphasis: Lower pitch + slight volume boost on "melanin", "budget-friendly", "dark spots"

---

🔹 DIALOGUE (SIMPLIFIED + CONCRETE PACING)

0:00:00 "Invest in your melanin." [Pause 1 full second after. Lower pitch on "melanin". Neutral face. Direct eye contact.]
0:00:02 "Here's five budget-friendly tips" [Pause 1 full second after "tips". Slight forward lean. Hand enters naturally.]
0:00:04 "you can start implementing today" [Pause 1 full second after "today". Blink naturally. Hand gesture syncs.]
0:00:06 "if you wanted to start fixing your dark spots." [Softer volume on "dark spots". Relaxed jaw. Hand exits naturally.]

→ Total words: 19. At slowed pace with pauses = fits 8 seconds.

---

🔹 AUDIO (SIMPLIFIED)

iPhone 15 Pro built-in front mic. Quiet car interior ambient. Dry phone mic sound: proximity warmth, mild compression. Natural breath sounds. No cleanup. No music. No engine noise (car parked).

---

🔹 REALISM CONSTRAINTS (PRIORITIZED)

❌ NO smiling, NO text overlays, NO product in frame, NO moving car, NO external mic, NO cinematic grading
❌ NO stabilization, NO tripod, NO gimbal, NO smooth tracking, NO robotic steadiness
✅ REQUIRED: Visible handheld micro-shake, neutral expression, slowed pacing with pauses, lowered pitch, raw iPhone audio, natural arm-fatigue tremor

---

🔹 RENDER NOTES

• Clip 1 of 4 | Parked car | iPhone mic | Neutral face | Handheld micro-shake | Slowed pacing
• If shake is still subtle: add "MORE visible grip tremor" to camera line
• If pacing is still fast: add "SLOWER — like speaking to someone who needs to remember every word"
• CAR IS PARKED — no road vibration, no suspension movement, no bumps

Here it so sorry for like this I don’t the subscription plan

[pulsar coral]

Man no wonder it's tripping on itself, but probably just encasing and simplifying the camera/composition slot would fix it, most the rest looks fine

pretty sure this one here's conflicting

🔹 SETTING
Driver seat of luxury vehicle — PARKED, NOT MOVING. Red leather seat right (#8B1A1A). Black interior trim (#0A0A0A). Driver window left: bright diffused daylight, green foliage (#2D5016) visible. STATIC background — no motion outside window.

[violet ruin]

Seriously

[pulsar coral]

Remove "Not moving" and only leave "Parked car"

[violet ruin]

I been irritating myself lately 4 hours

[pulsar coral]

Lemme see how to fix the camera slot

[violet ruin]

Why but ok, I lose hope

[violet ruin]

🙏 what’s the solution

[pulsar coral]

You got two "not moving" instructions there, gotta remove them

[violet ruin]

I never thought like that

[pulsar coral]

"Parked" is surely enough for the prompt to understand the car isn't moving, might just wanna tru it after removing all that says "not moving" but I'm still gonna try something else.

[violet ruin]

Yes I’m here

[pulsar coral]

Any AI be it chatting, videos or images is very sensitive to conflicts

[violet ruin]

What do you mean

[pulsar coral]

For example what we're doing, you can't tell it "Not moving" if you want things to move. It's like saying "go right" and "Go left" at the same time, the character would snap to both sides

In this case it's probably better to say something like

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

I don't think you're using reference tho, but it would understand it the same way

[violet ruin]

Brother I have a lot to learn. What you saying a facts that I never open to realise. Seriously. So right now I should remove the word “not moving”

[pulsar coral]

Yes pretty much, seeing how the vid went I think it will understand by itself. Look for everything that says "not moving" and take it out. "Parked" seems like it'd be enough

[violet ruin]

Okay I will remove it. Can I do it and give you feedback

[pulsar coral]

Sure

[violet ruin]

Okay give me a minute for me to redo and load the video

[violet ruin]

It still doesn’t work. It’s it’s showing me a transition which I don’t want the transition

[pulsar coral]

"Static background" is also a contradiction, gotta remove that. Parking is handling it too

[violet ruin]

Flip okay I’m wasting credits. I didn’t remove that

Handheld camera. The camera moves with organic human jitters. Slight instability.
Sarah. Black woman, early 30s. Medium brown skin (#A67C5B), symmetrical face, pronounced cheekbones, full lips, warm brown eyes (#5C4033), long wavy black hair (#1A1A1A) with natural flyaways. Navy sweater (#0A2342) with royal blue stripes (#4169E1) over white collared shirt (#F8F8FF). Delicate gold jewelry. Minimal natural makeup, visible pores, authentic hyperpigmentation.
Driver seat of luxury vehicle — PARKED. Red leather seat (#8B1A1A). Black interior trim (#0A0A0A). Bright diffused daylight through driver window, green foliage visible outside. Static background.
She records herself on iPhone front camera, one hand, arm slightly extended. Her grip is unstable. The phone visibly trembles from natural hand fatigue. Constant micro-tremors ripple through the frame — tiny unpredictable vertical and rotational shakes from a real human hand. The frame drifts, overcorrects, drifts again. She adjusts her grip mid-sentence and the whole frame shifts. Shake is visible and persistent throughout every single second.
She looks directly into camera. Serious. No smile. Like telling a friend something important.
She says slowly and deliberately:
“Invest in your melanin. Here’s five budget-friendly tips you can start implementing today — if you wanted to start fixing your dark spots.”
Bright diffused daylight from left window, cool-neutral, 5200K. Crushed blacks in interior shadows. Raw iPhone audio. Quiet car interior. No music. Natural breath. No color grading. No filters. Vertical 9:16. 8 seconds.

[violet ruin]

Are you still there

[silk brook]

me same

[pulsar coral]

Yeah just running some tests too, it'll take time

[pulsar coral]

I'm getting close, this whole prompt is huge and needs a redesign

I don't think most of what it had is needed, except for the character and voice

Weird that gemini doesn't understand itself... wonder how much is VEO really connected to it

[violet ruin]

You’re honestly extremely close like this is a great prototype and how did you do it what was your prompt and I also have no problem I’m trying to work on it as well

[pulsar coral]

Told it in every encase to be shaky, gonna get rid of those weird transition cuts

Yeah this can be done, how's this

Amazingly while looking for a fix noticed a lot of people need this kinda thing

Actually yeah you're on the right track starting your prompt by

"Handheld camera. The camera moves with organic human jitters. Slight instability."

[violet ruin]

I’m going to send you a reference clip

[pulsar coral]

Yeah I see the vid reference now, it does start looking like this, gotta make it more shaky

[violet ruin]

Yeah that’s what I was thinking I was gonna try to like what you did and then I tried to just iterate to make it more shaky I like Claude or ChatGPT and to make it more shaky.

[pulsar coral]

Sometimes will do it sometimes not, but more often is better to start by telling it to shake, prompts at the top do have more priority

[violet ruin]

Oh my God, this is what I’ve been struggling you’re such a good friend nah for real 🙏

[pulsar coral]

...must've made like 12 tests, funny enough it never added music. Dunno if it's something in the prompts or they just fixed it.

Nah who am I kidding, of course it must've been the prompts. Fixing, good joke.

Must've been something here

[🔹 AUDIO (SIMPLIFIED)
iPhone 15 Pro built-in front mic. Quiet car interior ambient. Dry phone mic sound: proximity warmth, mild compression. Natural breath sounds. No cleanup. No music. No engine noise (car parked).]

[🔹 REALISM CONSTRAINTS (PRIORITIZED)
❌ NO smiling, NO text overlays, NO product in frame, NO external mic, NO cinematic grading]

Maybe it's changing the music for ambient noise

It aligns a bit with some of Wrigley's no music prompts, since it cuts sounds in various ways.

Well thats the point of these things, gotta learn for yourself too.

[pulsar coral]

Maybe a start to end simpler encased [Scene: ] prompt would keep it stable, but yeah if you're the one who can't get it stable we're probably effed here. They should really not take the older models out of the dropdown menu selection, that would solve so many issues. It's just common sense to keep them to avoid these type of problems.

They put up completely untested image and video models and just cross fingers for it to make money.

Like these for example are not image models but chat models at the Chinese Chatbot site I help at, they keep it all neat there with every version and they don't even have Google's resources, so I guess it's possible so what's the excuse for Flow not to do it this way.

Set cover is working if you select an image inside a project as cover, but the cover doesn't show the changes until you refresh the whole Flow page.

And yes that's sailor moon, I was just challenging Flow's prominent people error ok.

[lone seal]

Perfect, thanks! Of course it was something simple and stupid that I completely overlooked.

And yeah the prominent people thing is so horribly done--it's flagging OCs/random generated characters as prominent people, but I have no problem dropping The Deep or Homelander, etc into images with my characters if I want to 😂

[pulsar coral]

The VEO watermark is there due to policies about AI from each country, USA residents don't get it for Ultra plan and seems in Malaysia they don't either. Otherwise you need a specific VEO watermark remover because at this point the problems about it are too many just gonna say this one worked for me https://reel.money/tools/remove-veo-watermark

No watermark. If you use a random watermark remover for other things it'd give a moving blurr there or worse, needs to be specific for VEO

[exotic shell]

thank you for your response. have you tried using a vpn to not have watermarks?

[pulsar coral]

If you have a clean residential IP/proxy to a country without that restriction it sometimes worked for some people

Once a connection is marked as VPN it wont work anymore seems

And most commercial VPNs alike Nord etc are all flagged, they almost never work

Lemme test with this one, it's 1080 p

There we go, clean as a newborn no blurr or anything. Then again there could be better ones floating around, but again needs to be specific.

...tbh removing watermarks is bread and butter for everything AI atm, and rules are very inconsistent.

Like at Flow images don't have watermarks, but at Gemini they do. Who knows how it works.

Then again more likely when the Flow website was generated it lost the ability to add the watermark to images more as a bug than anything else.

I love adding lensflare to stuff

[pulsar coral]

The unusual activity errors can be fixed by clearing cookies, every google and labs cookie. When it starts working again don't generate more than 10 images at once and it will happen less or not at all.

It varies according to daily resources available for Flow, some days you can generate 25 images at once from free accounts without problems, other days it won't let you generate more than 6.

But paid plans Plus and above are more resilient to it, doesn't mean it wont also happen to those if you try hard enough by clicking the generation button too fast.

Also staff has said it may happen if you're at a VPN or sharing your account with someone, at a VPN someone else could be at Flow in your same network so yeah that could be adding all that extra activity.

A lot of the artistic difference between Whisk and Flow comes from the image models they use underneath. Flow uses Nano Banana 2/Pro, which seems to get instructions from Gemini to recreate or heavily reference images from its trained dataset, though the actual generation itself is probably being handled by an Imagen model underneath.

If the model recognizes a character or image pattern, it can make something extremely close to what it thinks that image is supposed to look like. That gives it really strong benchmark scores and high consistency, but in practice it can feel less like actual image generation and more like tracing or reconstructing things it has already seen in its training data. But yeah tracing is not art exactly.

Meanwhile, Whisk was using Nano Banana (Gemini 2.5 flash image) for it's precise mode if you had it turned on, if not more likely it'd use Imagen 3. Both are very similar to Imagen 4, but Imagen 4 seems bugged and confuses a massive amount of safe prompts with unsafe and blocks them, making it almost unusable as it is.

Both Whisk's models were more freely artistical instead of being locked into tracing, but you've already seen it if you tried it.

Flow used to have Gemini 2.5 flash image too but one day they disabled it without explanation and ignoring all questions about it.

Nano Banana (Gemini 2.5 flash image) is still useable at Google AI studio as a free introductory model but it's not at all as comfy to use there as it was at Whisk. It's also possible to create a similar website to Whisk that can use the image model by using Gemini's canvas mode, it's similar to prompting for an image but you ask it for like a website instead.

And then Nano Banana (Gemini 2.5 Flash image) is also available all over the internet in sites that use credits, but it's always unreasobly expensive to use there. More often it's just called "Nano Banana" not "Nano Banana 2" or Pro.

Also about Flow's Nano Banana 2/Pro's current subpar performance, the other day a mod around here said the Nano Banana models are NOT a Flow product and if there's complaints about it gotta go to the Gemini Discord to tell them... (?) which I was like...

"Then why did you bring every Whisk user here if they wanted to generate images with it? and who are all these Google staff idle here if they're not Google too and in charge of it? Everything I know is a lie"

So yeah point is if you don't like how Nano Banana 2/Pro works, gotta go tell them at their Gemini Discord.

[quick oasis]

hi. Uhm question. can you do Ip marvel comic characters in google flow with nano banana 2?. Because I really wanted to do like reference sheet of like they're iconic suits studies and stuff

[glad yacht]

Here is your answer!

[quick oasis]

man, that sucks alot. Oh will, just forget it

[pulsar coral]

actually as long as it's comics style and not realistic actors it might do it just fine.

Like don't try to make Hugh Jackman's Wolverine portrayal.

NB2/Pro is really bad at anatomy, so you have to compensate by prompt beforehand. NB1 was better at it.

Still is, they just don't have it here for whatever really bad reason.

[pulsar coral]

Oh yeah and don't write "Marvel Comics" in the image... that copyright more likely comes more from their cinematic universe than the comics.

But they wouldn't mind each character's name.

[glad yacht]

I didn't write Marvel or anything, simply ask for a Captain America portrait.

[hearty hound]

If you use a vpn it doesn't have copyright protection, i can generate Marvel characters just fine lol

[pulsar coral]

Sure you're not using NB1? Cause your account has it somehow

It wouldn't surprise me a VPN would work tho

[hearty hound]

No i'm using Nano 2, when Whisk shut down Nano 1 disappeared in Flow but i don't have Imagen 4 what you had i only have Nano 2 and pro now

[pulsar coral]

The problem with NB2 is it goes realistic from the getgo very often, so more likely it tried to get movie picture of captain america and derped itself. But these comic styles and ambiguous prompting passes more often.

Captain America wasn't my first idea tho cough

Shoulda tossed muscle in it, but ok for a quick test

That said yeah very likely even like that it would fail sometimes, NB2 is worse atm

Let's shoot like 20

Yeah dude's ok

Added muscle this time and it sucked less

[paper ermine]

Bro use my image

[crisp basalt]

No

[paper ermine]

wdym no?

[crisp basalt]

That’s not the point here lmao

[paper ermine]

K

[crisp basalt]

The main issue is Veo not being able to generate seamless transition

[paper ermine]

It did tho

Lemme send you real quick hold on

[crisp basalt]

I don’t care about the pic 😭

If you manage to generate the video then send the video and prompt

[paper ermine]

https://www.facebook.com/reel/1948228402731915

[crisp basalt]

Bro 😭 post the damn video , I don’t wanna open the Facebook

[paper ermine]

company property cant send it here so either you go to my comp facebook or non dude

VEO DID a really good job last month but now? Pure of crap

[pulsar coral]

Well I'll be a monkey's

He's indeed saying it's the model's fault that it can't do it this way now.

[hearty hound]

Company property what lmao

[pulsar coral]

Which is believable with all the weird changes it had

[hearty hound]

Would people already be using AI to make movies or something or trailers?

I wonder

[pulsar coral]

I've already seen really awesome AI short movies but with Seedance.

[hearty hound]

Damn so it's actually possible

[pulsar coral]

Or like they're making series, and Batman or Captain America can be there without issue

[hearty hound]

I wouldn't be surprised anime will soon be made with AI or partly with AI

[pulsar coral]

Like you can look in youtube for it stuff alike "
Seedance 2.0 is the Craziest AI Model"

Makes it feel like we're with sticks and stones here

[paper ermine]

Bro fr. Im not even joking Cant posting stuff around wthout perm

[hearty hound]

I thought kling was the craziest

[hearty hound]

How is AI generated content company property lel

[paper ermine]

The design

[hearty hound]

Ahhhh

That makes sense

[paper ermine]

The architecture design are comp Prop man

[pulsar coral]

I got rid of it now, was just a quick peek

[paper ermine]

Thanks dawg.

[pulsar coral]

Wolverine is stupidly copyrighted, nothing next to his name and the word "comic" passes. More likely cause there's movies with his name.

My derrierre tho, there's no way to block every word to copyright and sensor things, all they did with all this is destroy the consistency of the image generations.

Nice 3 feet yeah.

Never takes more more than 5 minutes to still get what I want... it'll more often be ugly digital art tho, so it's not worth it over NB1 quality.

Can prompt proper anatomy to avoid 3 feet, but that's quick

[glad yacht]

@pulsar coral

Test this

[pulsar coral]

Aw man no thanks

Just with wolverine I'm sure it's not gonna pass, one can see it from the workarounds I had to do up there.

But thing is a wolverine is also a real animal, it's kind of brain dead to have blocked the word

"Cinematic " is not gonna let it pass either... less I'd really need to find a workaround.

Words like "vintage filmation" etc, something must pass or the model would have too many words blocked and that'd be the end of it.

If it's not already.

Spiderman is also blocked? Haha hold my beer

[quick oasis]

yes and no at the same time. Because usually it only does tom holland's spiderman alot. Not much of like the classic steve ditko or scott johnson spiderman style. Just mostly mcu and sony movies for some small cases if you ever decide to do realism style. Even when you write a description instead putting spiderman name on google flow sometimes

[pulsar coral]

Let's see how many animal/critter names they can block before this becomes obsolete

It just can't be done

But yeah Tom Holland I'm not even gonna try.. but I bet braining it really hard there's also ambiguous workarounds for that

idk if this would be even easier with reference images tbh

The little details can be prompted to be correct after that, when it wasn't this blocked it'd also miss them often.

... that's just with the hero stuff, you don't even wanna know what that prompting style can do to the supposed nsfw blocks.

Was that even 5 mins.

I imagine that's also why "Captain America" wasn't so blocked, imagine not being able to use just those two words, stonks would go down to the floor.

We're lucky there's no superhero called "Otter man", there'd be even less to post around here, if you know what I mean. And goodnight.

[pulsar coral]

Ok then

[opaque sand]

Copyright restrictions vary from country to country. Therefore .

[pulsar coral]

Well in mine it was blocked. I have made 0 tests with VPNs tho, so no idea if works or not.

I don't think it would work tho, it can't be that easy.

This is probably a good way to be sure if it's blocked or not in your country, but again I don't believe yet in that superstition.

Jarvis what's the country with less copyright blocks

How the hell do I proxy to Somalia, I think it'd be easy to fly over there

And what's Eritrea

Of course it's not gonna block anything if the country is pretty much next to Latveria

I tried from Russia, Thailand, Malaysia, Indonesia, only got like 1 more image, but seems to have been by chance. I think I can just change the artstyle to photorealistic if I wanted an image like that anyway.

1 outta 40 in Vietnam... yeah I think it's just an illusion by chance.

But man this must be straining Flow, no wonder one can only generate like 8 images at once atm, dunno if it happens to Ultra.

[violet ruin]

Hello everyone I tried with help to make my veo3.1 handheld not static but it still doesn’t work

[pulsar coral]

Are you using a reference image for this

[violet ruin]

Yes

[pulsar coral]

That explains, that woman looked nothing like the ones VEO was generating to me. Send me the image.

[violet ruin]

Okay

[karmic moat]

Hi guys, I wanna ask a question. I'm using Google AI Ultra to use Flow. But I saw a post, that Google no longer sell this kind of package anymore and change to AI Expanded Access. But this AI Expanded Access add-on no longer includes higher access to Flow. So which plan should I buy now ?

[violet ruin]

Seriously

[violet ruin]

Are you still there

[pulsar coral]

Yeah running tests again, it's as fast as videos generate so can't be helped.

[violet ruin]

Thanks was checking. Arg, man I hear your running tests again and can’t be helped. So I can’t create a handheld camera movement but you did it😓

[pulsar coral]

Veo just thinks differently with a reference image, gonna rearrange it

[violet ruin]

Thank you

[pulsar coral]

Look at er'

[violet ruin]

God dam you always seem to impress me but not in a weird way. The only issue is that her hand is very far but I’m being picky this is great. You don’t give up. You a problem solver that takes on a challenge 🙏

[violet ruin]

How did you you manage to do it?

[gloomy verge]

Hello. Can I ask one questrion. I don´t understand one feature in Flow. How difference is between, when I only load some image to reference and when I add to prompt "@... " too? Are there some differences? Thank you for help. I hope, that you undertand my question.

[violet ruin]

Can you show the prompt and image

[gloomy verge]

For how reason is there feature "@..."? I don´t understand it.

[violet ruin]

So do you want a women in a shirt wearing that red bow

[gloomy verge]

Yes, but when I don´t used "@...", it generated bow too.

[violet ruin]

MrMegaBuster are you there

[gloomy verge]

If it generate female with bowtie image without "@bowtie", so for how reason is there this feature. I don´t understand it.

[stone garden]

Bro, this is crazy, I did two generations of img and now I get that? What's the rule behind this thing? So annoying

[pulsar coral]

Yeah it started doing some weird things but I'm taking them out.

[violet ruin]

Dam.

[violet ruin]

Why don’t you. Remove the bowtie then out from the prompt and continue what you were doing

[gloomy verge]

I just wanted to know what this "@..." function is for.

[violet ruin]

It checks the placeholder text or docs for the specific image tool/interface you’re using.

[gloomy verge]

It means, that when I use "@...", I get better result images instead without "@..."?

[glad yacht]

People more often use it in video generation to @ different characters or ingredients etc..
In image generation, it doesn't improve or get better results.

[gloomy verge]

So, in image generation, it is useless. Yes?

[glad yacht]

When you Prompt reference image 1 character etc can be simplified as @the character name you created etc when creating an image.

[stone garden]

It's useful if you have more than one image reference and the more complex the prompt becomes, the more useful it becomes

[opaque sand]

What the hell is going on with the lip-sync since yesterday? I stopped working yesterday, thinking that something would change today.... There isn’t enough grass in the world for me to touch to understand why even the most basic prompt and image for lip-syncing leads to generation errors in 99 (REAL 99 ATTEMPTS) out of 100

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

Over and over again... any characters, phrases, prompts. I'm just blown away..............

[exotic shell]

Since the day of the deletion of veo 3.1 fast low priority i would again like ask google to think about it and keep it. the lite version is a massive downgrade which makes a ultra subscription useless. i need to provide consistent content with good quality. i cannot suddenly go back to trash content.
please think about that decision. 🙂

[blissful abyss]

Did Google fix.the.gemini chat that says I can't edit images at this point.

[opaque sand]

What does that 5/10 even mean???

[exotic shell]

i never figured out. that numbers never changed. it will be deleted in 1-2 days anyways. 🙁

[opaque sand]

Let's see what kind of “quality” it offers... I hope I don't get banned for saying that Dreamina Seedance 2.0 is already a cut above the rest, and if you don't use it as a conveyor belt for neuroslope, the unlimited subscription fee is worth it for just one person a month...

I just spent three hours trying to get a 111111single1111111 lip-synced shot for a project I was supposed to finish yesterday, so I'm really angry... really...

[exotic shell]

to which unlimited-subscription are you relating? seedance or veo?
If seedance is already a competitor, making veo even more uselss by downgrading it, might not be the best idea.

[exotic shell]

i feel you. lite isnt really needed. most regenerations come from getting a good result, not from prototyping as they say.

[candid bay]

what exactly do you mean? which limit error?

[analog lava]

Please remember our first rule: Be compassionate and treat others with kindness and respect. We have #1238621511401734254 for feedback specific questions/ queries.

[tropic laurel]

10 of may - 5/10

[exotic shell]

i just tried the remover you mentioned again, for me it creates blurr. sadly

[glad yacht]

They will discontinued this model on May 10th.

[frosty niche]

Making way for 3.2 perhaps. Who knows.

[exotic shell]

doesnt help to create a gap

[frosty niche]

The google gemini server may have a better anwser for that.

[exotic shell]

googles pocket as well, 280$ for a subscription is not enough!

[violet ruin]

I tried the prompts be still didn't work

[ancient vigil]

Just throwing my sadness into the void.
I had images from Whisk and ImageFX I was working on. I hadn't touched them in a little bit. Didn't know about them being retired. Missed the announcement about doing a data transfer.

[pulsar coral]

Weird how it acted differently for me yesterday... Veo just acts differently everyday, well will see what to do with it

[violet ruin]

Hello everyone can someone please help me to make my base image to be handheld camera motion behaviour if this is even possible I’m using VE03 as my video generation. https://drive.google.com/file/d/1qHG0oQ2beID1f4WVdMlZPQRoo7n5Q_Gy/view?usp=sharing

[wide moon]

Dammm they really messed up NB2 and Pro

The prompts that usually works are generated into something different

It’s like Imagen 4 it

[wide moon]

I meant the roles are reversed

[crisp basalt]

have u tried using "ingredients" to generate that type of video u wanted?

[crisp basalt]

https://labs.google/fx/tools/flow/shared/video/042c443d-fc2f-4c6d-a11b-1c655b1101b8

@violet ruin https://labs.google/fx/tools/flow/shared/video/1709b232-8c0e-4892-ac35-001c320bef70

[chrome folio]

hello Staff, <@&1366831844502274180> - Are there any restrictions on automating workflows using a desktop app extension that interacts with Flow?

[pulsar coral]

Mine's also have those weird little stutter cuts atm

Could be related to the rest of the prompt, I tried simplifying it but ended up just using Gemini's again

I think it has to do with the word "Shaky" veo seems to understand it as stutter

Veo is so intentionally random, almost as if there was a chat model attached to it... which most likely is.

Like who even asked for a subtitle. But yeah shaky is understood different than "trembly"

I know what's causing it now, it always happens after the first time she says "melanin" there must be a pause or something there

Mines also do it there.

For some reason it understands "pause" as frame skip

Then again sometimes might be pure luck when it works

[bitter wyvern]

When will the raffle for an ultra account for members take place?🫡

[violet ruin]

I feel you.

[violet ruin]

No honestly I have yet.

[violet ruin]

Thank you for hearing me out.

[violet ruin]

I give VEO3.1 a master with ChatGPT and Claude, still I get output video wrong 😑

[violet ruin]

How did you manage to it off. Was it because the dialogue and pause in it. I heard you mentioned it

[pulsar coral]

Yeah I'm trying by removing the pauses and timing from that one "melanin" word, but then other stuff fails sometimes. That one comes 1 outta 4 or so

Pause probably pauses the whole vid for a fraction, but sometimes it understands it differently and doesnt

Maybe it's better to use "quiet" instead of pause.

Gonna give her that accent too, I think I'm close this time

[violet ruin]

I hear you, since you understand my prompt do you feel that it’s too long or it’s okay. Ya, the fact that the one comes outta 4 that’s a low rate meaning less likely to get the best results again.

[pulsar coral]

Nah gemini seems to have done most of it right actually, it's veo's randomness what is unpredictable

[violet ruin]

Definitely, she has looks the same but different voice. • Accent: Standard American (neutral) with natural AAVE cadence and inflection. Clear diction, warm resonance, no upward vocal fry. Authentic cultural rhythm without caricature.

[pulsar coral]

Veo is blocking the word "fonts"... I don't think these people can be in this business anymore, what's the point really

[violet ruin]

Ag, I never thought 💭 of it like that. I thought honestly are are the same. Same as in same model but I use Veo 3.1 because it more cheaper and best lip sync and audio.

[violet ruin]

What do you mean.

[pulsar coral]

The filter that should block copyright and NSFW terms thinks it should block the word "fonts"

I was using it to avoid subititles since it seemed logical, but wont allow to use it

[pulsar coral]

There's other ways anyway, but its kinda dumb

There's this random dude which I can't locate where he's coming from sometimes

"Tips" is still stuttering

0:00:04

These things are causing it at random

[violet ruin]

From the listening there was an audio break or stuttering on the tips, voice is good, camera shake I see. I hate the fact it cause it to be so random

[pulsar coral]

I was adding "no zooming effects" and it was adding "Zoom" sounds instead of understanding it as no closeups.

With Wan woulda been done immediately sheesh.

[crisp basalt]

I feel like they intentionally dumbed down veo understanding level towards user prompt ,it’s annoying ngl

The constant zooming in , glitch effect that I’ve been getting everytime I prompted it “amateur footage. Style”

[pulsar coral]

Probably asking Gemini to make the prompts is not the best idea actually, unless it worked for someone then it did.

But yeah mostly seems like the Veo model's fault, it seems sometimes it's not being allowed to do what it's supposed to do, same as it happens to the Nano Banana models now.

[pulsar coral]

Anyway maybe "professional" would work better than "Amateur"... even tho that's not really the intention, but maybe it would cause less random distortions and stuff

Veo just understands things in a really weird way

Best attemtp so far, let's see what's the success ratio now.

But yeah it's been useful to finally sit down to learn about this, got tons of feedback.

Chatgpt has been more useful at this than Gemini

Second went better

Stutter at melanin at third attempt

[grim nest]

Why im getting that error help.plz

[pulsar coral]

Like it says it might have accidentally tried to generate something copyrighted. It'll do that. If doesn't work after various attempts you'll have to edit your prompt, again, like it says.

Fourth video attempt went fine

5th went fine with a little effect.

6th went fine. One more and that's enough ratio

[proper epoch]

Can we generate audio in flow?

[silk gulch]

Anyone using flow for youtube automation

[glad yacht]

You two better create a Thread to better focus on this project.

[pulsar coral]

It's more of a gacha than a project anyway.

[glad yacht]

Prompt challenge May, 9th

[neon leaf]

ey bro flow didnt work? hit generate? prompt in ekstensi?

need info

[icy halo]

hi there

[wheat wren]

@icy halo hello

[icy halo]

hi there

[remote bay]

From what I've been told, you can do this via google AI studio, in which you can get an API key for a per project basis, however, it may be costly!

[lavish glen]

has anyone else noticed that the number of concurrent generations on Flow just got nerfed? A few days ago I could run up to 20 in parallel, now I'm capped way lower

[exotic shell]

btw its the last day of veo 3.1 fast (low priority) existing. tomorrow it will be deleted and we have to pay for it.
enjoy your last few hours of quality

[icy halo]

hi

[pulsar coral]

Yeah even with paid plans it only allows like 8 generatios at once atm. That summed to the augments in copyright and censorship (which are pretty much just causing bugs at this point instead of censoring things) made worflows 3 times slower or almost impossible for some people.

Funny enough Whisk using Nano Banana (Gemini 2.5 flash image) would only allow to generate 5 images at once and it would still work better than that with way many more useful results.

[frosty niche]

Is that w/ or w/o ultra?

[pulsar coral]

Apparently with all plans free or not. Could be different for someone here and there but yeah.

And I guess it might change soon or not with daily demand, hasn't been the first time.

Anyway go above 8 at once and you're gonna start getting error messages, which more often still get fixed by clearing cookies untill you do actually reach your accounts daily limit, then not even that will work, until next day at least.

[frosty niche]

I'm a bit terrified to even test that out. It always craps out at, at minimum 3 concurrent most of the time. But i havent zoned in on it lately.

[wide moon]

Its like the roles are reversed for NB 2 and Pro like your using up 2 more than Pro which give you a limited amount of generations for the free tier

[rigid nest]

Google team, May I have an answer please?

I am sure other community users also wonder about it.

[pulsar coral]

According to what you're doing NB2 is gonna sometimes be better than Pro mode. Pro is better at literal realism, NB2 is better for unrealistic, semi-realistic and anime.

[rigid nest]

Flow is better for finding new bugs everyday

[pulsar coral]

Like I do beach shots a lot, Pro flattens the ocean alike a real photo, NB2 looks more cartoony, digital art or anime

Would seem like a good thing, but you don't want a photo of an ocean at the background where you're generating an anime character, looks awful

It's more likely cause cause NB2/Pro traces real photos of oceans from their datasets, rather than artistically drawing things like NB1 and Imagen models do by themselves

But idk if you make photos of coffee mugs , tomato soup cans or whatever products with text and alike, Pro is gonna work nicely

[pulsar coral]

But yeah that doesn't excuse the lack of answer to feedback or fixing bugs. Lately some error messages state better their causes, not that it's a fix to them.

[rigid nest]

The team reporting bugs seem to be doing a good job, but I am disappointed at the limited tracking and support for bugs.

Some bugs have been reported 50 plus times, yet we do not know what is causing it, what work is going on and if there is a timeline for a resolution.

In the mean time the quality of generations are going down by the day.
Flow seems like a weekend project at this point or perhaps heavily under resourced.

[candid bay]

some bugs, as in which ones?

[rigid nest]

Oops Something went wrong
Whisk Migration Issues
Quality of generation going down after few edits
Famous people false positives
New feature ideas

To quote a few

[candid bay]

yeah, you're right lol. the famous people false positives is so strange to me. also i haven't the one with quality going down after a few edits

BUT i've seen the one in which the model doesn't follow the prompt at all and makes some random nonsense for some reason

[rigid nest]

I have seen this too. It's too rare to bother me at the moment.

But the quality and experience have definitely gone down in last 1.5 months.
It was amazing during Whisk days and when Flow was separate.

[exotic shell]

and starting today we are forced to stick to lite, another downgrade 😄

[pulsar coral]

Google gets their money from whatever other sources too, but I'd like any of the staff to sit down with Flow for a while to try to make income from it in the usual ways as they left it for the rest... they seriously need some empathy around here.

That or they might think most people just pay for plans to sit down and play.

[candid bay]

i've made some tests. seems like filters are more normal now? but i'm not sure yet

have you guys ever noticed filters are more strict or lenient at certain times of the day?

[pulsar coral]

Neh with old prompts feels samey still casually producing an image sometimes 1 outta 40, nothing like it was like a month ago. And it's still limited to 8 image generations at once.

Doesn't mean ambiguous re-prompting doesn't still work.

... which made all this hassle kinda pointless. I still rather use NB1 wherever.

[candid bay]

bro, have you seen that you can use NB1 on Flow now? maybe i'm being silly but now you can use "Imagen 4" in Flow to also edit. up until very recently it would only generate, not also edit

[pulsar coral]

Flow's Imagen 4 sorta looks like NB1 yes, but it's flawed for long prompts even when they're really safe.

It's not really the same model tho sometimes looks like it

[candid bay]

it looks like the NB1 used in Whisk, but of lower quality? idk, i remember it was slightly better at prompt-following

but visually it looks exactly like NB1

[pulsar coral]

Here's proof of it, this is the reporter girl vintage anime images prompt, it's completely safe, NB1 and NB2/Pro accept it, but Flow's imagen 4 doesn't

But NB2 makes them ugly digital art while NB1 keeps the vintage acetate filter.

Not gonna post all that again, I've done it enough times

But still here's just an old image of Whisk running the prompt with it's NB1/precise mode without issues. Flow's imagen 4 completely refuses it, it's bugged.

Imagen 4 is just not the same as NB1.

Wherever you'd find NB1 the prompt would still run without issues too. This is just one example tho- Whisk thread was full of issues alike, but they deleted it here.

Then again I think Flow's imagen 4 still accepts it without reference image. But that's not always useful either.

which this was it

[glad yacht]

No, you can't use Imagen ⁴ to edit image. It doesn't accept reference photos!
You can only use it for text to image generation.

[pulsar coral]

... I think that's a little of a deal breaker yes. NB1 doesn't have the prob

[pulsar coral]

Imagen 4 itself wouldn't, but somehow Flow's does accept reference images tho, it just derps itself on longer even safest prompts because it seems to have been badly tested, so that might make it seem like it doesn't accept references.

It's like they here think this replaced NB1, didn't even check it doesn't and called it a day (?)

I've called them on it before at feedback too, but we all know the story there.

[glad yacht]

That's because when image edit mode , only NB2/Pro available, Imagen 4 is Gray out. It accepted doesn't meant it's use Imagen⁴.

[pulsar coral]

Yeah imagen4 is indeed not capable of editing references, this here is more likely similar to Whisk's mode when you turned precision off, and using some sort of Gemini model to achieve this.

It wouldn't be beyond how things work here that they wrongly called it Imagen 4, due to not wanting to add a precision mode button again.

[glad yacht]

Please stop spreading rumors about imagen⁴ can be use in edit mode!

[candid bay]

want me to prove it to you? lol

[pulsar coral]

Tell that to them, like i said, you're not wrong about it.

No Yaska is right, Imagen is an artistry model and shouldn't be able to edit a reference.

Flow people more likely called this image model "Imagen 4" wrongly

[candid bay]

because it's not Imagen 4 anymore, it's NB1 with the wrong name...

All imagen models are strictly generation models, no editing. this is NB1

[glad yacht]

This is a reference image mode , you can select an image in the folder generated by Imagen ⁴ eariler.

[pulsar coral]

Yeah but that's totally done from Flow's selection. And no this not NB1, Idk what it is tbh. You can run the reporter girl experiment to see it.

[candid bay]

if you try to edit the image by clicking on it first, there's no "Imagen 4"

just add it to the prompt then select Imagen 4

[candid bay]

it's exactly NB1 from Whisk. same bad quality and artifacts lol

it's not the NB1 from Gemini.. that was decent

[glad yacht]

A Very simple test.
Try editing photo with none English content, see your imagen ⁴ can generate correctly then it's not imagen ⁴.

[pulsar coral]

You see, NB2/Pro work by using a Gemini chat model to instruct them how to draw things, that's why their real names are something like "Gemini flash image"

The instruction goes to an "Imagen" model that's also, why it's in the name

[candid bay]

well, nanobanana was initially just an internal codename lol

[pulsar coral]

It seems whatever "Imagen 4" flow has there is attached to a newer Gemini model.

[candid bay]

gemini flash image... so on, was always their real names

[pulsar coral]

But it has been badly coded or whatever

[candid bay]

idk what it is but it's a watered down version of something

because omg it's so full of artifacts

[pulsar coral]

Because it's understandable it would censor NSFW, but its also blocking longer safe prompts

Its just bad design

[candid bay]

imagen 4 was always more filtered than NB2 and pro

imagen3 was the real deal lol

[pulsar coral]

... Yaska just made me notice what's pretty much another bug here, and it's been in our noses all this time

Flow's Imagen 4 has the wrong name, whatever the thing could be called now

Also it's possible the real NB1 is still running on Imagen3 + Gemini 2,5 , that's why it still accepts longer safe prompts normally.

[candid bay]

NB1 is kinda Imagen 4 with Gemini 2.5 slapped on top of it. Imagen 3 is so different from 4 visually

[pulsar coral]

Yeah that sounds more like it too

... question now is, now that we know or suspect this, how do we use it in our favor

[candid bay]

Don't feel like going back to Imagen 4 now lol. Would have to change completely how I prompt and edit images lol

[glad yacht]

The Bottom Line: You can "edit" an Imagen 4 result in Flow, but technically, the Nano Banana²/pro architecture is likely doing the all the lifting for the actual movement or changes you're asking for.

[candid bay]

because it's not Imagen4, it's NB1... wrong name in the UI

[pulsar coral]

Eitherway, that shouldn't be called Imagen4 there, funny.

[candid bay]

its not possible for an image model to do any lifting for a different image model...

[pulsar coral]

Or like, which is where I'm going to leave it

[candid bay]

would be kinda funny if they brought back an ancient model (by NB2 and Pro standards) to Flow by not so popular demand...

most people are crying after Whisk's UI and functionality than specifically NB1

but if you kinda think about it, Whisk was kinda revolving around NB1 and imagen3/4

[pulsar coral]

NB1 is just not so easy to find now tho. How would it be possible to slap Gemini 2.5 back to Imagen 4 or 3. I've only gotten it as it is from a Gemini canvas.

[candid bay]

eh, if they kept Whisk but added NB2 and Pro to it, it would've been just perfect

[pulsar coral]

Bro I just noticed in your image, our interfaces work differently.

Ours do have the upload button, yours doesn't

Jeebus

Not the first time I see it tho, but with other things

[glad yacht]

This is the beta version of Flow

[pulsar coral]

This is what ours look like with the upload button for "Imagen 4"

[glad yacht]

Ask it to add Chinese or Japanese text to the photo.
Example: (add "流动" text in background)

[pulsar coral]

And no doubt that's an Imagen 4 based generation. Because when you compare it to an NB2 image which I did now, it immediately recognizes the character from it's trained images as "Weiss Schnee" from RWBY and gives it an accurate wiki image face. Imagen 4's is completely different and more original, unless you'd prompt to give it the same facial structure precisely.

[sage otter]

oh u guys are having hte same convesation we are 😄

[pulsar coral]

It might look better at first glance but yeah... in this case works for this particular canon character, but when you're trying to make an original character with no relation, that's not useful.

And doesn't look that good either, than hand was awful.

[sage otter]

imagen4* is definitely better in some areas, like drawing peopels faces sometimes

[glad yacht]

Imagen ⁴ only speak English. If you tell it to generate anything else it will be all geribish text. The simplest test, but none of you are willing to try.

[pulsar coral]

Yeah it couldn't do it, well sucks for the noodle side of the world

[candid bay]

i tried, this is imagen4

it's... correct? omg it's kinda good

[pulsar coral]

If you go simple enough it will give you something yes, and it does look good

No reason to be crippled just because they want it like that here tho

[glad yacht]

Now tell it to replace the sign with Chinese

[pulsar coral]

It can probably copy it like this, I've seen it doing weird fonts too.

[candid bay]

it doesnt do it. it adds these strange light blue artifacts on top of it

but thats it

[pulsar coral]

But yeah it did fail to understand it

[glad yacht]

Imagen⁴ 100%

[pulsar coral]

Whatever Gemini model that's attached to it seems to be prob... but yeah since surely it will never be fixed it's not very useful.

[crisp basalt]

DAMN they are sunsetting Imagen model too (this is only for Google Vertex AI API ), but if Google is doing that for Imagen… do we have another video model that’s…similar like Gemini Flash Image (Nano Banana)? What if there’s actually Gemini flash Video …

.

[pulsar coral]

Goodbye Google AI studio subscriptions if they do that... worse in NB2/Pro's current state. Who knows why do they hate themselves so much tbh...

Then again it doesn't look so bad, actually...

I think it actually means they know they can't get rid of 2.5

... then why in cuckoos is it still disabled at Flow

[sage otter]

will likely stop using altogether if they get rid of imagen

was bad enough they removed imagen 3

[frosty niche]

Yeah, i thought this was an interesting move to. Saw it happening early last month, it's a bit unfortunate.

[pulsar coral]

Yeah

Also now that I try Nano Banana Pro from a free account, the daily limit for it has extended a lot, dunno how much yet tho. And it's indeed just like very little less censored. Almost nothing tho, specially compared to like a month ago.

But the generation limit is still just like 8 at once, and if you don't wait till all of them generate and for like a couple minutes before generating more again you get the unusual activity error.

It's the never ending story with this, I'm just going back to a Gemini canvas, not worth the waste of time.

[lusty sinew]

They really need to fix this 8 at once error, it's been irritating

[crisp basalt]

Oops

[pulsar coral]

It's probably intentional tho, typical that they'd do changes without testing them on friday and then don't show up till monday.

[crisp basalt]

Fun fact: the actual code name of imagen 4 wasn’t really 4 it was imagen3_5 🙂

[frosty niche]

NBP is usually my goto but even then I'm usually working on mulitple things so waiting for the gen output isnt the worse thing, but it helps in looking for quality or more of what tastes you for your prompt des.

[frosty niche]

In this day & age codenames are for the birds.

[crisp basalt]

@frosty niche have u tried using nano banana on aistudio

[pulsar coral]

Makes sense.

[pulsar coral]

Only Whisk users got enough practice with prompts, people who got to Flow first don't seem to get used so well to it

[frosty niche]

Shoot been a bit, I was messing around with it for a specific project a couple of months ago, nice testing session.

I feel like the watermark in AI studio is extra watery.

[crisp basalt]

TIL nb2 has the capabilities to do image search

Makes me wonder if nb2 in flow has that capability

[frosty niche]

I do sense a learning curve as well.

[crisp basalt]

Wait u got watermark?

[pulsar coral]

I thought that's what it was doing since the start... but then settled thinking it had just trained the images.

Weird they'd disclose this

[crisp basalt]

I can confirm nb2 actually uses image search

[frosty niche]

That image search is what makes it more legal compliant basically or opens it up more?

[crisp basalt]

I think makes it more accurate?

Idk what context but let me try something

[frosty niche]

Acouple months back i did, when i was using AI studio like everyday.

[pulsar coral]

Just tell it to do a character from a popular enough anime, without reference image. I do it all the time, that's how it searches.

Or a videogame, or cartoon

In fact I spent yesterday doing marvel characters here, but those don't go through so easily

[crisp basalt]

DAMN

Minus the krusty krab being randomly placed behind theee

WAIT that explains why in flow one time I got nb2 to generate tony stark scene from endgame

https://labs.google/fx/tools/flow/shared/image/98d9745d-1169-4b4f-9e10-bddeec3fa98e

It was using image search capabilities

[frosty niche]

The light/bright mode is frying my eyes!
You don't do dark mode? I mean do your thing.

[crisp basalt]

https://tenor.com/view/job-application-jump-scare-gif-11326455699963497416

[pulsar coral]

When you make it generate some canon character you can literally go to Google images and try to find the image it used

[crisp basalt]

JOB APPLICATIKN?????

[pulsar coral]

I do sometimes find them

[crisp basalt]

The gif 😭

[pulsar coral]

... my experiments about it were at the Whisk thread tho, so it poofed now

[frosty niche]

Yup sums it up well.

[frosty niche]

May be a correlation, who knows thats backend stuff there.

[crisp basalt]

Hold on there’s something weird but I couldn’t put my finger

[frosty niche]

Danggg all these are in AI studio, thats nutty.

[crisp basalt]

Watch them nerfing this tomorrow

[pulsar coral]

Art station actually allows to use it's images for training image models, it's in the user terms when you create an account there

[frosty niche]

They probably got their claude+antigravity bot reading this & pushing a pr now.

[pulsar coral]

This info is useful actually, because people always complain that the image models steal images

[frosty niche]

Sure is.

[pulsar coral]

Not very sure what Reddit thinks about it

[crisp basalt]

Ok aistudio is actually really good

[frosty niche]

Japan never had a "The Office" yet? Now thats nutty.

[crisp basalt]

If Arab got one , I need to see Japan version 👁️

Oh it’s still generating

[frosty niche]

We all know how the UK version went...

[crisp basalt]

https://tenor.com/view/love-you-happy-birthday-gif-19440393

[frosty niche]

I think the longest I had AI studio running was like 15 minutes.

[crisp basalt]

Wait it’s still generating , wow am I pushing it limitation or

[frosty niche]

It usually times out, at 15 mins basically kind of how jules hullcinates really.

[pulsar coral]

I think they were gonna make it the same at Twitter, and put in the terms that images are ok to be used for training AI... must've never happened cause everyone would have fled.

And it would explain why NB2 still struggles to recognize some more niche canon characters, it really isn't trained to everything yet

If training is even what it was doing anymore

[crisp basalt]

Timed out

[frosty niche]

Yeah, with cameras being everywhere now-a-days, doesn't surprise me, theres probably datasets from cctv footage by now.

[frosty niche]

what time did you max out at?

[crisp basalt]

4m

[frosty niche]

Dang I was stress testing acouple months ago getting 15 easy thats interesting.

[crisp basalt]

Oh

I think the consequences is catching up

[pulsar coral]

... if they'd force allow to use twitter as training source the model would be almost sentient perfect by itself without references at all.

[frosty niche]

I guess that's how one earns their pay packages now-a-days.

[pulsar coral]

But then again that wouldn't matter much since everyone wants to make their own art, not copy it.

Google's got the marketing backwards

[crisp basalt]

I think they banned me

[frosty niche]

That claude+antigravity clamped you it looks like.

[pulsar coral]

Nah it's probably the cookie error.

[frosty niche]

refresh, or try a different browser.

[pulsar coral]

Nano2/Pro models have been annoying everyone with it all day.

[crisp basalt]

Everything is accurate

[pulsar coral]

... wiki images, I've been saying all the time that's where images were coming from mainly

They must have the same terms at the wikis

[frosty niche]

Isn't wikipedia basically an X or soon to be X product?

[pulsar coral]

idk but elon ain't involved with it atm

[frosty niche]

I was expecting sushi & sake...but my japanese is horrendous.

[frosty niche]

He was earlier, probably sniffing up datasets or something. Interesting pushback & timing.

[valid nymph]

IMAX scales?

Nah it's 4:3 cut

[frosty niche]

This is nice, it has that art energy where they dance with paint in front of a crowd.

[pulsar coral]

He's indeed been telling people to stop donating to it.. and hates Wikipedia a lot for some reason. But Elon's dramas always happen out of nowhere anyway, so it's probably nothing

[frosty niche]

Interesting timing if you ask me.
Yeah, things in tech usually change pretty quickly & sour up.
I just can believe some academic institutions were allowing citable information to come from there anyways. The edit button always gave me the creeps tbh.

[pulsar coral]

Eitherway point is wikis are probably the easiest endpoints for Flow's NB2/Pro to get images, they always appear first in searches, also they come with all sorts of information to know if the image is copyrighted, prominent person or not

I knew that's how they had to be doing it

[frosty niche]

https://tenor.com/view/sapo-gif-15065622896911975780

[frosty niche]

Thats one way, sure.

[pulsar coral]

Wikis also have galleries of images, specially for canon fantasy characters, all those come from there

[frosty niche]

wikipedia should change their name to dataset-pedia then.

[pulsar coral]

Like I generate this character often, she always looks a lot like those first 6 images.

... problem is it this is also how the prominent people error happens, while trying to make your character wiki accurate it might accidentally confuse it with someone famous and then you're effed.

[frosty niche]

https://tenor.com/view/squirel-red-walnut-nut-shelling-gif-15975705

[frosty niche]

Interesting find here.

& yup dataset-pedia it is.

[dusty bay]

Hey everyone 👋

I recently finished a war-based AI cinematic short film called THE GRAVEMARCH.
It focuses on immersive battlefield storytelling, cinematic visuals, and AI-assisted production techniques.
I’d really appreciate any constructive feedback from fellow filmmakers and editors.
Watch here:
https://youtu.be/1e009mzsfH0

[exotic shell]

last few hours of 3.1 fast (low prio) and i have to deal with this:
"We noticed some unusual activity. Please visit the Help Center for more information." 😄

[shell wadi]

Are we now going to start paying for the delete button?

[vague island]

same

with nano banana as well

[exotic shell]

im curious how those who are using flow for work are going to proceed when 3.1 fast low prio is gone

[wintry pecan]

I’ve paid for the Ultra tier, yet it’s been very glitchy lately and I can’t get any work done at all

[candid bay]

try again, reload the page, change browsers

[limpid ether]

Even if I do that, this error comes back again.

Does anyone know the reason for this error?

[grave granite]

Animate in runing Activity.

[clear cipher]

This error is quite common. It happens because you likely used a word that involves copyright, such as when you include a famous person’s name or something linked to a brand or trademark.

Try rewriting your prompt asking to remove all words that contain copyright or celebrity names and replace those words with characteristics. For example, if it is a brand with two colors, instead of using the brand name, you describe a brand and include the two colors.

[hybrid pawn]

Google flow you are seriously making me question my subscription multiples times a day!!!

[brittle bloom]

Animate? Sorry u use in flow video not begging

[candid bay]

it's untrue, it's a bug.. lol

[limpid ether]

There is nothing related to copyright issues, famous names, or anything of that nature in these prompts. So, why is this error appearing?

[limpid ether]

So Is this their bug?

[candid bay]

it's one of them. so if all those fixes I told you don't work, maybe give a try to a vpn or something

though i've heard in some cases, vpns cause such errors...

[grim nest]

Getting that error at my ultra acc from 2 days

Im too tired how to fix that is this a glitch?

@brittle bloom boss help plz

[clear cipher]

Well, you might think it is not true, but after I understood this and started applying it to my prompts, I never had this type of error again or, as you said, a bug.

I do not have this type of problem anymore because I understood this and apply it to my prompts. But you can think it is not true and keep believing it is a bug and stay stuck in the same place.

[brittle bloom]

Sorry I'm not mod or else #1238621511401734254 it would be helpful

[violet ruin]

Hey everyone, I wanted to please ask does anyone know in flow VEO3 what is the difference between frames and ingredients I’ve been using frames and yeah it comes it works thanks very well when I give it my prompt and then I start image but I’m not sure of ingredients what is that

[candid bay]

it's really not so serious bro, so no need to get upset. frankly i rarely have any sort of issues or I find it easy to circumvent them. but this is also based on what the staff said, not what you believe

your specific solution maybe helps with the filters, but has nothing to do with the bugs

but "unusual activity" errors has nothing to do with copyright or celebs lol. come on

[wet horizon]

https://labs.google/fx/tools/flow/shared/video/ed91b46c-efa3-4383-8b5b-af1ad391b4d7

There may be some issues but overall the platform's cool

[exotic shell]

definitely not. happens to me as well and i never did.

[brittle bloom]

for i see just reprompt and retry if i saw ununsual activity

[inner gulch]

Ever since I received this warning a week ago, everything has been more sensitive. Now I can only generate a few images before I get stopped. Any reason why?

[tacit aspen]

Yes, feeling this... Mother's Day was an interesting choice.

[fading dirge]

the prominent people filter is way overtuned in flow

[candid bay]

it really is. other filters too

[icy halo]

hello

[pulsar coral]

Allright, to all those getting the "Unusual activity error", this happens because you're generating images/videos too fast.

The limit right now is 8 generations at once per 2 minutes, as in it's better to wait till all your images/videos finish to generate or fail to do so and 2 minutes before requesting more.

If you go faster than that you'll get the "You're going too fast" error message and if you try to generate more sooner than that like twice then you'll get the "Unusual activity" error.

To fix this error you need to clear cookies, it can be every cookie or just every cookie related to labs and Google, and so on if it happens to you again.

If after that you still get the errors you do need to wait a couple minutes after clearing cookies. And if after 5/10 minutes trying you still get the "You're going too quick" message, that just means your account hit the daily limit for image/video generation attempts.

So in this case If you're generating images try swapping models from Nano Banana 2 to Pro and vice versa, you might have some more image attempts there. Otherwise yeah the account is dry for the day, it will reset not exactly after 24 hours but like 30 mins after you generated your last image/video the day prior whatever the hour was.

According to your tier plan from free to Ultra you get a set daily amount of image/video generation attempts, no matter if you still have credits or not or if you generated an image or not, the failed attempts count towards the limit. It's around 200 attempts for free then double that for each tier, I think Ultra has many more than that.

This might change by monday/next week or not according to daily resource availability, they're always changing it but right now yeah it's very limited.

And Nano Banana Pro increased it's free daily limit a lot, used to be just like 10 a day, yesterday I got easily around 60 Nano Banana Pro images... while still having those bugs of course.

[pulsar coral]

That said, all those other error messages do count as bugs since they're not tuned in a way that would help users understand what's going on. Like Whisk used to have a concurrent image generation limit of 5 at a time and would never bother people with these issues while still giving better results.

[rigid nest]

Forgot to add from and to dates 😉

[proven onyx]

yes. Useless

[crisp basalt]

Oh

Omg the video is better

[pulsar coral]

Two different characters seem to be talking where intended, so I get it

[rough halo]

does anyone know of a solution to the unusual activity bug? i mean a permanent one. currently I just open a new tab and go back to my work copying my prompt from the old tab but its very disruptive

[crisp basalt]

Nothing you can do, it’s server backend issue

[pulsar coral]

Also try all the stuff I wrote like 8 replies above. Yeah there's no permanent fix to it but you can somewhat adjust to deal with it.

[ornate pine]

want a way that users can close the paywall do I create this on revenue app or do I create this on googleai studio where i am creating my aoo

[rough halo]

ty. i know of the cookies part. Even just opening in a new tab helps. Transfering over to a new browser altogether makes it last longer. Incognito also. But its still cat and mouse and is just frustrating

especially when my prompt is long and has "@" tags in it

[pulsar coral]

I also saw a user said she blocked cookies in a separate browser or separate browser account and that dealt with it, but haven't tried it yet really.

[rough halo]

it deals with it until it catches up with you. then its just the same. i dont know why it links to help center when help center has nothing about it.

and im not even using any vpn.

but im not surprised. given they rarely update their website. only recently did their changelogs online reflect up-to-date updates. before if you checked their webpage it was months behind the changelogs on discord

i suppose they have their staff working on the tool's upgrades itself and nobody is really dedicated to addressing existing bugs unless they are really breaking the functionality. if there's a work-around it can wait it seems

[pulsar coral]

So far they only update Flow to add bugs or remove websites and functionalities that were working fine enough before and that no one wanted gone.

Tho yeah maybe VEO might have gotten a little better now, gonna try after the daily reset.

[rough halo]

im not holding my breath lol

hope you're not disappointed

[pulsar coral]

It does seem NB Pro has become a little more uncensored than NB2 for some reason... which would be devilish considering that gotta pay to use a higher NB Pro limit.

But it's still not as good as as it was last month, so still not worth paying really.

It used to be way more than censored enough last month anyway, all that the recent exaggerated blocks did is cripple the model for simple things NB1 can still do much better.

[brittle bloom]

Looks like too quick instant Generation just one by one

[void crater]

is something wrong with flow?? in morning i was using and for everytime it shows ooops something went wrong message and now i tried again and it is showing prompt must be provided. but I already provided prompt

[crisp basalt]

try again

and provide prompt

jk flow is broken yes

[void crater]

i am trying since yesterday. it was giving me error messages so i stopped . and now i am trying again and it is giving prompt must be provided.

[void crater]

ok

[tribal star]

how to avoid veo watermark on videos?

[vocal kayak]

you can use content aware fill of aftereffects else you've to purchase the pro version of flow

[violet ruin]

Hey everyone, I want to know whether VO 3.1 response better with Jason prompt or just normal more sentence

[gloomy verge]

Hello. I want to ask. Imagen 4 doesn´t understand my prompt? Because, when I written to prompt: "Make image, where shower will be turned off", Imagen 4 created images of shower, but shower was turned on and working (water dropping), but when I used NB 2, it created images of turned off shower. Why? I must use other way of prompt for better result for Imagen 4? I written it bad, when Imagen 4 created turned on shower, instead turned off? Or where can be problem?

[pulsar coral]

Imagen4 seems to have been broken unnecessarily when they were trying to censor it, I can surely confirm it's blocking completely safe prompts. Even worse when it thinks a reference image barely resembles a woman.

The pics don't even need to be unsafe, it's almost like a Netflix executive was in charge of it.

[gloomy verge]

But in my opinion Imagen 4 has better graphic. NB 2 images are so bad, grainy, no? You doesn´t have this feeling, me yes.

[pulsar coral]

But sometimes Imagen 4 would still let short prompts pass, the longer the prompt is the more likely it will break.

It's not at all as good as Nano Banana (Gemini 2.5 flash image) is, it's like they tried to make it like it but epically failed at it.

[pulsar coral]

Yes you're not wrong about it, that's because Imagen4 is closer to what Whisk's Nano Banana (Gemini 2.5 flash image) is. Or was since it's closed now, but the model still exists and no one's planning it to retire it anytime soon.

We all did get confirmation on it from Vertex couple days ago.

It's just Flow that has it disabled who knows why.

[gloomy verge]

What is disabled?

[pulsar coral]

The image model from Whisk's precision mode, Nano Banana (Gemini 2.5 flash image), it was disabled at Flow some months ago.

Man too early for me gotta go live, be back later.

[gloomy verge]

Yes, thanks for help.

[torpid cedar]

Definitely interested in the next level of Google video models. I've tested each product in the past even Genie.

[exotic shell]

who cancelled his ultra subscription already?

[wicked crater]

Why?

[sudden vortex]

I can't say I'm impressed... I had high hopes they'll match/surpass seedance

[sudden vortex]

what can you say about this video⁠?
is it actually the next model?

[torpid cedar]

Not sure

[sudden vortex]

that's good news

[exotic shell]

3.1 not free anymore

[urban solstice]

hi there

[sage otter]

There is still no way to simply seartch your projects in flow?

[calm spoke]

i have an issue with flow
i can't generate anything, it stops at 99% with this message

[burnt oriole]

is someone having an issue where sending a prompt or pasting an image inside of a collection is sending it outside the collection?

[split lintel]

That's a known bug and the dev team seems already been working on it... pls see "labs-bugs-feedback" (unfortunately, I cannot post links here).

[burnt oriole]

yea a few minutes ago I posted my bugs and FRs combined together in there

[rigid nest]

Yes, it's been reported before

[burnt oriole]

flow feels a lot buggy today. Some buttons aren't working and also sometimes I can't Rename or delete anything

[rigid nest]

It's been buggy for last 1.5 months 🙂
You may check the #1238621511401734254 channel to see if your problems have already been reported.

[burnt oriole]

ooh i see

[pulsar coral]

...it's amazing how so many things are buggy yet somehow they'd still find a way to make it worse.

Then again most of it more often comes from how the Nano Banana 2/Pro models have been "nerfed" lately which visibly wasn't done by someone very conscious and caused more bugs than censorship... I actually have no problems to still generate whatever I want

I bet many others don't have probs either, so it was pretty much done for nothing, it should just be rolled back to previous version and whoever changed it like this... yeah they should do something not very good to him.

Then once again apparently no one who's in charge of the Nano Banana models is at Flow (?) so begs the question then why every Whisk user was migrated to here, if anything they shoulda sent us to the Gemini Discord.

[rigid nest]

I just feel that we have reached a point where feedback is just a formality.

So I do not report it anymore. Sad, but there is truth in it.
Immense potential for this tool, but hey we have to cater to Gemini Pro users only. Flow is a weekend DIY project.

[chrome imp]

Yeah i think Nano Banana with Veo could be such a powerhouse, but there a a lot of things that they could improve, coming from google

I still use Google AI Suite becuse it gives you the most amount of videos/images per month for what youre paying, but right now, on Video and Image generation they are lacking

[spiral python]

When update veo 4

[rigid nest]

https://gemini.google.com/share/7d5dc678c80a - Gemini Veo4 Omni model is coming, I have heard. Is this true?

[merry mango]

🐛 Bug with nested collections – images always generate on parent collection

Message:
Hi team, I'm experiencing a major bug with nested collections. Every time I generate an image inside a sub-collection, it gets saved to the parent collection instead. This happens consistently, no matter what I do. It makes it impossible to organize images properly when using nested folders. Please fix this – it's breaking my workflow. Thanks.

[lusty sinew]

Wow, more than 4 hours and still no Flow reset, these new limits are ridiculous. If I wanted a 24 hour window I'd move back to Gemini

[remote bay]

Thanks for the feedback! Please send any feedback or bugs reports to ⁠⁠#1238621511401734254. This is so we have it all organized and it won't get lost 😄

[pulsar coral]

The account's reset is around the time when you tried your last generation attempt yesterday. Like if it was 7 pm it'll be around 7:30 or 8 pm.

[lusty sinew]

It says 11:50pm according to my history, I've just never been locked out for more than a few hours

[pulsar coral]

That could be cause you didn't reach your daily limit like 10 hours ago, if you did just now yeah you're gonna have to wait more than 4 hours

It's better to be sure you used them all if you plan to workflow at the same hour the next day

Yes it's that cruel

[lusty sinew]

I hit it around 6:17PM

[pulsar coral]

Did you try swapping model

[lusty sinew]

That's the last time I got a successful NB2 gen

[lusty sinew]

I did but I really don't like NB Pro, it messes up faces

I cleared cookies too. I think I'm just at limit for the day but I assume it'll reset around 12:30

[pulsar coral]

Yeah all this is weird, but it will eventually reset for sure. It does need a timer, and it's indeed possible to prompt it to these gemini sites tbh.

[lusty sinew]

Do we know why they changed the amount of images we can do at a time? They're claiming high demand times but that was clearly never an issue before. And we still never got a great answer about this "prominent person" filter that's tuned WAY too high.

[pulsar coral]

It's still 8 generations at a time, and the prominent person error is still there. More likely it's Google's fault for sure since it seems NB2/Pro is locked onto some sort of image search mode for accuracy, that surely could be fixed by allowing users to turn it off manually or re-enabling Nano Banana (Gemini 2.5 flash image) here.

Wrigley did post the other day how it works with some screenshots from Vertex.

Like I knew it the whole time cause I recognized the wiki gallery images it was using to trace the faces of the characters I generate, but that did just confirm it.

[lusty sinew]

I would be less bothered if those errors didn't count toward my total generations for the day. I have a pro account so it would be nice if I didn't have to worry about it.

I know that before, I was hardly if ever reaching gen limits on flow, and even then it took only a few hours to reset.

[glad yacht]

Here is official documentation about Google Gemini models available in canvas mode!
(Imagen ⁴ and Nano Banana aka flash 2.5 image)

[unborn shale]

How do I get more veo credits without paying 😭 😭 😭 😭 😭 😭 😭 😭 😭

[hollow furnace]

many error!

[vague island]

high demand rn

[cobalt scroll]

Same here bro 😭
Flow’s been bugging out all day for me too. Random errors, black frames, failed generations… feels way worse whenever demand spikes.

[brittle bloom]

Just wait if high demand maybe hours, or later or refresh the art before refresh make sure Copy ur prompt in notepad

[cobalt scroll]

Nah honestly that doesn’t even fix it anymore 😭
I’ve waited hours, refreshed, copied prompts, switched browsers… it still randomly breaks again. Super unstable lately.

[opaque sand]

Even in “quality” mode, the video generation fails...

[brittle bloom]

Hmm if so try to report bugs in #1238621511401734254

[rigid nest]

Joining google DEV team 😉

[idle shale]

use multiple acc

[hollow furnace]

It works worse and worse every day, but the subscription price has increased

[modest terrace]

I had 1000 credit for one month
Fun fact 😂 I didn't pay 💰 a single penny

[unborn shale]

I get it every month I have Ai pro plan. 😭🤣 on two accounts still it’s less

[candid bay]

pro plan on 2 accounts? damn

[unborn shale]

1 I got from my network service provider free for 18 months

[exotic shell]

since the downgrade of the ultra plans it might be the best way to go for, having 20 gmail accounts. each one gives 50 coins a day. thats 1000 coins a day. the price for 3.1 fast generations are the same, for ultra and free accounts.

[unreal viper]

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

[modest terrace]

😎 if I want i can get in all my 4 gmail account but for now I had 2

[frigid pier]

When Veo 4?

[modest terrace]

For literally $4, they’re giving:
Google Gemini 3.1 Pro
5TB Google One Cloud Storage
Flow
NotebookLM
Jules
Like… this bundle’s total value is around $370–$400 USD and they’re casually attaching it to a recharge plan 😎

[shell frigate]

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

[unborn shale]

add a greenscreen

[peak sluice]

WHY😭

[pulsar coral]

Now without Whisk the Google subscription is worthless anyway, which is funny cause one can still create a similar app with the canvas mode from Gemini... but they don't tell that to anybody so it's the same.

[peak sluice]

how to fix it🥹

[pulsar coral]

Clear cookies, specially related to labs and Google, when it works again don't generate more than 8 images/videos at once per 2 minutes after the last one generates or fails to generate. Or this will keep happennig to you and you'll have to fix it the same way.

[peak sluice]

Thanks 🙏

[violet ruin]

Hello everyone you have a video and I’m trying to do a UGC video be honest with me what are your thoughts on it and how can I improve on it

[violet ruin]

NO TEXT OVERLAYS NO SUBTITLES NO CAPTIONS NO LETTERS NO WORDS NO TYPOGRAPHY NO ON-SCREEN GRAPHICS. NO BACKGROUND MUSIC NO SCORE NO AMBIENT MUSIC NO SOUNDTRACK. [BASE IMAGE: Lily start frame]

Young Black woman early-mid 30s, medium brown skin, symmetrical face, pronounced cheekbones, full lips, expressive brown eyes, extremely close-cropped buzz cut natural raw hair growth, bare ears no jewelry, clean skin visible pores minimal makeup, navy sweater with royal blue stripes over white collared shirt, delicate gold rings thin gold necklace. Seated at warm walnut table lower third horizontal, neutral beige walls white door right green potted palm behind, soft diffused front-facing natural window light.

Static tripod iPhone selfie, perfectly straight-on frontal axis zero tilt/skew, medium shot subject seated further back generous negative space above/around shoulders full upper body visible, face-priority focus eyes sharpest, iPhone 15 Pro front camera sim 26mm f/2.2 ISO Auto 800-1200 1/60s, organic luminance noise shadows, computational portrait bokeh uneven blur.

Raw iPhone selfie candid unfiltered "shot 5 minutes ago" vibe, visible pores natural skin texture slight iOS smoothing, authentic phone dynamic range crushed blacks blown highlights, cool-neutral palette, clean frame no on-screen graphics no pop-up text.

Dialogue: "Just three simple steps that actually deliver lasting results. Specifically, if you want the bare minimum" | 14 words ~5.8s @145 WPM | Downward inflection on "results" 0.3s pause, slight pitch lift on "Specifically," staccato rhythm calm authoritative tone crisp consonants.

Action: Hands rest flat on walnut table direct eye contact slight forward lean on hook, on "Specifically" right index finger extends toward lens head tilts ~5° right eyebrows lift slightly breath out matches pitch lift, natural micro-movements slight head tilt eyebrow animation finger extension lagging audio by 0.2s organic blink rate.

[pulsar coral]

The limit for image generation is always changing with daily demand, then if you clear cookies and you still get the "You're requesting generations too fast" error after like 5 minutes trying, that just means your account is empty for the day, you gotta come back tomorrow around the same hour you generated something today, a little later is better. The failed attempts count towards your daily generation limit, it's around 200 for free accounts and more around double that for each tier, could be more for Ultra.

Man this looks like what we were testing. And looks great in the video tbh.

But yeah whatever the characters need to do or not seems to work better as instructions at top of the prompt. Veo is a little too creative,

I forgot to put it it the other day, but the command "Without any creativity freedom" in general deals with the randomness from VEO and the Nano Banana models

[violet ruin]

Thank you very much yeah we’ve been testing on it and then I’ve just been iterating on it I appreciate it. 🙏

[pulsar coral]

This is so extended and looks real... realism is what people have said to like from VEO

[violet ruin]

Do you feel if the lighting is off as in like it still looks fake or is it so good yeah I’m still doing a lot of Disney as we are working on it but our lady take me very long but I’m just improving on my prompt engineering

[pulsar coral]

Yeah this totally looks real, would take someone who's really into AI zooming into every detail to notice it's not

Which is not everybody, worse for a random add

[violet ruin]

Okay yeah I’ll keep on working

[pulsar coral]

For example if an angle changes at the start it can be locked as "camera must be locked to same angle from reference image without any creativity freedom"

it can still be stubborn but with that it's way less

[violet ruin]

Understood, camera tends to change so I will add into the prompt

[pulsar coral]

It's not just for the camera tho, can be tossed into anything that goes random, also works for Nano's images

... tho yeah yelling to Veo at the start of the prompt does show to be working, that's funny

[violet ruin]

Okay, I also forgot to mention apologies that I feel as though she doesn’t feel itchy enough or tonality is not also enough but her not like smiling a lot how can I make a sound more engaging and also a little bit of fast
Pace in speaking

[pulsar coral]

You might wanna change that on the reference image first, it's not such a good idea to start with it at the video. Tho the first prompt you sent me does have pointers about sound pitch and skin tones, you could add some of that back.

The one Gemini made with all the little emojis and symbols

[violet ruin]

Okay 🫡

[pulsar coral]

In that case Gemini can indeed be good to tell you how to do it, but gotta remove the extra clutter and emojis it sends

[crisp basalt]

so....

youtube already launched "ai avatars" (its like openai sora2 cameo/character) , are we gonna be expecting to see this in flow? (assuming you guys will add it along with omni/agent)

[stuck adder]

might we be getting veo 4 soon?

[crisp basalt]

christmas came early or something 😭

[stuck adder]

Nice you got it???

This from the webiteapp or the mobile app?

[meager edge]

Is VEO 4 out

[stone garden]

Is there any plans to actually give us a timer to that thing? I'm tired of being blocked after launching 8 generations consistently.

And it would be nice to know which rules is 'thought' to be broken too.

Especially since launching two generations and one goes through perfectly while the other gets this 'broke the rules' error

[delicate flame]

i never once signed up for any membership with you guys only a free account and connected my google account and you somehow subscribed me to your pro account and charged my card when i never even put in my card details you fk* frauds

where the f*** are all the staff

[severe lily]

Keep clearing my cache like a 🤡

This is on Ultra and on NBP or NB2

[unreal solstice]

Did Fast (low priority) got removed for ULTRA?

[stuck adder]

who on earth is even using lite

[sudden vortex]

Let's see a video

[uneven sand]

Your fomrs suck big time... you gathering wrongful or not cmpete data

[split lintel]

Interesting but unfortunately, no chance for members of the EU... 😭

[unreal viper]

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

[main saddle]

hi Jose! the Form is optional, you do not need to fill it out.

[main saddle]

hey there! stay tuned for future opportunities! 🙏

[uneven sand]

I did because I'm interested I'm just sating thAT is poorly buiilt

[split lintel]

Will do...! Thx! 🙏😎👌

[lunar root]

At one point veo was bad a text, then got much better with fast. now lite is back to garbage gibberish text like they broke Veo. Any client face image uploaded is flagged and you have to do mouse and browser magic tricks to even get renders now without that useless "unusual activity" block on every and any render. THIS IS BAD FOR ULTRA USERS paying the high rate for quality but getting less. Process is now 100 times slower using Flow.

[exotic shell]

youre a very quick one. got announced a month ago and i came here almost daily to remind people that ultra accounts are useless soon 😄

[pulsar coral]

I'll try, like your poll about videos is missing Seedance and Wan... and adding many no one's using much atm

[main saddle]

passing this along to the crew who made the Form! thanks 🙂 🫡

[pulsar coral]

LTX is being used a lot since last month too.

[main summit]

Ongoing UXR's Panel: I can't sign up. Spain isn't in the list of countries.

[deft tree]

I use it to do minor edits

[kind dirge]

The annoying part is mostly went the image generation stuck on 99% and you you have to refresh the page to make it work again.

[lapis ridge]

Hello

[keen hinge]

is there a specific channel for feature requests? ... I would love to be able to send a full project to a folder on my google drive. Ideally with the option to sync a folder (bi-directional?).

[bold falcon]

We already have that. Once you have your project open, go up to the 3 dots, click Download, and choose My Drive is what I do, which means that my desktop mirrors my Google Drive, and they sync both ways. That's already part of Google Drive. Saves as a Zip File. Then use an automation to extract or manual.

[rugged drum]

Has anybody else been having issues with frequent failures due to "policies about prominent people", despite using novel characters?

I feel like they might have erred a bit too far on the side of caution there when it comes go incidental resemblance.

[bold falcon]

Yes, I tell it that it's me and then at least for that session I'm okay.

[keen hinge]

I entirely didn't realize that there was a project wide download option - that is worth a lot to me, thank you! I do not have "my drive" installed in that way and always go there through the web interface, but this solves a lot of issue for me anyway. Thank you! 🙂

[bold falcon]

You are quite welcome and glad it helped you out. Have a great day

[sterile fulcrum]

Excuse me...but I've been using flow just fine for weeks now, and suddenly I keep getting this "unusual activity" message again and again! Can someone tell me what causes it? I can't find answers.

[pulsar coral]

Clear cookies, specially related to labs and Google, when it works again don't generate more than 8 images/videos at once per 2 minutes after the last one generates or fails to generate. Or this will keep happening to you and you'll have to fix it the same way.

The limit for image generation is always changing with daily demand, then if you clear cookies and you still get the "You're requesting generations too fast" error after like 5 minutes trying, that just means your account is empty for the day, you gotta come back tomorrow around the same hour you generated something today, a little later is better.

The failed attempts count towards your daily generation limit, it's around 200 for free accounts and more around double that for each tier, could be more for Ultra.

[bold falcon]

Hi, You have to be more specific, what plan are you on, what are you trying to do, what do your prompts look like, how many iterations, images or video, and within images or video, which model are you choosing? How often does it happen? ETC. Otherwise, honestly, everyone is just guessing.

[sterile fulcrum]

Thanks for replying! I don't have any plan, just the free one. It was happening once in a while at first, and now it's happening every time- but since you mentioned my prompt, I tried changing it a little, and it finally worked! I didn't think that would fix it since I used the same prompt for my last pictures and it worked fine.

Okay, some of them are still getting the error, so here's the prompt I'm using: "young woman flying down the street in a tiny personal-size hot air balloon. She's sitting on her knees so that she's only visible from the waist up"

I'm not trying to generate lots and lots of images, and I've never gotten the error about doing that or generating too many per day. I'm way under 200 today for sure!

And I'm just trying to make pictures. I've tried all of the models too.

[pulsar coral]

It's just the speed on which you're generating, Flow doesn't give clear limits when one should stop before being flagged as spamming.

And the error started just few days ago, before that you could generate 25 images at once without issue.

[sterile fulcrum]

Okay, I don't think I'm generating fast at all, but I guess the speed I need is kind of random?

[bold falcon]

Mega has good advice. I have ULTRA, and it has been happening to me every time I use flow. 7-8 videos a month. What I have learned is only try 1 or 2 iterations. Sometimes only 1 if I'm doing 8 second videos because I do avatars. The infographics I have no problem. I always create 2. I use fast but I have 25,000 credits a month so I don't worry about that. And then I wait until those images are generated. Then I click download, choose my resolution, and then I post the next 1

[sterile fulcrum]

You mean I should only use the same prompt once or twice?

[bold falcon]

. So I literally have to wait for every test to complete before starting the next 1 that cuts down on the errors. But they still happen every time. I always get prominent people for myself as well as as unusual activity. For example, today I just did 16 separate avatar videos of myself and I probably had 40 to 45 errors for unusual activity and prominent people and failures. and my prompts are solid because I've updated or I'm sorry I've uploaded both the image guide and the video guide to my prompt maker in Claude

[sterile fulcrum]

Oh no, I didn't realize it was so bad for everyone! It was working fine for me until yesterday.

[bold falcon]

No for my videos my prompts are pretty much always the same because I've already created several 100 B-rolls in prior sessions. I still love everything it does though.

[sterile fulcrum]

Maybe yours is working because you're paying for a plan! I think I've only gotten one image done in the last hour.

[bold falcon]

For example, this is what I've done so far today for my next video on co-work within Claude, and that's actually taken me a couple hours.

If you do the pro plan I think you get the first month free and that gets you a 1000 credits and then just start keeping track of how many credits it takes for each thing that you do and then just make a judgment call on how many you need to do. In my case I code as well so I use anti-gravity so I need the ultra account but to me honestly 250 isn't that bad but I'm also in my third month of 124.95 for the first 3 months. I haven't decided at the end of this month whether I'm going to continue or try to drop down the pro.

[sterile fulcrum]

Well, since I'm not generating too fast or anything, I think all I can do is wait for tomorrow and see if it's any better! Thanks again for trying to help.

[bold falcon]

Oh, you're welcome. I have some flow videos on my YouTube channel, so you might find some decent tips there as well.

[sterile fulcrum]

Oh, okay!! Is it your username?

I'll look it up.

[bold falcon]

yep

Also up at the top right of Discord is the search, so you can do it a couple ways if you see people that are continually answering questions, type their username up there and you can see what other questions they've answered. And of course, searching for a topic that really helped me out when I first got here.

[spare palm]

Some one had posted a link to a prompt generator from an image. I forgot to bookmark it and can't find it now. It was better than any I have been able to find myself. Anyone able to help me out TIA

[bold falcon]

I can only tell you what I did. If you're using Gemini you can create a gem. Do a search and download the Nano Banana Pro prompt guide as well as the VEO 3.1 prompt guide from Google. Upload those as knowledge for your gem and then all you have to do is describe it and it will follow the rules that Google wants you to use to make your stuff.

[pulsar coral]

It might be the one Wrigley prompted, best solution I think we have atm, you need a Gemini account https://gemini.google.com/gem/176dRicJ74XR85ux0WZU8SlBk2EXQPkpG?usp=sharing

[bold falcon]

One of Mine, But this is specific.

[pulsar coral]

But yeah best thing is probably to try to prompt one yourself at Gemini's canvas mode like that

you don't need to know any programming or anything, it's like prompting for an image, but it creates like a little website/app instead

[bold falcon]

I agree those are great for folks starting out I actually have Claude write all my prompts. For example, here's 1 of my avatar ones for my opening. For my video uploading tomorrow, I have 33 assets total. Plus voiceovers I have to do inside DaVinci Resolve Studio. And because I'm trying to do things at scale, having structure is a requirement for my pipeline that eventually I plan to have completely automated LOL. Claude and antigravity for the win

[pulsar coral]

That one looks great man

[manic depot]

Hi there

[gray pilot]

Hey, everyone, Google Flow(Beta)(Early Access) app is available, now.

[vestal shore]

Can you share us the link to it...

[brittle bloom]

I did bro

[cerulean bone]

Halo @unreal viper , Im interesting with UXRs Panels program on latest #labs-announcements・📣

But unfortunately I couldn't find my region Indonesia in demographic. So its mean I can't join with it?
Or any solution about it?

[unreal viper]

hey! I've escalated this to the team!

[cerulean bone]

Okay thankyou

[main saddle]

hey there! thanks for your interest 🙂 we’re working on expanding which regions we can do UXR in, please stay tuned for updates!

[stone garden]

No "France" either? Is it only for specific/targeted countries?

[desert berry]

Is it Flow V4 coming? or just a Flow V3.X?

[candid bay]

<@&1366831844502274180> this is triggering to eyes, all this flashiness

[cerulean bone]

Appreciate your response and await your further news 🙏

[exotic shell]

is it a new thing that it shows a generation as "queued" for more than 10 minutes?

ah, it failed like 15 minutes later.

Half an hour later, still no refund for my used credits.

[unique citrus]

We noticed some unusual activity. Please visit the Help Center for more information.It keeps doing this every time I try to generate a video.

[wide ether]

what's the story with all the 'prominent people' errors? I generate randmon images of people without using references, and then use those generated images for videos and the errors coming back stating 'prominent people' lol!

[candid bay]

i think filters are over tuned

[wide ether]

you mean the censors?

[candid bay]

same thing lol

[tardy ridge]

https://youtu.be/jozQfsExtbE?si=Rw6al57xwY7HvaC6 i made this with ingredients, seems like Omni will be an evolution of ingredients, better fidelity, this is imo close to what you'd see on an indy film, I'm looking forward to it.

[pliant zodiac]

Does anyone know how to use this indefinitely? I've tried on several accounts, but they're all locked.

[sage otter]

did they just change imagen 4...?

[arctic kelp]

im fine

[cerulean bone]

<@&1366831844502274180> can't click any button. register and sign in

[crimson garden]

last couple of days the NB is terrible (USELESS)

On ULTRA too.

[distant ferry]

veo lord 🤡

[uneven sand]

So now Flow fails but charges anyway despite saying that the generation is not going to be charged

[blissful abyss]

When is the flow android version going to come in full release!!

[pulsar coral]

If this is how Flow's AI image generation is going to be handled, then what should be discontinued isn’t the functionalities that actually were working perfectly before — it’s the excessive ideological woke pandering.

[unreal solstice]

Cant generate videos?

[solemn isle]

I have been trying to generate since yesterday it keeps on saying queued and failed

[unique citrus]

mee too

[obtuse flint]

what the hell happed. have they lowered the limit to use it and censored the model. its absolutely s**** to use.

[unique citrus]

veo3 is down in flow? i cant generate videos

[stone garden]

Still same problem as every single day, "asking for generations too fast".... Come on... How long do I even need to wait?!

[fallen lily]

Yup it’s failing i tried creating something but it was not

Now my 24k credit gonna waste 😕

[spring kite]

mee too

[fallen lily]

How much credits you got ?

[unique citrus]

How can a large company like Google not be able to maintain Veo3? It's always in high demand.

[unreal viper]

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

[fallen lily]

There are much more users than you think … the server is down i guess

[fallen lily]

Alrr Brodie

[arctic kelp]

not working

[unique citrus]

that's very bad

[spring kite]

23k

[severe lily]

this is a joke at this point.

[wide ether]

Have you tried changing your VPN location a bit ... I suspect it's related to VPN usage

[soft tiger]

I think it’s broke, I still get that every prompt I do

[astral fern]

No VPN, no other windows open with flow.. 🙁

[spring kite]

The VPN is not active.

[spring kite]

yea me too

[exotic shell]

noticed the same and mentioned it a few messages above.
generations went to "queued" for a long time (15 min +), then failed, never received the coins.

[exotic shell]

in general it seems to look like almost all generations are stuck on "queued"

[bleak agate]

I get a "unusual activity" message every time I click and create a video... How long do I have to wait? 🙁 ?

[wicked crater]

Takes too long, "The generation might be taking longer than expected. Please check again in a moment. You will not be charged for failed generations." but I got charged for a failed generation... 😭😭😭

[tulip pilot]

wazzup y'all

[severe lily]

This has been basically unusable for me since yesterday afternoon and for long stints the past few weeks. I'm paying for ultra and clearing my browser is NOT a viable solution as the issue returns within an hour.

[opaque sand]

Has NanobananaPRO crashed again? Only version 2 is working(((

[keen hinge]

right now pretty much everything seems to fail - is there some sort of status website that lets us know if this expected behavior?

[pulsar coral]

.... they seriously need to enable NB1 back, then they can leave NB2/Pro for the vultures.

[mighty bison]

I'm guessing "We notice some unusual activity" is just me running out of credits

[winged bolt]

Coinstant Failures....This prompt might violate our policies about generating prominent people. Please try a different prompt or send feedback.

[tacit knoll]

Is there other ultra subscribers that can't generate a thing? It's been almost a week now, or the unusual activity message or "you're generating too fast" from the first attempt. I'm not sure why am I paying 250 dollars a month anymore. If this persist for one more day, I'm closing all my 3 ultra accounts and moving my team to Kling. Absolutely unacceptable

[fossil python]

it's been month, but they don't care. our Ultra subscription is just mean we pay more for them, nothing else. they treat us the same with free users abusing the service.

[split shoal]

Anyone else experiencing Veo taking the credits but no output was made?

[acoustic grove]

@rain ivy 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!

[rain ivy]

ack! ymy drunker ginnberish isn't that bad, is it? perhaps, i guess it is! ack!

[bold falcon]

Okay, I have been frustrated over my last 3 videos with this. (3 days). I think I may finally have solved it, (for me anyway), and thought I would share with you. First, the caveat: I've only been at this a 2.5 months. I'm making my third video of this week. I do have the Google One Ultra account in my third month at the $124.95 level. I'll be keeping the $250 level.

1. I uploaded the VEO 3.1 prompt guide as well as the Nano Banana 2 prompt guide into my LLM.
2. I have had all kinds of unusual activity and prominent person warnings. The prominent person warning you have to put something in that references your image as the subject if you're using yourself which is what I do.
3. The Unusual Activity one turned out to be the tough one to crack. The prompt above has been working perfect for me. And this is the image I'm using with it for avatar creation.
4. Keep in mind I had been doing the lite and the lite low priority and multiple iterations getting a ton of failures, like many here.
5. Now, I am using VEO 3.1 fast for 10 credits each, only one iteration. (Let's hear it for 1 shot) And then I am upscaling that to 4K for 50 credits, so that's 60 credits per video. And it's working fantastic. That would be 416 Videos at 4k (8 sec) per month. I also cheat and put in approximately 12 normal speech seconds and then I go 50 % speed in editing so I'm getting 2 for 1 basically.
6. I also have my LLM not make prompts when I'm off screen and only give me the script if I'm running B-roll or a graphic or something or a screen recording. So typically 9 to 15 avatar videos per project.

CONTINUED: I do understand if you're on Pro that that's a heavy price but honestly the not have to jump around a whole bunch of LLMs anymore is very cool. So I am running only Ultra and Claude Pro. My entire content creation pipeline is built with Claude. I'm just running the videos and graphics through my Ultra.

So, if you have any questions that you're just curious about with doing things this way, the best advice I can give you is try 1 or 2 iterations only. At the highest you can afford for credit, let the image or video completely generate and then completely download or upscale and download before you send the next prompt. Do whatever you can even though it takes longer to not stack the processes and that has honestly worked great for me and I know it takes a lot of patience so I totally know exactly where we are all coming from.

[bold falcon]

So far tonight I've solved it with some more tightening of prompt and structure and using VEO 3.1 fast for 10 credits on the video and then upscaling to 4K for another 50. But I'm only doing 1 iteration, so 60 credits each. And I cram in about 12 seconds of talking in text, and then in editing, I lower the speed to roughly 60%.

Here's an example of one I did earlier Cramming more text in. 8 seconds. That's not my voice. I use voice convert in DaVinci Resolve Studio post to convert it over to me. I'll try anything to save credits lol

[crisp basalt]

they havent .. update the website for almost a week now

i wonder if theyre getting ready for google i/o + .. flow agent / omni model

[bold falcon]

Absolutely they are. Actually, I have a video coming out about that tomorrow morning that I'm working on right now. I'm pretty excited about what's coming.

[lilac hearth]

Getting hit by the prominent people problem this morning too. I generate a picture with banana 2 with just a simple a beautiful young woman description. I then animate it and get the prominent people problem. This is weird. Even when I that the picture should be in anime style it is the same. And when I get over the prominent people problem I run into failed to generate due minor ones. When this is the filter training for veo4 Then I think I should go to stick man animations hoping that this will work 🙂

[wide moon]

When the iOS app coming?

[tame falcon]

The censorship is getting stricter and stricter. I remember last October, it was almost impossible to generate any kind of image.

[azure peak]

guys wanna ask anyone use this ai before? and what do you think of this ai?

https://www.lovart.ai

[whole escarp]

I was CPP when they had a program. They are nice people. Their AI is a little loopy but if you give it the right prompts it's very good. You can generate a lot of stuff if you get a plan when they have a good deal like unlimited photos. Personally I would use magnific.

HELP! Hi it says my images transferred but under flow I only see videos. If I change it to imagefx it takes me to a welcome to flow page.

[ember pawn]

Guys, I still have the "queued" issue, how can I solve it? nothing works since yesterday

[late totem]

@bold falcon

[shell rover]

whats the issue ?

[ember pawn]

Every video I'm trying to make goes in que

[shell rover]

what app it is ?

[ember pawn]

Flow

[ember pawn]

And then it goes like this

[sick sparrow]

image generation is down it seems

[shell rover]

servers are overloaded

[ember pawn]

since yesterday? i'm getting the same error since yesterday

[shell rover]

maybe its just timing

or maybe server will be down for long time

[ember pawn]

😫 😭

[shell rover]

chill it happens

[ember pawn]

I know, thank you for your response ❤️

[shell rover]

check your DM

[tropic laurel]

servers are fried?

[candid bay]

yup

[high plume]

anyone else getting this error

[compact wedge]

Are you blind?

[high plume]

no

[shell rover]

everyone

[compact wedge]

Just wait until the service is restored.

[shell rover]

@ember pawn @tropic laurel asked same questions

[frosty niche]

Anyone else getting strange outputs this morning?

[frosty niche]

Make sure to log these bugs if you come across them, it helps the team see the whole layout when users see these outputs. #1238621511401734254
Thx!

[unique citrus]

If you don't want any more free users, then remove them immediately, because it's impossible to do anything, it just keeps giving these ridiculous errors.

[wide ether]

Yes .... all generations are getting that error .... I'd rather get that error though is that's the true issue then those other random errors

[cerulean bone]

any update sir?

[wide ether]

Yeah the platform is totally down now

like full outage

[frosty niche]

https://tenor.com/view/robot-plug-unplug-arm-kys-death-gif-11213717263814528707

[wide ether]

Seems like a DDoS .... anybody from google in here to comment?

It's really not working now .... like DDoS not working

[neon raven]

since when is it not working?

its been a few hours? or more?

it was working fine for me yesterday

today i am not able to generate any video at all

[candid bay]

don't need to spam and ping everyone...

plus be nice

<@&1366831844502274180> he's being rude and disrespectful

well I'm more active than you, for starters

I know how to behave, also, unlike you

[frosty niche]

Woah, simmer down now.
Make sure to read up on the #start-here・📍 & #rules-and-guidelines・📋 to make sure your up-to-date with how things work here.

Oh @acoustic grove

[idle pasture]

Whos that pokemon

[frosty niche]

Or @pallid violet

[candid bay]

also, this is an official Google discord server, not a place to act like you are at a street corner

[idle pasture]

Its another pokemon!

[pallid violet]

Hey, thank you for pings, we are handling it!

Please also keep in mind that Moderators are here to handle server side issues and we aren't able to help you directly with Google product related support. If you have any issue, please post about it in #1238621511401734254 and Google team will look into it.

[stone garden]

Is there some specific problem today?

[teal phoenix]

Yeah

[frosty niche]

https://tenor.com/view/mosquito-giant-gif-14169911

[pulsar coral]

....gotta say, before buying any product or subscribing to any kind of service, next time gotta read customer reviews at whatever site that has them... because at this point yeah if there's basically horror stories there why even bother.

[modest terrace]

How can i get extra credit 😓

[river rapids]

yes same

[stone garden]

Well, you can still generate a wall apparently

[somber needle]

google, did you try to turn off and on whisk 🤭

[river rapids]

very bad

[pulsar coral]

Flow wasn't working so bad like a month ago, it's the recent changes they did to the Nano Banana 2/Pro and VEO models which is causing all these bugs, it's not even really server probs, cause yeah I can also generate "walls" on my end without issues.

This needs an emergency rollback to previous version asap

[true hare]

For me Whisk was so better 🙁

[river rapids]

The wall generation works pretty well...

[stone garden]

I tried to check if everything was down by asking for a wall but besides that, all my generations of 'worth' requiring elements are not going through today, yeh. I agree with you that it's been working less and less lately too in the past month yeh.

[pulsar coral]

They ignored every feedback about Whisk too... it's not only the models that need a fix, probably some staff with more empathy would be necessary

I mean I thought Google would be the first company to run psychological empathy tests before hiring staff

[stone garden]

Nah, I wouldn't take anger on staff, people just be doing their jobs.

Beyond that tho, I'd love to be able to use the product correctly x) It's actually quite insane that Nano is working better when using another website/platform/intermdiary (whatever it's called) than Google's.

[river rapids]

Nano Banana on Gemini is working. The problem is actually with Flow

[pulsar coral]

Idk maybe not the ones around here, but I can't imagine someone didn't pass the feedback to higher ups... someone's visibly not taking the paying customer's in mind.

[stone garden]

True it does, I forgot about that one.
Unfortunately, it does not have as many options as in Flow to correctly generate the sizes of images and such