partner-models

[glass cradle]

This channel is for discussing partner models in Firefly. Feel free to discuss models that are already accessible via Firefly as well as ones you'd like to see added. For a comprehensive list of partner models available in Adobe products and breakdown of their credit consumption, check out this helpx article

[analog ginkgo]

is there any one who runs api for firefly luma ray-3 ?

[small plover]

Luma ray 3 is an partner brand model not from firefly officially so no there's no api service related to that via firefly

[small plover]

Any chances of getting ( Seadream 4 / Qwen Image / Reve 1 Fast ) images model to be added ?

[agile elm]

Nano banana pro?

[small plover]

Already added

[agile elm]

Add BANANA 4K generation! 🍌 Its only 1k now.

[echo fossil]

I chose a 2:3 aspect ratio, but nano banana Pro generates 768 x 1376 images which are in 9:16 aspect ratio. Other ratios of both banana versions are wrong too.

[odd heath]

Yeah... that's sux a lot.

[narrow cedar]

@odd heath I spoke with the team yesterday and they are aware and looking into it.

[odd heath]

Additionally, will we have (at some point) an option to select a specified element/area to be changed in the Edit option - like a brush in Midjournal? or just simply draw an instruction on the photo?

[ashen prism]

It was a sneak at MAX I believe, but they usually don’t say when it’s gonna be released before it is out

[narrow cedar]

Even if we know, we are not allowed to say. NDA and all that good stuff.

But make sure to post your requests here and to the Adobe forums so Adobe is aware of what users would like.

https://www.adobe.com/community

[static fog]

One way to go around it is to use a reference image cropped to your desired output aspect ratio

[rich quartz]

add seedream pls

[dreamy pecan]

in here to see nano banana pro native resolution solved, esp for paying users....

[calm shuttle]

I gave Nano Banana Pro this prompt: "a high quality watercolor illustration for the chapter "The Elephant's Child" in Rudyard Kipling's Just So Stories." Here's what it generated:

Those who know the story will see several wrong details, but it got the correct story, the correct key scene from that story, the main characters (the young elephant, the crocodile, and the rock python), and a reasonable representation of the tropical river setting.

After a couple of rounds of fine tuning to remove extraneous text and fix the errors, I ended up with a good, usable image:

[wind pollen]

[true prism]

When seedream is coming?

It's so amazing model

[frank nimbus]

still no sign of Flux 2?

[wind geode]

why nano banana pro is not generating 2k res in firefly ?

[finite cosmos]

#partner-models message

[small plover]

What's your team thought about adding flux 2 series ? It's been many days flux 2 released almost every other platform added it's api still firefly lacking in adding the APIs.

[narrow cedar]

@small plover I have not heard. I will bring it up next meeting.

[small plover]

Yeah pls

[serene swift]

I have a trial subscription and am extremely disappointed. If nothing changes soon, I will cancel it again. As a pro user, I need 4K. I like 1K previews but finally it must be 4K.

[small plover]

They haven't added flux 2 too yet

[rich quartz]

Why nano banana is creating low resolution images?

[tawny laurel]

Team is looking into it per last I heard

[acoustic glacier]

@rich quartz Most likely, they will do so when the free generation period ends.

[rich quartz]

Oooo I see

[acoustic glacier]

It's a bit glitchy right now, and loads of people use it for free, but like a lot of pros, I wait the 4K quality. I have to use upscale.

[swift steppe]

Firefly NBP currently only supports 1K images.

[dusk portal]

i made this

i love beauty of firefly

i love make ideas real

[fleet delta]

This Santa and Rudolph were my creations and then I put ‘em into Firefly video… Not too shabby

[fresh peak]

#ai-video kk

[finite cosmos]

@fresh peak firefly is not a discord bot, please visit https://firefly.adobe.com/ where you can generate

[finite cosmos]

sorry but I can't translate this so I've removed it out of caution @sullen sigil

[quick shadow]

This is mission impossible on high level and the traffic sign is correct, in this case i think Nano Banana 3 taked help from Google Maps to solve traffic signs, that is the only explanation I can see to get this result. IMG 1 Original IMG 2 Nano Banana 3 Pro. I can only say WOW to this result. I made this in Photshop 2026

[calm shuttle]

That is quite astonishing. I would be surprised if Gemini didn't pull information from Google Maps, but even so, this is an unexpected consequence.

[quick shadow]

This is a good example then AI can do something that you can't do manually without a lot of houer or even weeks work.

[calm shuttle]

This is something that would be basically impossible without an actual reshoot of at least the view from the vehicle.

[warm current]

anyone else having issue with Nano Banana pro using previous reference images instead of the the current of whavever new image you want it to use

[narrow cedar]

@warm current This discord is for Adobe Firefly. Not 3rd party issues. Please contact the other models for their problems.

[warm current]

this is using FIREFLY WITH NANO ....

i dont have the issue when i use with gemini

only with Firefly

therefore why would i contact them?

[narrow cedar]

@warm current Adobe has no influence over Gemini. It is not an Adobe app. Its model is just part of the Firefly ecosystem.

[warm current]

my issue is when i use FIREFLY. NOT GEMINI.

[quick shadow]

As long Adobe include Partners models in Firefly or other apps, questions about or answerers about problem with this apps, has its place in this serve as long they is included in Adobe's ecosystem has there place on this server, or in the Adobe servers in Discord, it's same as you ask a questions about Fresco in Photoshop server. I can agree with you if it a technical question that Adobe not have influence over you ask the 3rd party direct, but the user can then be directed there.

Sometimes can a user have answer or tips with how to solve or get around the problem and you will never see it, if you answer the 3rd party direct. My content of this is that it shall never be wrong to post a question or answerer to the community on Adobe's Discord servers as long is on the right channel ,even it's only part of Adobe's ecosystem.

[warm current]

well said. thank you 🙂

I even posted this under the "partner-models" channel.... lol

[finite cosmos]

the partner models can be slow to load in firefly, you need to give it a minute when you remove and add a new reference to make sure you've actually done so before hitting generate, you also need to wait a minute for the image to start generating, as its pretty easy to hit it multiple times thinking it's not working

[warm current]

i'm very familiar with how to use it. it's like it's reusing a previous seed or cache isnt clearing... very odd and frustrating

[narrow cedar]

@warm current As you stated you had issues with the Gemini model in Firefly. #partner-models message

If you have questions about using 3rd party models in Firefly, we are more than happy to help. But if you have issues with 3rd party models directly, Adobe has nothing to do with these 3rd party models. That are not Adobe.

[warm current]

my issue only occurs when I use them thru Firefly. Doesn't happen when I go to Gemini directly. But heaven forbid I check on here to see if anyone else is having the same issue when using Firefly instead of using Gemini directly

Google isnt going to help me because I am using their product thru a 3rd party vendor.

[swift steppe]

I've had it completely ignore my reference photos. And I've experienced that using multiple 3rd party surfaces when generating with Gemini; not just the Adobe Firefly surface. I've not had it use reference images from a previous job. Not sure but this sounds like a browser caching issue. And when I've experienced odd behaviors, I've tried hard refreshes Shift Ctrl R and that usually solves it.

[quick shadow]

I have use Nano Banana pro 3 at about 25 of my images , all have been other of focus or shaky. My conclusion , of best use is, even if you only will correct spelling and format. Ex. "Correct the text on the image to xyz, capital letters ", select the hole image. Or if you only will change the subjects clothes, color or something, isolate the subject true select subject and then us Nano Banana pro 3 on the selection.
For me it worked 24 times of 25 without compromise the original image. So Nano Banana is very good on fix an correct mistake you done then you take the image. But a little warning you need to have a lot of credits, on my 20 latest image I have used 1.530 credits of 6000 from 20 December 2025 to now.

Most of the time I use Nano Banana for edit my own images not for render new one, so I use always reference image

[languid cipher]

Why does the listing of Partners models sorting is different in Firefly (left) than in Board (right)?

[finite cosmos]

probably because 2 different teams working on the different products chose to organize things differently

[gleaming bloom]

anyone else having issues with gemini 3 in photoshop? I keep getting "something went wrong". All other models work fine. Happens in latest photoshop release and latest beta - also in the bit earlier versions I had before updating just now. Yesterday all worked fine.

[finite cosmos]

yep, there's been intermittent issues with google's api, for adobe and all other services that use their api

[gleaming bloom]

seems to work fine on both freepik and krea.ai :/

[mystic beacon]

Is anyone successfully generating using Gemini 3 via the Firefly web app? The last 24 hours i keep receiving "Something went wrong. Please try again later."

[finite cosmos]

not currently, no but you can check the status here https://aistudio.google.com/status

[narrow cedar]

@cinder obsidian If you continue, you will be removed.

[rapid thunder]

Will Nano Banana pro forevere flop?

[narrow cedar]

@rapid thunder I guess it's up to google. It's just an API for Adobe.

[mighty knot]

i cant even get my generation history to load smh

[wraith pasture]

Hello
Am a graphic designer

[serene quartz]

new here, sorry if this has been answered before. is there a user guide that tells what each model is good for? like when is it better to use one over another? thanks!

[swift steppe]

There isn't anything like that, that I know of.

[finite cosmos]

there isn't an official guide, but things change so often/frequently that would be hard to maintain

[ashen prism]

Also some personal preference, like I only use gemini for fun, but for stuff that I’m gonna use in a project I stick to Firefly because of copyright and such

[swift steppe]

Using the Firefly model doesn't guarantee you copyright protection. Only actually filing for copyright and being accepted by the USPTO gives you actual copyright protection over work. Please do not equate what Adobe calls "commercially safe" with copyright protection. They are not the same.

[ashen prism]

Not mine, but misuse of others work

[swift steppe]

The way you stated it seemed to imply that Firefly gives you copyright protection. And this is a huge misconception that I keep hearing from people. Thus, I wanted to make sure no one got the wrong idea.

[latent lance]

why with any model I use, if I put in some reference images, they just copy the reference image instead of using them as reference

[tawny laurel]

@latent lance Can you share with use what model were you using, have you accompanied it with a prompt and similar?

[latent lance]

yes its accompanied with a prompt

its happening in nano banana and flux

and imagen

[tawny laurel]

can you provide us with visual cue showing both reference and prompt?

[latent lance]

does nobody else ever have this issue then?

where adobe firefly just regenerates the reference image no matter the model or prompt?

[tawny laurel]

I did catch it, but not with partner models. Asking for details is just further checking

[latent lance]

I am currently trying many different prompts, references, and models, and it keeps happening

[tawny laurel]

Try anew with a new seed, and keep us posted if you keep running into an issues with partner models (more details the better). For me it seems that Firefly reference is completely broken, but generally the way it works does differ from a start with what you can do with Nano for example; it wouldn't be transforming the reference to a new view/scene.

[latent lance]

nah it really doesnt matter what I do, whenever I put in any reference image its just stuck copying the reference image

[tawny laurel]

I do think that for more profound changes I'd have to make up better prompt instead of vaguely asking to change a style.
It's doing something at my end in comparison to what is happening with Firefly

[latent lance]

yeah I get that same issue with every single model pretty much, its just a copy of the same image with slightly different textures iguess, except yours seem to have a bit more variation at least.

I thought it was supposed to be a reference image instead of a copy template

[finite cosmos]

firefly 5 does tend to keep to the reference image as close as possible, firefly 4 will give you results that vary more and you can choose to use it as a composition reference or style reference, and adjust how closely you want it to match with the sliders in the ui

[tawny laurel]

If you'd like more drastic changes you'd prompt for them. What you are describing is how/should Firefly references work; style and composition. I am unsure of the specific of every model, but think that having multiple references as in Nano are more for compiling multiple subjects/objects to a scene

[finite cosmos]

^ the partner models usually you'll write in instructions for what/how you want the reference changed

[latent lance]

no I just want firefly to use it as a reference, not as a template

[finite cosmos]

then use firefly model 4 instead of model 5

[latent lance]

whenever I try model 4 it says this

it hasnt even generated anything yet

[finite cosmos]

are you currently generating something? you'd have to wait for that to finish before starting another, and if you've got multiple tabs running generations you might be getting rate limited and need to wait a minute

[latent lance]

it finally worked

but I asked for 1980s epic dark fantasy theatre posters

and I got this

very strange stuff

I dont remember adobe firefly being this useless

[tawny laurel]

You'd also have to adapt the prompt. With Firefly the most concise as you can get better the results should turn.

[latent lance]

I dont think you understand whats going on

[finite cosmos]

works fine with model 4

[latent lance]

im losing my mind lmao

[finite cosmos]

you're right, we don't, we don't know what model you're using, the reference image, the prompt....

[tawny laurel]

I responded to a prompt that you earlier shared, it's removed but what I suggested applies

[latent lance]

I just told yall, you did the exact same thing as I did as you were able to replicate in the screenshot you sent

you get the results I am looking for, I get the results I just showed you

[finite cosmos]

i'm using firefly model 4 and changed the settings, what model are you using

[latent lance]

firefly model 4, changed all settings multiple times, keep getting random results that have nothing to do with any of the settings or the prompt

[finite cosmos]

it could be the composition reference messing it up, you'd have to remove the references and just try the prompt

[latent lance]

already done that

nvm lol ill just use adobe for editing, its absolutely terrible at generating stuff but for editing its fine

ill get the base content somewhere else

[rapid wave]

no matter what image generator I use, regardless of FF native or partner models, FF's native issue in not recognizing negative prompting really nerfs the creativity and adds a layer of frustration that, to me, is not worth the subscription price.

[calm shuttle]

Firefly's limitations on negative prompting have no effect on partner models. Prompts are passed straight through (assuming they are within guidelines). On the other hand, FF doesn't give much guidance on partner models, so you'll have more success if you consult their websites to find out their best practices, which are different from one model to another. The Gemini models definitely respect negative prompting.

[rapid wave]

Hmmm do you mean that I can neg prompt Image GPT 1.5 in Firefly's limitations? SWEET.

Thank you

[naive dune]

So… quick question for the AI gods / Runway Gen 4.5 enjoyers.
I used Adobe Firefly’s partner model: Runway Gen 4.5 with what I thought were fairly unambiguous instructions. You know, things like:

“Keep the exact same face, facial features, hair colour, clothing style and clothing colours as the reference image.”
One character.
Walk in from the left.
Stop at a medium shot.
Smile.
Open arms like, “Here I am.”
Absolutely no identity drift, face morphing, surprise wardrobe changes, or random extra humans.

All very reasonable. All written down. All stared at repeatedly.
And yet.
Somewhere between “walk diagonally toward the camera” and “welcoming gesture,” the model apparently read:
“Cool, but what if we also turn him into an entirely different Chinese man?”
Now, I’ve reviewed the prompt several times. Slowly. Carefully. With tea.
And I must ask:
👉 Where, exactly, did I request a full ethnicity swap?
👉 Was it between “natural lighting” and “no flicker”?
👉 Is it hidden in the word waist‑up?
Because as far as I can tell, this creative decision cost me 400 credits and resulted in a video that can generously be described as abstract performance art and less generously as absolute garbage.
Anyway — if anyone finds the secret line in my prompt that says “please replace my subject with a completely different man,” let me know. I’d love to stop accidentally funding these experiments.

[finite cosmos]

they do have their own prompt guide and various learning resources on their website https://academy.runwayml.com/guides/prompting-guide

[narrow cedar]

FYI: https://helpx.adobe.com/creative-cloud/apps/generative-ai/model-deprecation.html

[jade niche]

What’s the best model to use for ideating logos and word-marks? Still pretty new to all of the ai stuff and have run into a lot of models generating really poorly and mis-spelling words when it comes to text rendering. Any thoughts or suggestions appreciated!

[swift steppe]

Ideogram, Nano Banana 2 and now GPT Images 2 are all good if your designs require high-quality text treatments.

[carmine horizon]

MOTOR VEO 3.0

A IMAGEM É ESSA PERSONAGEM CARICATA EM 3D,

JÁ CRIEI QUASE 100 CENAS ONDE ELA VIVE HISTÓRIAS EM UM MUNDO MÁGICO E HOJE QUANDO MEUS CRÉDITOS RETORNARAM O MOTO SIMPLISMENTE NÃO ACEITA E TENHO PRAZO DE ENTREGA!

O PROMPT:
"A 3D transformation of an anthropomorphic creature. The animation starts with a touch and transitions into a chest-clutching pose. Vivid teal energy swirls around the creature while metallic wire clouds and lightning materialize in the sky. High-contrast light transition, smooth character movement, stabilized camera, 8k render."

Essas animações que faço é para um trabalho remunerado então eu preciso entrega-lo, caso contrário, demissão. Quem vou responsabilizar já que a ferramenta que eu PAGO e UTILIZO se recusa a funcionar?!

[finite cosmos]

please do not spam the same message across multiple channels, I replied to your post here https://discord.com/channels/1076190214369853510/1498737618999967965

[carmine horizon]

moderador pediu pra enviar aqui :/

[finite cosmos]

then why did you post it in at least 3 other places?

[tight hinge]

As someone who uses firefly almost daily I'm very pleased that the partner models have been provided by Adobe. I've always struggled just using the Firefly model (unlimited usage) as it seems to generate images not related to my prompt, while the other models excel at conforming much more consistently, particularly with images that require multiple variants of the same basic design.

[calm shuttle]

That said, their kerning leaves something to be desired.

[swift steppe]

They leave a lot to be desired. Good enough for ideation. If you're actually designing a logo, you'd do this in Ai (Adobe Illustrator) and not AI. heh

[calm shuttle]

I did exactly that with the Karavan D'Or wordmark that Nanoba came up with for the dromedary beauty ad!

[lilac mulch]

@calm shuttle