Skip to main content
cuela
cuela
Participant
July 18, 2025
Answered

Firefly not using my uploaded images as reference — what am I doing wrong?

  • July 18, 2025
  • 1 reply
  • 782 views

Hi everyone,

This is my first post, hope I’m in the right forum.


I’m trying to generate painted portraits in Firefly using my own photo as a reference for composition/structure and my own artwork as a style reference. But Firefly seems to completely ignore both — especially the composition — and ends up creating a totally different face, seemingly based on someone else’s photo, with no resemblance to mine. I’ve generated many variations, but nothing even comes close to the original likeness.

I’m not aiming for photorealism, but rather an artistic or abstract painting with some degree of likeness. From what I understand, Model 3 might be better for this than Model 4, which feels more photographic. Is that true?

I’ve only used Firefly a few times since its release, but each time I’ve generated hundreds of images out of frustration trying different prompts, settings, and improved references just to get a hint of resemblance. I’ve also used the thumbs up/down with detailed feedback. I'm using all my credits and am left me with no usable results.

Last time, I downloaded many of the outputs to study and try to understand what might be going wrong.

I’ve watched several YouTube tutorials, but I’m not getting anything close to what they show. Maybe I’m misunderstanding how to use the sliders or reference tools?

Is there a clear step-by-step video or guide that explains how to get Firefly to actually use uploaded images for structure (composition/likeness) and style in image-to-image generation? And is image-to-image even the right term for what I’m trying to do?

Is someone having the same problem?

Any tips, workflows, or insight would be hugely appreciated.

 

Thanks so much!

 

Correct answer Oh.N8

Hey @cuela 

1st you are in the right forum. Welcome in and thanks for asking such good Qs.

2nd sharing an image w a screenshot of the results could help.

Regarding getting the desired results of having FF follow the image specific can be a bit tricky and challenging. FF is good w following designs, icons, and images drawn (better than an actual photo). Keep in mind that FF is trained on images in Adobe stock and other commericially safe images.

Drawn images help considerably.

That being said, place the image into Composition Ref and Style ref and play around w the sliders. 
2) if you are not getting the desired results after the 3rd or 4th generation go into Files and delete the generated images. Completely erase all the ones not close to your desired results. Close off the tab. Open a completely new tab. 

Sometimes FF gets into a bad seed. I find starting fresh helps considerably.


Cheers

Nate

1 reply

O
Community Expert
Oh.N8Community ExpertCorrect answer
July 19, 2025

Hey @cuela 

1st you are in the right forum. Welcome in and thanks for asking such good Qs.

2nd sharing an image w a screenshot of the results could help.

Regarding getting the desired results of having FF follow the image specific can be a bit tricky and challenging. FF is good w following designs, icons, and images drawn (better than an actual photo). Keep in mind that FF is trained on images in Adobe stock and other commericially safe images.

Drawn images help considerably.

That being said, place the image into Composition Ref and Style ref and play around w the sliders. 
2) if you are not getting the desired results after the 3rd or 4th generation go into Files and delete the generated images. Completely erase all the ones not close to your desired results. Close off the tab. Open a completely new tab. 

Sometimes FF gets into a bad seed. I find starting fresh helps considerably.


Cheers

Nate

    cuela
    cuelaAuthor
    Participant
    July 22, 2025

    Thank you so much for the warm welcome and helpful tips Nate!

     

    I appreciate the clarification about FF being specifically trained on Adobe Stock and other commercially safe images. I had assumed it would work more directly with our own uploaded photos using both the structure and style sliders, as some tutorials suggest.

     

    It will definitely be tricky to get any likeness in a painted portrait without using my own photos as reference, even though I’m aiming for a non-photographic, more abstract style. I’ll definitely follow your advice. Deleting unsatisfactory results and starting fresh in a new tab makes a lot of sense.

     

    I should have mentioned that I’ve worked from drawings briefly, but not systematically. I’ll dig deeper now that I understand they tend to give better results. Do you think Model 3 works better than Model 4 for these kinds of abstract, artistic portraits, or are they about the same?

     

    Also, I’m curious. Does the FF team have plans to better support working with user-uploaded images? Is that something they’re already working on? Hope so! FF is indeed a wonderful tool for feeding the imagination, really inspiring, and I’m looking forward to learning more about it.

     

    Thanks again for the guidance!

     

    Cheers,

    Cuela

    O
    Community Expert
    Oh.N8Community Expert
    July 22, 2025

    Hey @cuela 

    I find starting w Model 3 gives better results for this type of project. 
    since you have some background w drawings, consider experimenting w uploading the drawings and see what kind of results you would get. 

    Your others Qs: I do not know. 🙂

     

    Enjoy the creative process.

    Cheers

    Nate

    Terms & ConditionsAccessibility statement
    Using the community Terms of use Privacy Cookie preferences Do not sell or share my personal information ad choicesAdChoices