Skip to main content
silja23
Participating Frequently
February 5, 2026
Open for Voting

Referenzbild in Firefly

  • February 5, 2026
  • 2 replies
  • 19 views

Wenn man in Firefly ein Referenzbild hochlädt, sieht das generierte Bild exakt wie das Referenzbild aus, auch wenn der Prompt deutlich verändert wurde (Bett statt Krankenwagen zum Beispiel)

    2 replies

    Hajar Toutsi
    Community Manager
    Community Manager
    February 6, 2026

    Thanks silja23 for sharing the reference image, that helped a lot.

    I tested your exact workflow using the same image, and I was able to reproduce the issue. Firefly kept generating another ambulance instead of switching to a hospital bed icon.
     

    After a bit of digging, I found the cause. The reference image you uploaded was in WEBP format, and Firefly sometimes has trouble interpreting WEBP files correctly. When that happens, the model doesn’t extract the subject properly, so instead of using the reference for style, it ends up re‑generating the same subject again, which is why it kept giving you another ambulance.
     

    Once I converted the image to PNG, the reference behaved normally and Firefly generated a hospital bed icon as expected.
     


    If you run into this again, converting the reference image to PNG or JPG before uploading should solve it.
     

    As for your first generation of the ambulance, here are a couple of things that can help you achieve better results:

    • Firefly doesn’t respond well to negative prompting: Instead of saying “not childish,” it works better to describe the style you do want (clean, mature, professional, geometric, etc.).
    • For icons, Text to Vector often gives more consistent results: It’s designed for clean shapes and repeated iterations.
       

       

    • Take advantage of the advanced settings: Firefly models offer a variety of controls that can help you fine‑tune content type, composition, and overall style more easily. (For this test, I stayed with Text to Image to match your workflow)
       


    I hope this helps in your future generations. If you’d like, feel free to share the next result, I’m happy to help you fine tune it.

    Thanks.

    Hajar

    silja23
    silja23Author
    Participating Frequently
    February 9, 2026

    Good morning Hajar! Thank you a lot for doing research and solving the problem! As I just uploaded the result, firefox gave me before, I was not aware of the format. I am sure, that your solution will help me. 

    Have a great day!

    Silja :)

    Hajar Toutsi
    Community Manager
    Community Manager
    February 5, 2026

    Hi silja23,

    Thanks for letting us know.
     

    We’d love to help you get more reliable results. To do that, it would really help if you could share the reference image you used, along with the model and prompt you tried.
     

    Once we have that, we’ll be able to take a closer look and understand what’s going on.
     

    Looking forward to hearing from you.


    Hajar

    silja23
    silja23Author
    Participating Frequently
    February 6, 2026

    Hi Hajar! Thanks for your answer and for offering help. Yesterday I generated an illustration of an emergency car with this prompt: “generiere mir ein isometrisches Icon eines Krankenwagens, welches sich für ein Spiel (für Erwachsene eignet). Einfach gehalten, aber farbig, nicht zu kindlich” and got this picture