Firefly cannot make art anywhere close to my style, even with a direct image prompt

Question

I have been trying to replicate my own art style in Adobe Firefly to fill in gaps in my own artistic skill, with the intent of using these images as light reference to get better at drawing poses and settings that I'm not familiar with. But every time I have uploaded my art to the generator, it gives me increasingly bad results that honestly make me feel like the program reads my art as "amateur middle school bs" instead of a style I have been working on for 10+ years. Is the program just only capable of imitating hyper-detailed digital paintings? How do I get it to actually follow the image prompt I give it?

This is the sprite I uploaded as reference, drawn by me:

This is a screenshot of what it gave me when I tried to use my art as a reference:

I kind of feel insulted that it seems to think my art looks this bad. Even with strength at every different setting, the results were absolutely terrible. I didn't want to use this to emulate other people's art styles, I wanted to enhance my own in a way that I could work with, and I just can't get it to even acknowledge that mine isn't terrible.

rayek.elfin · Answer

I wouldn't take a genAI's result a personal attack on your art style nor feel insulted. 🙂

The result is entirely dependent on a particular genAI's model training. Adobe's model just isn't trained to deal with drawings like yours. I'd gather no existing genAI model would be able to emulate your art style exactly.

A genAI model has to be trained or extended with a particular style through a so-called LoRA to really do it justice. That requires more time and effort on the part of the user, and is not something you can do with Firefly or Midjourney, or other online genAIs.

It is therefore very dependent on the training of any particular AI model what kind of result to expect. For example with your prompt, it goes from:

(sorry for this one)

to

...and so on.

All of these were rendered using your drawing as a reference image and your prompt with a different trained genAI model.

So, in the end I don't think any of these fit your style. Completely different. The result would be similar with other genAI services. Only by creating a LoRA based on your own style would you be able to approach your drawing style more-or-less correctly. And that takes more knowledge and technical tinkering, and doesn't work with online services such as Firefly or Midjourney anyway.

Can't blame Firefly. GenAI is not magic nor intelligent or aware 😉

Which is a good thing, because original art styles will stay original (for as long as someone doesn't create an AI model or LoRA that is trained on that art style).

サインアップ

ソーシャルログイン

コミュニティへログイン

ソーシャルログイン

ファイルをウイルススキャンする。

このファイルはダウンロードできません