>>2067
The prompt isn't very important. I did use the debreast lora >>1909
(mutilated and fly), and 0.2 weight ayaswan.

The source pic is someone else's art. https://boards.guro.cx/g/src/1692854020287.png
The rest is photoshop inpainting (id, writing, tape)

Anytest essentially extracts an image's composition while keeping the characters and details flexible to prompting.
It's like controlnet for dummies, very simple to use. You can also think of it like i2i but without overwriting the source image, it analyzes the data in some mysterious way and only uses what's relevant to composition.

Latest ver:
https://huggingface.co/2vXpSwA7/iroiro-lora/blob/main/test_controlnet2/CN-anytest_v4-marged_pn_dim256.safetensors

Put it in models/controlnet/

You load the image into controlnet, select anytest and it's ready to go (no preprocessor needed).
Just make sure the image and gen size are the same.
Weight 1 is fine for strong influence, around 0.5 is a good "weak" influence spot, lower starts deviating too much. A lot higher and you're asking it to trace it. I recommend 0.5 to 1.2

Source image quality doesn't matter much either, you can process your own gens (from other models, useful sometimes), your rough sketches, quick photoshops, or bad art. it'll keep composition and change the style.