๐ง๐๐ฟ๐ป๐ถ๐ป๐ด ๐ ๐ ๐๐ฎ๐ ๐๐ป๐๐ผ ๐๐ป๐ถ๐บ๐ฒ ๐๐ฟ๐
I tried to turn my cat into anime art. The AI drew a human girl instead.
The AI saw the tuxedo colors of my cat. It turned them into a black dress and white collar.
I used IPAdapter to fix this. IPAdapter uses a photo as a visual guide.
The AI still struggled. Here is the process to fix it:
- I raised the reference strength to 0.85.
- I added humans and girls to the negative prompt.
- I removed words like big eyes.
Two things caused this problem:
- Anime models prefer drawing people.
- My cat sits upright. This look tricked the AI into drawing a person.
Every style needs a different balance. Cyberpunk worked with a photo. Storybook watercolor failed. It made humans or two cats.
My tools:
- DGX Spark machine.
- ComfyUI.
- IPAdapter Plus node.
- AnythingV5 and Realistic Vision V6 models.
I cropped the photo to a square. I kept the seed and steps the same for a fair test.
Next I will try video.
Optional learning community: https://t.me/GyaanSetuAi