slaanesh

Experiment - Red head dllc excercise

Added 2025-01-26 09:19:27 +0000 UTC

Hey,

roughly a year ago I asked you if I should try out the new AI tools.

The poll suggests that most of you were open to the idea of training it to draw in my style. Well, that didn t work out and also the control is way to time-consuming/impossible. Even creating backgrounds is not yet something I can do. I can t get fitting results (should give that a try again in the future). I also tested if it can speed up coloring and no not really. But I can use AI to give color suggestions / create color pallets - no arguing stable diffusion can do that very well - and slowly I see my coloring improve. so that helps to speed up things a bit.

I also tried to use different LLMs in combination to create casted ladies - mostly starting with drawings or paintings or alterations of existing images. Though Bing changed recently it could be used sometimes as the base. Image FX shows some interesting results as well (AIcast on twitter) that seem less "artificial". I have not tested it (to lazy to deal with geoblocking) but it seems less restrictive than the bing image creator. It also keeps characters more consistent.

Recently several services popped up that allow animation/video creation (5-10 second clips). Two or three weeks ago Hailu improved their "anime/cartoon" model to a level where the results can be good when you are lucky. There are still often artifacts or strange and impossible movements and half of the time it bends casts in places that should be stiff. It is still like a gatcha game but the hit rates are not as bad as 0.97%.

Anyway, I attached my best results. (after more than a year of a try and error) What do you think?

The base image for the attached video clips: