Home Artists Posts Import Register

Content

Some stable diffusion action

Files

Alison Style Morph

Comments

Bernard

Nice Animation. One thing that I immediately noticed is, how all the variants of Alison differ not only in style but also in the emotion they portrait. Now I wonder if this is because the prompt only specified generic "smiling" and the AI just randomly choose different facial expressions or if AI will always have trouble generating different graphical styles that evoke the same emotions. Theoretically it should be just a matter of training data, yet not having the internal feedback loop of a human artist may be an insurmountable hurdle.

satin

I prompted for "open mouth, happy". The current models can do basic expressions like happy, sad, angry, yelling, blushing. It's much harder to get something like confused, surprised, afraid, puzzled, conflicted, worried, etc. It's also nigh impossible to get a character to look away or roll their eyes. I don't think training with manually labeled images is going to be enough to surmount this. People aren't consistent in describing complex expressions and we don't really have the words for everything a face can show. The system would need to be able to figure out the similarities between a cartoon face and a photo face expression on its own. We haven't reached the limit of current techniques yet, so we'll see what happens.