form2 (extended) -SD remix

 

Stable diffusion generative image synthesis animation run of the form2 storyline used in a previous CLIP guided direct RGB synthesis experiment.  The color pastel nude sketch thing at the end totally surprised me, since the text prompt at that point in the storyline is 'endless form'.  So don't watch the end of the video if you are offended by nudity.

Stable diffusion is not doing really well in this run to convey the abstract geometric patterning prompts in the 'form2' story line in visual form.  Direct CLIP guided synthesis seems to do way better at this kind of thing, as does the other latent diffusion code i've been running.  Maybe a byproduct of both their direct synthesis approach as well as the 'aesthetic' image database they did their training on?

How you 'modify' the previous output frame that you use as input for the next frame animation is also a huge factor in what ends up being conjured up by the synthesis algorithm.  Studio Artist users should be familiar with this recursive approach.  Interesting to see long time Studio Artist animation concepts become more mainstream in recent generative ai code.