Thread by @ShuaiYang1991

Combine #stableDifusion with style transfer models #VToonify and #DualStyleGAN. Toonify videos using #VToonify with the backbone #DualStyleGAN trained on face images generated by #stableDifusion. (1/5) #aiartist #deeplearning

First, generate random art images with #stableDifusion with prompt `A ultradetailed bust portrait painting of a <woman|man>, highly detailed, a full front face in the middle of canvas, upper body, digital painting, sharp focus, _adjective_ mood, illustration by #Artist` (2/5)

Next, train image style transfer model #DualStyleGAN with the these images. The results look pretty impressive! (3/5)

Finally, with the trained #DualStyleGAN as backbone, train video style transfer model #VToonify. Besides the illustration style in 1/5, here are the results of the impasto and fantasy styles. (4/5)

A nice way of customizing your own style transfer model with the SoTA diffusion model!
I just added the #VToonify illustration model to the huggingface web demo. Have fun! (5/5)
huggingface.co/spaces/PKUWill…