disco-diffusion
disco-diffusion copied to clipboard
alembics
Update Disco_Diffusion.ipynb
Added Sampling with Splitting Numerical Methods
From the space-nuko repository: https://github.com/space-nuko/guided-diffusion/tree/f91a8708b7b05b48ad01b56bbf5ba375fc66818a
The sampler speeds up the sampling process by reducing the step count (50 to 150 seems to work fin) Seems to work fine with animation and it seems to reduce Vram requirements. I was able to run ViTL14_336px on a free colab (with primary model).
Original repo: https://github.com/sWizad/split-diffusion
Wow absolutely amazing work Dr. Sen. Just setup a DD environment locally on Ubuntu with surprisingly little issue using pytorch 2.x and CUDA 12.3 with your repo's notebook running on dual (so one lmfao) 3090's. The new sampler's are incredibly fast compared to the vanilla DD experience.
Thank you so much for the work you've done on this, DD is always gonna have a special place in my heart and I will probably always be coming back to it over the years. Hope to see this get merged and maybe reignite some interest in others.
