Stable diffusion thread

[doogyhatts]

SVD is now available on ComfyUI.
Better output from SVD using ComfyUI instead of Colab?
The video shows SVD doing camera rotation around the character and the character's hair is not always still.


Update:
Not enough vram for my local machine on ComfyUI.
Going back to use Colab instead and test maximum image sizes.

 

[doogyhatts]

Link to SVD models that are based on half-float precision.
https://huggingface.co/becausecurious/stable-video-diffusion-img2vid-fp16

Update:
I got it to work at around 8gb vram.
It turned out that I was using a different workflow that had required higher vram.
Switched back to the one recommended in the video instead.

However, I noticed that the usage of Euler sampler will sometimes cause distortion.