Vision endcoder
#14
by
kiarie
- opened
Hi there,
Thanks again for all the good work!
Is there a plan to release the CLIP image encoder that trained Stable diffusion 2?
The text encoder is present, but for image variation experiments, it would be awesome to also have access to the image embedder.
Cheers,
Ki
kiarie
changed discussion status to
closed
Nevermind! Just realised you must've frozen the CLIP image encoder (OpenCLIP-ViT/H), and only updated the weights in the text encoder during training.