Replace CLIP in Stable Diffusion with Jina CLIP to remove 77 tokens limitation

#17
by off6atomic - opened

I saw that Jina CLIP supports 8000 tokens so I'm wondering whether it would be a good idea to replace CLIP in Stable Diffusion with Jina CLIP to remove 77 tokens limitation and increase prompt adherence.

Do you think this would work well after training SD?
Any possible obstacles?

Jina AI org

hi @off6atomic indeed this is a very interesting case, to be frankly our team only have basic understanding of CLIP as a component in difussion process, we can not judge. I would be really interesting if you can conduct some experiment.

Bo

Sign up or log in to comment