songrise/Artist

Any luck with sdxl based model?

jsg921019 opened this issue · 3 comments

jsg921019 commented

First, Thank you for interesting work.
Have you tried using SDXL based models? I found traits of playgroundv2 in the code.
But i could not find any mentions in README or paper.

  • Also curious about experiments using image style reference instead of text style references.
songrise commented

Thanks for your comment. 1. Yes I am still trying to extend it to SDXL-based models. Currently I find Playground v2 works well for some image and prompts, but the performance is not as rubost as the SD2.1. So I did not include it in current version. 2. The reference image based style transfer is another intereseting future work, which I also tried before (not included in this repo).

This paper is still a WIP project, so I may update the two features latter depends on experiment results.

jsg921019 commented

thank you for the answer!

songrise commented

added playground v2 config, you can try it out