Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

SD3+ #2688

Merged
merged 13 commits into from
Feb 27, 2025
Merged

SD3+ #2688

merged 13 commits into from
Feb 27, 2025

Conversation

DenOfEquity
Copy link
Collaborator

@DenOfEquity DenOfEquity commented Feb 25, 2025

Building on the work of graemeniedermayer.

Doesn't need change to huggingface_guess repo.
Doesn't need duplicates of all the config files.

Tested successfully:

  • SD3 (all in one checkpoint)
  • SD3.5-medium (transformer+vae model, with CLIPs and T5 specified as additional modules)
  • SD3.5-large (transformer only model in fp8, with VAE and TEs specified as additional modules)
  • img2img, hiresfix
  • custom CLIPS
  • batch
  • nf4 (as created by Save UNet in Checkpoint Merger tab)
  • LoRAs (only tested a couple)
  • GGUF (tested: medium Q4_K_S; large-turbo Q4_1)

Tested without success:

  • GGUF (Q4_K_S)

not tested (yet):

  • SD3.5-large-turbo - should work, transformer config matches SD3.5-large
  • nf4
  • GGUF
  • LoRAs

Will merge tomorrow unless something occurs. Testing welcome, of course.

and add a CLIP-L key not handled from one model type
(no difference to test image)
just cast some input parameters to int
include percent-to-sigma function: useful for SLG implementation, which can be done in a post_cfg function.
@DenOfEquity DenOfEquity merged commit f23bc80 into lllyasviel:main Feb 27, 2025
@DenOfEquity DenOfEquity deleted the sd3+ branch February 27, 2025 21:43
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant