Performance PR

#2
by multimodalart HF staff - opened
  • Swap LCM LoRA to SDXL Lightening 2 steps (faster, more quality)
  • Switch regular VAE to tiny TAESD VAE
  • Add header and mention to blog
  • Add keyboard navigation (A to Dislike, Space for Neither and L to like)
  • Disable Safety Filter (redundant in SDXL for this use-case and lots of false positives)

Performance result on A10G:

  • < 1s per image

Nice!

rynmurdock changed pull request status to merged

Sign up or log in to comment