Study: Pixel-Space Diffusion Training Better for Distillation
A new empirical study finds that pixel-space training of text-to-image diffusion models converges slower than latent-space training, but this makes it better suited for distillation, speeding it up by 3x.
2026-08-20 ~ 2026-08-20 · 2 related posts
- Pixel-Space Diffusion Training Slower, Ideal for Distillation with 3x Speedup — bdsqlsz · 2026-08-20
- Study: Pixel-Space Training Suited for Diffusion Distillation — bdsqlsz · 2026-08-20