Miles-diffusion introduces LoRA SFT for fast post-training of diffusion models

ying11231 · x · 2026-08-28

Miles-diffusion introduces LoRA SFT for fast, targeted post-training of diffusion models. A rank-64 LoRA was trained for MiniMax H3 using 254 curated windows in under 3 hours on 8 GPUs, improving physical realism. The lightweight adapter can be exported to safetensors and served directly with SGLang.

Original post →

More from Multimodal

Multimodal channel →