DFlow2 distillation LoRA cuts sampling from 50 to 4 steps with Torch and MLX ComfyUI workflows
TimeTruth2490 · reddit · 2026-11-22
A Reddit user released DFlow2, a distillation LoRA using eDMD + LADD to compress diffusion sampling from 50 steps down to 4, shipping ComfyUI workflows for both Torch and MLX so it runs on Apple Silicon too.
Related event: Community Devs Distill DFlow2 to Generate Images in 4 Steps(2 posts)→
More from Multimodal
- Open-source AI video editor Scarlet adds Gemini support, frame annotation and lip-sync fixes — MatthewBerman · 2026-11-22
- DFlow2 2B T2I model distilled from 50 to 4 steps with eDMD + LADD LoRA — TimeTruth2490 · 2026-11-22
- A 2:40 AI-generated Lord of the Rings x Harry Potter dark-fantasy crossover teaser — Quiet_Cobra · 2026-11-22
- ComfyUI launches model usage rankings: Qwen Image 2.1 tops generations at 16% — cremebait15 · 2026-11-22
- pi-H3, a flow-matching video model built on MiniMax-H3, trends on Hugging Face — Lakonik · 2026-11-22
- GPT 6.2 Astra composes Beethoven-style scherzo, with all 604 key strikes from the score — imjustnewatai · 2026-11-22