Distribution Matching Distillation

DMD

다단계 말고 으로 냅다 가버리는데 distillation으로 보완해서 부모 모델 upperbound

One-step Diffusion with Distribution Matching Distillation

https://tianweiy.github.io/dmd/

MOSS-Image-SFT: eight-evaluation data-space DMD

lodestones releases a work-in-progress distillation of MOSS-Image-SFT from 50 sampling steps with CFG 3 to eight network evaluations without CFG. The generator iterates on its clean estimate from fixed noise. The Apache-2.0 release includes LoRA files and ComfyUI loading instructions. The step count describes inference configuration; the model card does not establish equivalent image quality or a matching wall-clock speedup.

huggingface.co

https://huggingface.co/lodestones/MOSS-Image-SFT-8-steps-dmd

AI generates high-quality images 30 times faster in a single step | MIT CSAIL

In our current age of artificial intelligence, computers can generate their own “art” by way of diffusion models, iteratively adding structure to a noisy initial state until a clear image or video emerges. Diffusion models have suddenly grabbed a seat at everyone’s table: Enter a few words and experience instantaneous, dopamine-spiking dreamscapes at the intersection of reality and fantasy. Behind the scenes, it involves a complex, time-intensive process requiring numerous iterations for the algorithm to perfect the image.

https://www.csail.mit.edu/news/ai-generates-high-quality-images-30-times-faster-single-step

Distribution Matching Distillation

DMD

MOSS-Image-SFT: eight-evaluation data-space DMD

Backlinks

Recommendations