How image generation got good (diffusion derivations)
Comments
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Ranked by a public rubric, not by votes: what a comment adds counts, not who wrote it. How comments are ranked · Comment rules
Summary
This post explains why image generation became reliable, focusing on diffusion models. It argues that L2 regression blurs outputs because it averages over multimodal possibilities, while diffusion iteratively denoises toward a specific sample.