Split sigmas speed up video reference in Stable Diffusion
A Reddit user proposes splitting sampling across resolutions to speed up video reference generation, noting that video ref injects thousands of tokens, making low-res steps cheaper. The technique puts the split early to reduce cost.
1 source
Daily brief
Get tomorrow's AI brief in your inbox
More stories today
- Claude Code silently deletes local history older than 30 days
- MiniMax tool turns anything into realistic human video
- MirroS' Code-as-World rewrites videos into executable MuJoCo programs
- PRAXIST research system boosts MLE-bench scores 44%
- Chollet: General intelligence means adapting to any problem