Search Results
7/12/2025, 6:54:09 PM
big
>PromptTea improves caching for video diffusion models by adapting reuse thresholds based on prompt complexity. It introduces PCA-TeaCache (noise-reduced inputs, learned thresholds) and DynCFGCache (adaptive guidance reuse). Achieves up to 2.79× speedup with minimal quality loss.
https://github.com/zishen-ucap/PromptTea
>PromptTea improves caching for video diffusion models by adapting reuse thresholds based on prompt complexity. It introduces PCA-TeaCache (noise-reduced inputs, learned thresholds) and DynCFGCache (adaptive guidance reuse). Achieves up to 2.79× speedup with minimal quality loss.
https://github.com/zishen-ucap/PromptTea
Page 1