<ul><li>TeaCache is a caching technique that leverages the similarity between input and output of attention blocks in diffusion models, speeding up ComfyUI image and video generation.</li><li>TeaCache, in combination with model compiling, can result in 2.5X to 3X speed improvements in API inference and ComfyUI workloads.</li><li>TeaCache node from the ComfyUI-TeaCache node pack includes parameters like rel_l1_thresh and max_skip_steps to control cache refreshing and generation speed.</li><li>Setting proper values for rel_l1_thresh and max_skip_steps can maintain quality while improving generation speed, but values above certain thresholds may lead to loss of details.</li></ul>

Speed up ComfyUI Image and Video generation with TeaCache

Discover more