[CVPR 2026] Adaptive Spectral Feature Forecasting for Diffusion Sampling Acceleration
-
Updated
Apr 30, 2026 - Python
#
diffusion-acceleration
Here are 6 public repositories matching this topic...
Drop-in TaylorSeer/HiCache basis upgrade — training-free diffusion acceleration via a Dynamic Mode Decomposition (Prony) exponential feature-forecast basis. Not the SGLang KV-cache HiCache.
-
Updated
Jul 25, 2026 - Python
Training-free HiCache acceleration for TRELLIS.2 image-to-3D in ComfyUI (~2x, near-lossless). Pairs with ComfyUI-Trellis2.
-
Updated
Jul 25, 2026 - Python
[ICLR 2026] The official release of paper "Astraea: A Token-wise Acceleration Framework for Video Diffusion Transformers"
-
Updated
Mar 11, 2026
Training-free Hunyuan3D acceleration node for ComfyUI: skip DiT steps, forecast the flow-matching velocity (HiCache Hermite / HiCache++ DMD via hicache-pp). Pairs with ComfyUI-Hunyuan3DWrapper.
-
Updated
Jul 25, 2026 - Python
Training-free HiCache acceleration for TRELLIS image-to-3D in ComfyUI (~2x, near-lossless). Pairs with ComfyUI_TRELLIS.
-
Updated
Jul 25, 2026 - Python
Add this topic to your repo
To associate your repository with the diffusion-acceleration topic, visit your repo's landing page and select "manage topics."