1.Tencent-Hunyuan / PrismWe propose Prism, a dynamic sparse attention framework for natively training joint video-audio generation models at 2K.diffusion-modelshigh-resolutionsparse-attentionvideo-generationPythonOther license1.5K total12 forks+377 starsfrom 1,128 · score 5.93
2.KooroshRH / AReAOfficial implementation of "AReA: Adaptive Receptive-Field Attention for Efficient High-Resolution Diffusion Transformers"diffusion-modelsefficient-inferencehigh-resolutionimage-generationprompt-datasetPythonAPACHE-2.0155 total5 forks+112 starsfrom 40 · score 4.73