Possible video upscale with Hunyuan 1.5: hunyuan-15-upscale-1, hunyuan-15-upscale-2.
PR 11212 merged to ComfyUI native fixing burn-out issue with Hunyuan 1.5 distilled and possibly SR (upscaler) models.
Distilled version of 480 model released by Tencent and repackaged for Comfy: link Both cfg and step distilled, allows generations in 4-12 steps.
seems to need at least 8 steps most of the time, shift 7.0; avoid the fp8 unless you have to use it; full model, but possibly Lora could be extracted; no 720 distilled model yet
Note: “sr” in model names on HF means “super resolution”, e.g. upscaler
has anyone compared HunyuanVideo-1.5 with Wan 2.2? Does HunyuanVideo-1.5 have any particular advantages? Not even close; camera very often goes WEEEEEEEEEE. BYEEEE, if you dont prompt it; blurriness
Latest is version 1.5 up to 10 sec generations.
Load models from HF:Comfy-Org/HunyuanVideo_1.5_repackaged Place into ComfyUI folder names following the structure on HF.
Native Comfy implementation available in latest main branch
SageAttention supposedly better quality than EasyCache
