Vid2Vid Hunyuan GGUF + Upscaler 12GB & 16GB & 24GB V2V Workflow — V01 GGUF

Use this Hunyuan Video workflow for video-to-video processing and upscaling: provide an input video and a manually entered prompt, and receive a video output; the described examples use 240x320 footage.

For input videos not generated by Hunyuan, use the frame-load cap arrows to follow Hunyuan’s 4*x+1 frame rule.

The described examples start with 240x320 footage and list approximate upscale factors of 1.3× at 12GB VRAM, 2.5× at 16GB, and 3.5× at 24GB.

Provide an input video and enter the prompt manually for the video-to-video process.

Install Kijai nodes in addition to nodes available through ComfyUI Manager.

Named model files are: boreal-hl-v1.safetensors; hunyuan-video-t2v-720p-q4_k_m.gguf; hunyuanVideoSafetensors_comfyVAEFP8.safetensors; llava-llama-3-8b-v1_1.Q4_K_M.gguf; and Long-ViT-L-14-GmP-SAE-TE-only.safetensors.

Named workflow components include JWFloatToInteger, ImageScale, VHS_LoadVideo, workflow>Config ⚙️, GetImageSizeAndCount, DetailDaemonSamplerNode, UnetLoaderGGUF, VAEDecodeTiled, RandomNoise, VHS_VideoInfo, and Label (rgthree).

Other named workflow components include HunyuanVideoLoraLoader, ImageSharpen, workflow>Sampling Config, CLIPTextEncode, DualClipLoaderGGUF, Anything Everywhere, TeaCacheHunyuanVideoSampler, VAELoader, GetNode, mxSlider, SetNode, ColorMatch, Seed (rgthree), VHS_VideoCombine, VAEใช

Enter prompts manually in the workflow.

The documented example uses the Hunyuan T2V 720p GGUF Q4_K_M, an average GGUF, and 73 generated frames.

Before using it, verify that the named model files and Kijai nodes are available in your ComfyUI setup.

Vorschlag · nicht geprüft

The listed ~14min on an RTX3090 is tied to the 24GB, 3.5× example; treat it as a reference, not a guaranteed runtime.

Vorschlag · nicht geprüft