MiniMax H3 Workflows (T2V, I2V, First-Last-Frame, Ref2V, Lipsync) with SageAttention — v1.1 (faster+upgraded) — v2.0 (modelpreview)

This MiniMax H3 ComfyUI workflow set is for making videos from text prompts and image or video references, with text-to-video, image-to-video, and reference-to-video paths that save video output.

The v2.0 archive lists seven workflows: base and Ref2V prompt generators, three I2V workflows, and two Ref2V workflows.

For I2V, choose the standard workflow, Larryvrh Turbo Nodes, or PDD-Acc 8-Steps Turbo Nodes; a PDD-Acc 8-Steps option is also listed for Ref2V.

The included graphs have LoadImage, VHS_LoadVideo, and VHS_LoadAudioUpload inputs, plus a PixaromaSaveVideo node for saving video.

The publisher points to Comfy-Org/MiniMax-H3 for model downloads and lists separate Heretic text encoder downloads in INT8 and NVFP4.

The included workflows list these node packs: comfyui_nvidia_rtx_nodes, comfyui_steudio, comfyui-kjnodes, comfyui-lora-manager, comfyui-minimax-h3-turbo, comfyui-pixaroma, minimax-h3-longmedia, rgthree-comfy, comfyui-videohelpersuite, comfyui-crystools, comfy-image-saver, comfyui

Model filenames listed across the workflows include gemma4_12b_it_uncensored_heretic_int8_convrot.safetensors, minimax_h3_audio_vae_fp32.safetensors, minimax_h3_fl2va_int8_convrot.safetensors, minimax_h3_turbo_v4_step600_ema.safetensors, minimax_h3_video_vae_int8_convrot.safet, q

The I2V graphs include Model Preview Override and SageAttention patch nodes.

The title advertises First-Last-Frame and Lipsync, while the seven listed v2.0 entries name prompt-generator, I2V, Ref2V, and long-video workflows; check the download contents if you need those options.

Vorschlag · nicht geprüft