MiniMax H3 Workflows (T2V, I2V, First-Last-Frame, Ref2V, Lipsync) with SageAttention — v1.1 (faster+upgraded) — v3.0 (RTXHDR+Interpolate)

Use this ComfyUI workflow pack to create MiniMax H3 videos from text or images. It includes first/last-frame, reference-to-video, face-refinement, and lipsync variants; depending on the workflow, inputs can include text, images, video, or audio, and the output is video.

The bundle includes six workflows: Face Refiner V2V, MiniMax H3 I2V, MiniMax H3 Ref2V, and Base Prompt, Live Wallpaper Prompt, and Ref2V Prompt generators.

Across the included graphs, inputs shown include images, video, audio, and multiline text; video workflows include a video-combine node for output.

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

The workflow files list these ComfyUI node packs: comfyui_fearnworksnodes, comfyui-kjnodes, comfyui-pixaroma, comfyui-videohelpersuite, comfy-image-saver, comfyui-easy-use, comfyui-rmbg, comfyui_nvidia_rtx_nodes, comfyui_steudio, comfyui-AGSoft, comfyui-lora-manager, and rgthree–

Listed model files include face_yolov8m.pt, person_yolov8m-seg.pt, minimax_h3_audio_vae_fp32.safetensors, minimax_h3_fl2v_turbo_4step_v1.2_768p_comfyui_bf16.safetensors, minimax_h3_fl2va_int8_convrot.safetensors, minimax_h3_ref2va_int8_convrot.safetensors, minimax_h3_video_vae_–

The v3.0 notes say RTX HDR and RTX Frame Interpolation can be switched on; the I2V and Ref2V graphs also include RTXTrueHDR, RTXVideoFrameGeneration, and RTXVideoSuperResolution nodes.

The listed model filenames have no individual download URLs; confirm you have the exact files before using a matching workflow.

Suggestion · not verified

The I2V and Ref2V workflows list comfyui_nvidia_rtx_nodes; confirm that node pack is available in your ComfyUI setup.

Suggestion · not verified