MiniMax-H3
Comfy Org's ComfyUI-ready single-file repackaging of MiniMax's H3 audio-video generation model.
Base model
Model Description
Repackaged model files for ComfyUI.
Original model repository:
- https://huggingface.co/MiniMaxAI/MiniMax-H3
- https://huggingface.co/lightx2v/Minimax-h3-Turbo
- https://huggingface.co/alibaba-pai/MiniMax-H3-Fun-Controlnet-Union
- https://huggingface.co/Kijai/MiniMax-H3-experimental
The Qwen3-VL-32B nvfp4_awq quant is converted from: https://huggingface.co/cybermotaz/Qwen3-VL-32B-Instruct-NVFP4
This nvfp4 text encoder does not require Blackwell GPU to use.
For diffusion models prefer int8_convrot if you are able to use pytorch with cu130.
fp8_scaled should only be used if you cannot use int8_convrot.
Place the files in the following folders:
π ComfyUI/
βββ π models/
β βββ π diffusion_models/
β β βββ minimax_h3_fl2va_bf16.safetensors
β β βββ minimax_h3_fl2va_int8_convrot.safetensors
β β βββ minimax_h3_fl2va_pruned_bf16.safetensors
β β βββ minimax_h3_fl2va_pruned_int8_convrot.safetensors
β β βββ minimax_h3_fl2va_pruned_fp8_scaled.safetensors
β β βββ minimax_h3_ref2va_bf16.safetensors
β β βββ minimax_h3_ref2va_int8_convrot.safetensors
β β βββ minimax_h3_ref2va_pruned_bf16.safetensors
β β βββ minimax_h3_ref2va_pruned_int8_convrot.safetensors
β β βββ minimax_h3_ref2va_pruned_fp8_scaled.safetensors
β βββ π text_encoders/
β β βββ qwen3vl_32b_minimax_h3_bf16.safetensors
β β βββ qwen3vl_32b_minimax_h3_int8_convrot.safetensors
β β βββ qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
β βββ π loras/
β β βββ minimax_h3_fl2v_turbo_4step_v1.0_768p_comfyui_bf16.safetensors
β β βββ minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors
β β βββ minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors
β βββ π vae/
β β βββ minimax_h3_audio_vae_fp32.safetensors
β β βββ minimax_h3_video_vae_fp16.safetensors
β βββ π model_patches/
β β βββ minimax_h3_fun_controlnet_union_pruned_bf16.safetensors
β β βββ minimax_h3_fun_controlnet_union_pruned_int8_convrot.safetensors
β βββ π embeddings/
β β βββ minimaxh3_art_is_explosion.safetensors
β β βββ minimaxh3_blooming_flowers.safetensors
β β βββ minimaxh3_bullet_time.safetensors
β β βββ minimaxh3_dark_magic.safetensors
β β βββ minimaxh3_fire_breath.safetensors
β β βββ minimaxh3_four_seasons.safetensors
β β βββ minimaxh3_kiss_camera.safetensors
β β βββ minimaxh3_spiral_ascent.safetensors
β β βββ minimaxh3_storm_magic.safetensors
β β βββ minimaxh3_truman_show.safetensors
Workflows
- I2VοΌhttps://github.com/Comfy-Org/workflow_templates/blob/main/templates/video_minimax_h3_i2v.json
- T2VοΌhttps://github.com/Comfy-Org/workflow_templates/blob/main/templates/video_minimax_h3_t2v.json
- R2VοΌhttps://github.com/Comfy-Org/workflow_templates/blob/main/templates/video_minimax_h3_r2v.json
Doc: https://docs.comfy.org/tutorials/video/minimax/minimax-h3
Embeddings
To use the embeddings in a text prompt, invoke them in the CLIPTextEncode node by
combining embedding: with the filename of the embedding like this for minimaxh3_art_is_explosion:
embedding:minimaxh3_art_is_explosion
Just exchange the minimaxh3_art_is_explosion part with the filename of the embedding you want to use.
Sign up to read complete case studies, access detailed metrics, and unlock all use cases.
Sign up to read complete case studies, access detailed metrics, and unlock all use cases.