Repackaged model files for ComfyUI.
Original model repository:
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:
📂 ComfyUI/
├── 📂 models/
│ ├── 📂 diffusion_models/
│ │ ├── minimax_h3_fl2va_bf16.safetensors
│ │ ├── minimax_h3_fl2va_int8_convrot.safetensors
│ │ ├── minimax_h3_fl2va_pruned_bf16.safetensors
│ │ ├── minimax_h3_fl2va_pruned_int8_convrot.safetensors
│ │ ├── minimax_h3_fl2va_pruned_fp8_scaled.safetensors
│ │ ├── minimax_h3_ref2va_bf16.safetensors
│ │ ├── minimax_h3_ref2va_int8_convrot.safetensors
│ │ ├── minimax_h3_ref2va_pruned_bf16.safetensors
│ │ ├── minimax_h3_ref2va_pruned_int8_convrot.safetensors
│ │ └── minimax_h3_ref2va_pruned_fp8_scaled.safetensors
│ ├── 📂 text_encoders/
│ │ ├── qwen3vl_32b_minimax_h3_bf16.safetensors
│ │ ├── qwen3vl_32b_minimax_h3_int8_convrot.safetensors
│ │ └── qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors
│ ├── 📂 loras/
│ │ ├── minimax_h3_fl2v_turbo_4step_v1.0_768p_comfyui_bf16.safetensors
│ │ ├── minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors
│ │ └── minimax_h3_ref2v_turbo_4step_v0.1_comfyui_bf16.safetensors
│ ├── 📂 vae/
│ │ ├── minimax_h3_audio_vae_fp32.safetensors
│ │ └── minimax_h3_video_vae_fp16.safetensors
│ ├── 📂 embeddings/
│ │ ├── minimaxh3_art_is_explosion.safetensors
│ │ ├── minimaxh3_blooming_flowers.safetensors
│ │ ├── minimaxh3_bullet_time.safetensors
│ │ ├── minimaxh3_dark_magic.safetensors
│ │ ├── minimaxh3_fire_breath.safetensors
│ │ ├── minimaxh3_four_seasons.safetensors
│ │ ├── minimaxh3_kiss_camera.safetensors
│ │ ├── minimaxh3_spiral_ascent.safetensors
│ │ ├── minimaxh3_storm_magic.safetensors
│ │ └── minimaxh3_truman_show.safetensorsDoc: https://docs.comfy.org/tutorials/video/minimax/minimax-h3
To use the embeddings in a text prompt, invoke them in the CLIPTextEncode node like this (you can omit the .pt extension): embedding:embedding_filename.pt.