MiniMax-H3 Ref2VA: Two-Pass HD Sampling

Note: All model files and custom nodes below must be downloaded via the client. If you do not have the client, Download >>>

MiniMax-H3 is a cutting-edge native omni-modal AI generation model developed by MiniMax, designed to seamlessly integrate video and audio synthesis in one go. Its standout feature is the ability to generate hyper-realistic, fluid 4K videos from simple text or image inputs while simultaneously creating perfectly synchronized natural sound effects, ambient background audio, and even voiceovers. Additionally, H3 boasts impressive text-rendering precision, effortlessly producing crisp, sharp written text directly within the video frames—making it an all-in-one powerhouse for creating cinematic, fully-realized video content effortlessly. The Image Reference-to-Video functionality enables users to upload one or more images as visual anchors, guiding the AI to generate dynamic videos that maintain high consistency with the original image(s) regarding artistic style, character identity, or scene layout.

MiniMax-H3-TAE >1 files

Name	Size	Target Dir
taeh3.safetensors	22.71 MB	vae_approx

MiniMax-H3-vae >3 files

Name	Size	Target Dir
minimax_h3_audio_vae_fp32.safetensors	605.25 MB	vae
minimax_h3_video_vae_int8_convrot.safetensors	2.81 GB	vae
minimax_h3_video_vae_fp16.safetensors	5.21 GB	vae

MiniMax-H3 >8 files

Name	Size	Target Dir
minimax_h3_fl2va_pruned_fp8_scaled.safetensors	20.96 GB	diffusion_models
minimax_h3_ref2va_pruned_fp8_scaled.safetensors	20.96 GB	diffusion_models
minimax_h3_fl2va_pruned_int8_convrot.safetensors	20.97 GB	diffusion_models
minimax_h3_ref2va_pruned_int8_convrot.safetensors	20.97 GB	diffusion_models
minimax_h3_fl2va_int8_convrot.safetensors	34.04 GB	diffusion_models
minimax_h3_ref2va_int8_convrot.safetensors	34.04 GB	diffusion_models
minimax_h3_fl2va_bf16.safetensors	66.28 GB	diffusion_models
minimax_h3_ref2va_bf16.safetensors	66.28 GB	diffusion_models

Minimax-h3-Turbo >6 files

Name	Size	Target Dir
minimax_h3_fl2v_turbo_4step_v1.0_768p_comfyui_bf16.safetensors	1.96 GB	loras
minimax_h3_fl2v_turbo_4step_v1.1_768p_comfyui_bf16.safetensors	1.96 GB	loras
minimax_h3_fl2v_turbo_4step_v1.2_768p_comfyui_bf16.safetensors	1.96 GB	loras
minimax_h3_fl2v_turbo_8step_v1.0_768p_comfyui_bf16.safetensors	1.96 GB	loras
minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensors	1.96 GB	loras
minimax_h3_ref2v_turbo_8step_v1.0_768p_comfyui_bf16.safetensors	1.96 GB	loras

Minimax H3 Latent Upscaler >2 files

Name	Size	Target Dir
minimax_h3_latent_upscaler_3d_fp16.safetensors	690.59 MB	latent_upscale_models
minimax_h3_latent_upscaler_3d_bf16.safetensors	690.59 MB	latent_upscale_models

MiniMax-H3-text_encoders >3 files

Name	Size	Target Dir
qwen3vl_32b_minimax_h3_nvfp4_awq.safetensors	15.69 GB	text_encoders
qwen3vl_32b_minimax_h3_int8_convrot.safetensors	27.14 GB	text_encoders
qwen3vl_32b_minimax_h3_bf16.safetensors	51.51 GB	text_encoders