MiniMax-H3-Tiny for ComfyUI

Project Page GitHub Collection

H3-Tiny transformers repackaged as single-file checkpoints for ComfyUI's native MiniMax H3 support. Part of H3-Tiny: structured pruning of MiniMax-H3 for joint video + audio generation.

Model Workflow Params File
MiniMax-H3-Tiny-17B-Pruned-Ref2VA Ref2VA 17.05B minimax_h3_tiny_17b_ref2va_bf16.safetensors · 34.2 GB
MiniMax-H3-Tiny-17B-Pruned-FL2VA FL2VA / T2VA 17.05B minimax_h3_tiny_17b_fl2va_bf16.safetensors · 34.2 GB
MiniMax-H3-Tiny-14B-OKD-Ref2VA-v0.1 Ref2VA 13.98B minimax_h3_tiny_14b_okd_ref2va_bf16.safetensors · 28.1 GB
MiniMax-H3-Tiny-17B-FewStep-Ref2VA-v0.1 Ref2VA 17.08B minimax_h3_tiny_17b_fewstep_ref2va_bf16.safetensors · 34.2 GB

FewStep samples in 6 steps (4 or more work) and needs our nodes, see below; the others use 50 steps with stock ComfyUI.

These are the same weights as the diffusers checkpoints, in ComfyUI's key layout. ComfyUI reads the pruned depth and FFN width from the tensors, so no custom node is needed.

Usage

  1. Put a file in ComfyUI/models/diffusion_models/.
  2. Get the text encoder and VAEs from Comfy-Org/MiniMax-H3: text_encoders/qwen3vl_32b_minimax_h3_*.safetensors → models/text_encoders/, vae/minimax_h3_video_vae_fp16.safetensors and vae/minimax_h3_audio_vae_fp32.safetensors → models/vae/.
  3. Open ComfyUI's MiniMax H3 workflow (MiniMax H3 Image to Video for FL2VA / T2VA, MiniMax H3 Reference to Video for Ref2VA) and select the H3-Tiny file in Load Diffusion Model.
  4. Sample without the turbo LoRA: 50 steps, simple scheduler, guider without CFG (BasicGuider). We tested euler.

Use the ref2va file with the Reference to Video node and the fl2va file with the Image to Video node (with first and last frames for FL2VA, without for T2VA).

We tested these files on ComfyUI commit afcb984 (2026-11-21) at 1344×768 and 243 frames on one H200: the outputs match our diffusers pipeline. The Comfy-Org turbo LoRAs were trained on the 33B model and are untested with H3-Tiny.

The files were made with scripts/convert_comfyui.py. It writes AdaLN in ComfyUI's curve form (adaln_t_table), which reproduces the diffusers AdaLN to about 1e-10 relative error.

FewStep

The few-step checkpoint also needs the step endpoint, which stock ComfyUI does not pass to the model. Install the H3-Tiny nodes:

git clone https://github.com/fuyi25/MiniMax-H3-Tiny
ln -s "$PWD/MiniMax-H3-Tiny/comfyui" ComfyUI/custom_nodes/ComfyUI-MiniMax-H3-Tiny

In the Reference to Video workflow, load the file with H3-Tiny Few-Step Loader instead of Load Diffusion Model, and feed H3-Tiny Few-Step Sigmas (6 steps by default) to SamplerCustomAdvanced, with the euler sampler and BasicGuider.

Contact

Questions and bug reports: GitHub issues or the Community tab here. For anything else, email Haoshuai Fu at fohoo@outlook.com.

License

These weights are Model Derivatives of MiniMax H3 and are distributed under the MiniMax H3 Community License Agreement; see NOTICE. Use of the weights must follow that agreement, including its territorial scope and acceptable-use policy.

H3-Tiny is an independent project and is not affiliated with or endorsed by MiniMax or Comfy Org.

Downloads last month
4,384
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for H3-Tiny/MiniMax-H3-Tiny-ComfyUI

Finetuned
(251)
this model

Collection including H3-Tiny/MiniMax-H3-Tiny-ComfyUI