dt-minimax-h3
MiniMax H3 weights in the Draw Things (s4nnc .ckpt) format, mirrored
unmodified from the Draw Things model CDN for use by JoyFusion's on-device inference. The conversion and 6-bit
quantization were done by the Draw Things project; the files here are byte-identical to the upstream ones.
| File | Size | SHA-256 |
|---|---|---|
minimax_h3_fl2va_q6p.ckpt |
30,149,910,528 | 5b7bd906974a15df94327bc48b1dc6785102d4ed1dff1bee3bb3c99e72e84b49 |
minimax_h3_ref2va_q6p.ckpt |
30,150,074,368 | d78902fb91e09606871b9e1964db5237ca6de6b8b3dde497299f6b6ef8fa22c2 |
minimax_h3_vae_f16.ckpt |
5,511,926,272 | d77c20bb0e092eb25e11d7b711f87293217350f129fa49010e9ee309f83ad14d |
The shared text encoder qwen_3_vl_32b_50_i8x.ckpt (Qwen3-VL-32B, Apache-2.0) lives in
JoyFusionAI/dt-components.
Acceleration LoRAs
lightx2v Minimax-h3-Turbo 8-step distillation LoRAs
(Apache-2.0), converted from the *_768p_comfyui_bf16.safetensors release to the Draw Things LoRA format with
the Draw Things LoRAImporter (auto-detected as minimaxH3). Weight 1.0, 8 sampler steps, guidance 1.0.
| File | Base | Size | SHA-256 |
|---|---|---|---|
minimax_h3_fl2v_turbo_8step_v1.0_lora_f16.ckpt |
fl2va | 1,458,900,992 | 0b712fb32775244d3099af7eed81386246bc9e96d18250522d833f527ae6df24 |
minimax_h3_ref2v_turbo_8step_v1.0_lora_f16.ckpt |
ref2va | 1,458,900,992 | 407c3c682a0f5a7c2a30cefeeef1dcfb723d98e9b004cbddd3b25a13bd9697cd |
Source files: minimax_h3_{fl2v,ref2v}_turbo_8step_v1.0_768p_comfyui_bf16.safetensors
(SHA-256 08cfe946…bf9980 / 6a56f41a…928c5a).
License
MiniMax H3 is licensed under the MiniMax H3 Community License Agreement, Copyright © 2026 MiniMax. All Rights Reserved.
See LICENSE and NOTICE in this repository. Note in particular that the license's Applicable Territory excludes the
European Union, the United Kingdom, the Republic of Korea and the United States of America, and that commercial products
with more than USD 20 million in yearly revenue need separate authorization from MiniMax.
- Downloads last month
- -
Model tree for JoyFusionAI/dt-minimax-h3
Base model
MiniMaxAI/MiniMax-H3