MiniMax H3 Studio assets (mirror)

Everything the H3 Studio web frontend needs on top of the base MiniMax H3 models, collected in one place so a Colab notebook can fetch it with a single hf_hub_download loop. The folder layout mirrors ComfyUI/models/, so each file goes to the same relative path.

The base models (the UNET, the text encoder and both VAEs) are not here. Get them from their original repos.

This repo is a mirror. Each file belongs to its original author and keeps that author's license and terms. The table below lists where each file came from. Every file is byte-identical to its source; the only exception is the rewriter, which is my own quantization.

File (path in this repo) Size Original source
loras/minimax_h3_dmad_full_dareties_v4_step600_comfy_fro.safetensors 957 MB Civitai model 2837571 (DMAD: several H3 turbo LoRAs merged with DARE-TIES)
loras/minimax_h3_turbo_v4_step600_ema_DasiwaREF2VAHybridV1_0_curveproj1025_compat_v001.safetensors 795 MB t8star/MinimaxH3-Dasiwa-10eros-hybird-8steps-turbo
loras/MysticXXX_MMH3-V2.safetensors 172 MB lynaNSFW/mysticxxx_MM_H3
loras/H3_Combat_V2.safetensors 155 MB JOKER141/MiniMax-H3-Combat-Base-V2
loras/Motion_Repair.safetensors 155 MB JOKER141/MiniMax-H3-General-Motion-Continuity-Repair
loras/H3_SPEED_slider.safetensors 149 MB siraxe/H3_slider_experiments
semantic_bridge/t8_compat/BUNNY_H3_ActionLogic_Bridge_V1_T8_Compat.safetensors 22 MB t8star/Semantic-Bridge-Comfy
latent_upscale_models/minimax_h3_latent_upscaler_3d_fp16_V2.safetensors 691 MB LBH-123-AI/Minimax_h3_latent_Upscaler (V2 file also at KENIC-1/minimax_h3_latent_upscaler_3d_fp16_V2)
text_encoders/minimax_h3_rewriter_ref2va_qwen3.5_9b_w4a8.safetensors 7.3 GB W4A8 quantization by dseditor of lightx2v/MiniMax-H3-Prompt-Rewriter-Qwen3.5-9B-Ref2VA (Apache-2.0, fine-tuned from Qwen3.5-9B). It loads in ComfyUI's built-in CLIPLoader and runs through TextGenerate.

Custom nodes these files are for

If you are an author and want a file removed from this mirror, open a discussion and it will be taken down.

Downloads last month
1,815
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support