Comfy-Org repackages MiniMax H3 Image for ComfyUI, with a separate revenue-based license
Comfy-Org has published repackaged MiniMax H3 Image model files for ComfyUI on Hugging Face. The MiniMax H3 Image license differs from the original MiniMax H3 license and requires separate written authorization above 1 million USD in annualized revenue or for hosted services.
Comfy-Org has published repackaged MiniMax H3 Image model files for ComfyUI on Hugging Face. The original model repository is MiniMaxAI/MiniMax-H3-Image.
Files and folder layout
The repackaged release places files in the following ComfyUI folders:
models/diffusion_models/:minimax_h3_image_pruned_int8_convrot.safetensorsmodels/text_encoders/:qwen3.5_9b_w4a8_gptq.safetensors,qwen3vl_32b_minimax_h3_int8_convrot.safetensorsandqwen3vl_32b_minimax_h3_nvfp4_awq.safetensorsmodels/vae/:minimax_h3_image_vae_fp16.safetensors
For prompt enhancement, Comfy-Org recommends the model qwen3.5_9b_w4a8_gptq used with the Generate Text node.
License terms
The MiniMax H3 Image license agreement is different from the original MiniMax H3 license. You must obtain separate, prior written authorization from MiniMax, requested through https://platform.minimax.io/license, in either of two cases:
- The annualized revenue of you and/or your affiliates exceeds 1 million USD.
- You provide hosted services to any third parties.
Hugging Face listing
The model tree lists MiniMaxAI/MiniMax-H3 as the base model and MiniMaxAI/MiniMax-H3-Image as a finetune of it. Two quantizations are listed. The page showed 18 downloads in the last month as of 2026-11-10. No Inference Provider deploys the model, and the page offers a link to request provider support. Context window, parameter count, benchmark scores and pricing are not listed in the feed summary.
What this means
This release is a packaging step for local use in ComfyUI, not a new model from MiniMax. The file names indicate int8 and NVFP4 variants of the text encoder and a pruned int8 diffusion model, so the files are prepared for ComfyUI workflows. The license is the main constraint: teams above the revenue threshold, and anyone offering the model as a hosted service, need MiniMax's written authorization before use. Because the H3 Image license differs from the original MiniMax H3 license, users should not assume terms carry over between the two.
Related Articles
Cursor adds multiplayer to Projects, rolling out in beta to all users
Cursor announced a multiplayer upgrade to Projects, its system for directing fleets of coding agents, on November 10, 2026. The beta is rolling out to all Cursor users. Teams can invite colleagues into a shared workspace where people and agents coordinate tasks, context and handoffs.
Google expands Gemini Intelligence to Pixel 10 and motorola signature 27, adds Create My Widget
Google is bringing Gemini Intelligence to Pixel 10, Pixel 10 Pro XL and Pixel 10 Pro Fold via a Pixel Drop, and to the motorola signature 27, which opens for U.S. pre-order on November 19. Task automation expands to the UK, India and Japan, and new features include Create My Widget and Device Assistance.
Google AI Edge Gallery adds on-device ASR and TTS, with ML Kit transcription across 100+ languages
Google introduced on-device Automatic Speech Recognition and Text-to-Speech in Google AI Edge Gallery on Nov. 10, 2026. ML Kit's ASR covers 100+ languages at 10x real-time speed, and its diffusion TTS begins speaking in about 100 ms. Cross-platform developers get open-weight speech bundles through MediaPipe Tasks and LiteRT.
Google ships Gemini Notebook API in public preview: REST and MCP server, source ingestion only
Google has released a public-preview REST API and MCP server for Gemini Notebook (formerly NotebookLM) that work with personal Google accounts. They can create notebooks and add web, YouTube, text and Markdown sources, but cannot yet query notebooks or generate outputs.
Comments
Loading...