Abiray releases GGUF quants of MiniMax-H3-Image
Original title: Abiray/MiniMax-H3-Image-GGUF — new model trending #29 on Hugging Face
Five pruned quants from 6.88 GB to 14.10 GB load in stock ComfyUI-GGUF and need a 24-channel VAE plus Qwen3-VL-32B.
Abiray published GGUF weights converted from Comfy-Org's pruned BF16 MiniMax-H3-Image checkpoint. The card describes a 33.1B-parameter single-stream diffusion transformer for text-to-image and multi-reference editing, sampled in 6 steps at up to 4096 × 4096. Sensitive layers stay in FP32. Users need ComfyUI v0.45.0 or later, ComfyUI-GGUF, a 24-channel image VAE, and Qwen3-VL-32B.
Unconfirmed: Visual-quality comparisons with full BF16 are the uploader's claims, not independently verified.
- Q2_K is 6.88 GB; Q4_K_M is 11.60 GB; Q5_K_M is 14.10 GB.
- The card recommends euler, 6 steps, and CFG scale 1.0.
- Files are said to load in stock ComfyUI-GGUF without custom patches.
- A dedicated 24-channel VAE is required; standard 4-channel VAEs are not.
- Vendors
- MiniMaxAI
- Products
- ComfyUIComfyUI-GGUF
- Organizations
- Comfy-Org
- AI models
- MiniMax-H3-ImageQwen3-VL-32B
From the original article
4 min read at huggingface.cotext-to-image · 127,803 downloads · 218 likes · license other
By Abiray · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.