WeeTodd MiniMax H3 MLX for ComfyUI
Collection
Paged and quantized MLX artifacts for WeeTodd Nodes MiniMax H3 synchronized video-and-audio generation on Apple Silicon. • 3 items • Updated
Acknowledge the included model license
Review the license and eligibility requirements before requesting access.
Access is provided only to users who are eligible under every included license. Do not request access if your location or intended use is excluded.
Log in or Sign Up to review the conditions and access this model content.
MiniMax H3 MLX q8-extended paged transformer
Project: This checkpoint is built for WeeTodd Nodes, an MLX-native MiniMax H3 custom-node suite for ComfyUI on Apple Silicon. Installation, workflow examples, loader behavior, and current compatibility notes are maintained in that project. Browse the companion artifacts in the WeeTodd MiniMax H3 MLX collection.
This repository contains a modified MiniMax H3 FL2VA transformer for WeeTodd Nodes on Apple Silicon. It is not a complete MiniMax H3 checkpoint. The processor, tokenizer, Qwen3-VL text encoder, video VAE, and audio VAE are required separately.
The checkpoint uses the WeeTodd q8_extended mixed-precision recipe. The runtime retains fixed
tensors and loads four transformer blocks at a time. The paged layout reduces active transformer
weight residency without changing the selected checkpoint precision.
Compatibility
- WeeTodd Nodes commit
789de36or later - MLX 0.32.0 or later
- Apple Silicon
- MiniMax H3 FL2VA task
Place the complete repository directory at:
ComfyUI/models/MiniMax-H3/transformers/q8_extended_paged/
Select MiniMax-H3/transformers/q8_extended_paged as the transformer override in the WeeTodd H3
Component Loader. The loader detects paged_manifest.json automatically.
Use it with the paged Qwen3-VL conditioner and, optionally, the Q8 video VAE. The remaining H3 processor, tokenizer, and audio VAE must be supplied separately under their applicable licenses.
Quantization and paging
- Profile:
q8_extended - Quantization: MLX affine Q8, group size 64
- Quantized modules: 82
- Paged format:
weetodd-h3-paged-v1 - Transformer pages: 50 block files plus one fixed file
- Execution window: four consecutive blocks
- Tensor storage: approximately 33.38 GB
The quantized modules cover both MLP projections in blocks 21 through 37 and all four core projections in blocks 38 through 49. Other tensors retain their source precision.
Measured result
A clean 640 by 384 ComfyUI run used this transformer with paged Qwen, five schedule points, four transformer evaluations, low-memory BF16 staging, and direct publication. Complete-process peak was 14.951 GB. This result is a capacity measurement, not a native-resolution quality claim.
Provenance
- Base model:
MiniMaxAI/MiniMax-H3 - Base revision reviewed:
bfc8ed0353f5a9733be73e6b2c98ec0948195b86 - Source transformer SHA-256:
85f57a3a7a26def920999f4be786ebb91d0e677ddf327d1848b7724ca829d187 - Conversion implementation:
wee-todd/WeeTodd-Nodes - Conversion profile and page hashes:
quant_config.jsonandpaged_manifest.json
Each page SHA-256 is recorded in paged_manifest.json. Modified-file details are in
MODIFICATIONS.md.
License
The weights are a MiniMax H3 Model Derivative. They remain subject to the MiniMax H3 Community
License Agreement in LICENSE. Review its territorial, redistribution, notice, and acceptable-use
requirements before downloading or using this repository.
WeeTodd Nodes is an independent project and is not affiliated with MiniMax.
- Downloads last month
- -
Hardware compatibility
Log In to add your hardware
Quantized
Model tree for Vayden/MiniMax-H3-MLX-q8-extended-paged
Base model
MiniMaxAI/MiniMax-H3