appautomaton/minimax-h3-base-8bit-mlx

MiniMax-H3-Base — MLX (8-bit)

MLX affine 8-bit conversion of H3-Base, the open stage of MiniMax-H3 — text to synchronized video and stereo audio, denoised together in one packed sequence. Runs on Apple silicon through the mlx-h3 runtime with no PyTorch, CUDA, or cloud API at inference time.

Contents

File	Size	Role
dit_fl2va_a8g32.safetensors	34.8 GiB	DiT for text and 0–2 keyframes
dit_ref2va_a8g32.safetensors	34.8 GiB	DiT for reference conditioning
te_qwen3vl_a8g32.safetensors	27.7 GiB	Qwen3-VL-32B text encoder

Pull one with --include; you do not need all three.

The two DiTs are structurally identical and differ only in input packing. Load one, selected by conditioning mode — not both.

Quantization

MLX affine, 8 bits, group size 32 — the a8g32 in each filename. One scale and one bias per 32 weights, stored as bf16 alongside the packed integers.

This is not ComfyUI's int8_convrot, which is tensor-wise int8 plus a 256-group rotation and cannot be loaded from MLX. The distinction is format, not precision.

Layers stay dense where quantizing them costs accuracy for no useful saving: anything under two dimensions, and anything whose last axis is not divisible by the group size. 8-bit is the serving configuration here rather than a compromise — the workload is compute-bound, so lower precision buys memory without buying speed.

This is H3-Base only

MiniMax-H3 ships as three stages. Only the middle one is released.

Stage	Role	Here
H3-Context-IR	prompt → structured brief	No
H3-Base	768p joint audio-video generation	Yes
H3-Regenerate-2K	768p → 2K upscale	No

Two consequences worth knowing before you download 97 GiB:

• Nothing rewrites your prompt. The hosted product expands a one-line request into a structured brief first. Here the encoder sees exactly what you send. See the prompting guide.
• Output is 768p. The 2K figures in hosted guides describe the third stage, which is not open.

Not included

The VAEs and tokenizer are required at runtime and are not in this repository. They are unmodified and available upstream from Comfy-Org/MiniMax-H3:

video VAE   fp16   4.8 GiB
audio VAE   fp32   577 MiB
tokenizer          6.7 MiB

How to get started

hf download appautomaton/minimax-h3-base-8bit-mlx --local-dir weights/mlx-8bit

Runtime install and usage: mlx-h3 on PyPI · project page · GitHub

Requirements

Apple silicon with enough unified memory to hold one model at a time. The DiT and text encoder are never co-resident — the runtime loads one, materializes its output, releases it, and asserts the memory came back. Unstaged they would be 62.5 GiB of weights before a single activation. Default active-memory budget is 70 GiB; swap activity is treated as a failure, not a slowdown.

Links

• Source code: appautomaton/mlx-h3
• Package: mlx-h3 on PyPI
• Project page: appautomaton.renocrypt.com/mlx-h3
• More from App Automaton: Project · GitHub · Hugging Face

License

These files are modified — quantized derivatives of the original release, not the original weights.

The two DiTs are governed by the MiniMax H3 Community License Agreement, which applies to you as a recipient. It carries territorial limits and an acceptable-use policy; read it before use.

MiniMax H3 is licensed under the MiniMax H3 Community License Agreement, Copyright © 2026 MiniMax. All Rights Reserved.

text-encoder/ derives from Qwen3-VL-32B, licensed under Apache 2.0.

The mlx-h3 runtime code is separately licensed under MIT.