awesome-minimax-h3-integration README

The project's README from github.com/MiniMax-AI/awesome-minimax-h3-integration. As one markdown file: README.md.

MiniMax H3

MiniMax H3 Integrations

A community-maintained index of checkpoints, tools, and workflows for MiniMax H3, ordered by developer interest.

Official resources

Start here

This navigation guide is not a complete compatibility list.

Goal Start with
Run on your GPU Run locally — pick a stack from the VRAM table
Work with audio comfyui-minimax-h3-audio-T8 and audio VAE
Build in ComfyUI Official tutorial · Workflows & nodes
Write better prompts Prompting
Make it faster Speed
Fine-tune it Training & LoRAs
Serve it as an API Serving H3
Run on Apple Silicon antirez/h3.c

Run it locally

MiniMax-H3 generates video with native stereo audio from text, images, video, and audio inputs. It supports clips up to 2K and 15 seconds. Two base variants exist:

Checkpoints are the same size. FL2VA, trained only with keyframes, typically yields better raw output. Ref2VA accepts more reference material but has lower base quality; the Ref Patch can partially bridge this gap.

By VRAM and hardware

Find your GPU in the table, then use the notes to inform your configuration.

Situation Stack Why this combination
24 GB, first run pruned_int8_convrot DiT (19.53 GiB) + TE nvfp4_awq (14.61 GiB) + ComfyUI-MiniMaxH3-Easy Easy routes T2V, I2V, first/last-frame, and reference input through a single Media port. Sampling, LoRAs, and decoding remain outside the node for later modification.
24 GB, want speed The above + TE-Speed-MiniMaxH3 + Turbo v4_step600_ema at 6–8 steps Block-cache acceleration; v3.2 adapts to current ComfyUI's block prefetch and no longer patches core files. 6–8 steps reduce Turbo motion smear.
22 GB, Turing (sm_75) minimax-h3-turing — W4A8 mixed + Turbo 4-step field handbook 5.7 min/clip baseline; T8 BlockCache drafts −43%; PDD 8-step + T8 combo 210 s/clip. sm_75 has no FP8/BF16 tensor cores — quantization, launch flags and kernel support all differ from Ampere+
12–16 GB Pruned Q4_K_M GGUF (10.64 GiB) or pruned nvfp4 (11.67 GiB) + TE Q2_K (7.91 GiB) + fp8mix VAE pair GGUF offers the most size options, beneficial for tight memory. IQ1_S is smaller at 3.78 GiB, but quality noticeably drops.
8 GB DiffSynth-Studio NF4 path The project states 8 GB as its minimum for this path. Offloading performs most work here; expect slow performance, not just small memory footprint.
RTX 50-series / Blackwell NVIDIA Sol-Attn 1.14–1.44× faster than SageAttention with −37 % MLP peak VRAM, measured on a 5090. SM89–SM121, Triton 3.6.0. Also unlocks Blackwell-only hybrid-NVFP4 checkpoints.
Multi-shot / long video ComfyUI-H3-Motion-Context H3 generates in blocks up to 15 s. Motion-Context feeds the previous block's final frame and audio forward, preserving motion direction and speed.
Storyboard / timeline ComfyUI_MiniMaxH3_Director Five importable templates: t2v, fl2v, r2v, v2v, and rv2v.
Inpaint / local edit scraed/LanPaint v2.1.0 fixed H3 support. Training-free video and audio inpainting.
Apple Silicon antirez/h3.c (MIT, Metal-native) h3.c supports T2V/A, first-last-frame, and ordered Ref2VA references end-to-end, with M3 Max / M5 Max performance optimization ongoing.
One-command local open-video-ai/open-video "Ollama for video models" — install · pull · run.

Checkpoints

Source What it is Files Total
MiniMaxAI/MiniMax-H3 Original diffusers weights — transformer/ (FL2VA) and transformer_ref/ (Ref2VA) at 14 shards / 61.73 GiB each, plus text encoder, video VAE, audio VAE, and self-contained FL2VA/ and Ref2VA/ pipeline folders 280 464.2 GiB
Comfy-Org/MiniMax-H3 ComfyUI-repackaged single-file weights — 10 diffusion models, 3 text encoders, video + audio VAE 17 433.2 GiB
Variant Name Precision Size Download
FL2VA minimax_h3_fl2va ![bf16][badge-bf16] 61.73 GiB ![][gh-Comfy--Org]
FL2VA minimax_h3_fl2va ![int8][badge-int8] 31.70 GiB ![][gh-Comfy--Org]
FL2VA minimax_h3_fl2va_pruned ![bf16][badge-bf16] 37.46 GiB ![][gh-Comfy--Org]
FL2VA minimax_h3_fl2va_pruned ![fp8][badge-fp8] 19.52 GiB ![][gh-Comfy--Org]
FL2VA minimax_h3_fl2va_pruned ![int8][badge-int8] 19.53 GiB ![][gh-Comfy--Org]
Ref2VA minimax_h3_ref2va ![bf16][badge-bf16] 61.73 GiB ![][gh-Comfy--Org]
Ref2VA minimax_h3_ref2va ![int8][badge-int8] 31.70 GiB ![][gh-Comfy--Org]
Ref2VA minimax_h3_ref2va_pruned ![bf16][badge-bf16] 37.46 GiB ![][gh-Comfy--Org]
Ref2VA minimax_h3_ref2va_pruned ![fp8][badge-fp8] 19.52 GiB ![][gh-Comfy--Org]
Ref2VA minimax_h3_ref2va_pruned ![int8][badge-int8] 19.53 GiB ![][gh-Comfy--Org]

"Pruned" means AdaLN-pruned. These models are roughly 40% smaller, work only in ComfyUI, and underpin most of the consumer-GPU quants below. For a 24 GB card, pruned_int8_convrot (19.53 GiB) is the usual starting point.

Quantized Models

MiniMax provides original BF16 checkpoints. The files below are community conversions and repackaged variants, not official MiniMax releases. Confirm compatibility with your runtime's documentation before downloading.

Community FL2VA conversions
Pruned Precision Method Size Download
![bf16][badge-bf16] BF16 61.73 GiB ![][gh-Comfy--Org]
![bf16][badge-bf16] Hybrid (fl2va base + ref2va adaln b15-49) 20.97 GiB ![][gh-smhfacct]
![bf16][badge-bf16] Hybrid (fl2va base + ref2va adaln b20-49) 20.97 GiB ![][gh-smhfacct]
![bf16][badge-bf16] Hybrid (fl2va base + ref2va adaln b25-49) 20.97 GiB ![][gh-smhfacct]
![bf16][badge-bf16] Hybrid (fl2va base + ref2va adaln b30-49) 20.97 GiB ![][gh-smhfacct]
![int8][badge-int8] ConvRot 31.70 GiB ![][gh-Comfy--Org]
![int8][badge-int8] ConvRot Lean (HQ) 21.91 GiB ![][gh-DmitryDB]
![int8][badge-int8] ConvRot 20.94 GiB ![][gh-DmitryDB]
![int8][badge-int8] ConvRot Lite 20.33 GiB ![][gh-DmitryDB]
![fp8][badge-fp8] FP8 E4M3FN 43.78 GiB ![][gh-rzgar]
![mxfp8][badge-mxfp8] MXFP8 44.34 GiB ![][gh-rzgar]
![fp8][badge-fp8] FP8 + FP16 attention 26.70 GiB ![][gh-rzgar]
![nvfp4][badge-nvfp4] NVFP4 (HQ) 13.60 GiB ![][gh-DmitryDB]
![nvfp4][badge-nvfp4] NVFP4 10.86 GiB ![][gh-DmitryDB]
![nvfp4][badge-nvfp4] NVFP4 32.05 GiB ![][gh-rockerBOO]
![int4][badge-int4] NF4 (DiffSynth) 15.98 GiB ![][gh-DiffSynth-Studio]
OrbitQuant W4A4 17.03 GiB ![][gh-WaveCut]
![int8][badge-int8] ⚠️ DT-sQKV ConvRot 21.00 GiB ![][gh-DmitryDB]
![int8][badge-int8] ⚠️ DT-sQKV ConvRot Lean 27.99 GiB ![][gh-DmitryDB]
✓ ![bf16][badge-bf16] BF16 37.46 GiB ![][gh-Comfy--Org]
✓ ![fp8][badge-fp8] FP8 scaled 19.52 GiB ![][gh-Comfy--Org]
✓ ![int8][badge-int8] ConvRot 19.53 GiB ![][gh-Comfy--Org] ┊ ![][gh-Abiray]
✓ ![nvfp4][badge-nvfp4] NVFP4 18.69 GiB ![][gh-rockerBOO]
✓ ![nvfp4][badge-nvfp4] NVFP4 + ConvRot INT8 18.69 GiB ![][gh-rockerBOO]
✓ ![nvfp4][badge-nvfp4] NVFP4 11.67 GiB ![][gh-Abiray]
✓ ![int4][badge-int4] Mixed INT4/INT8 ConvRot 14.81 GiB ![][gh-Abiray] ┊ ![][gh-tsolful]
✓ ![int4][badge-int4] Mixed INT4/INT8 ConvRot Lean 17.27 GiB ![][gh-tsolful]
✓ ![int4][badge-int4] INT4 ConvRot 15.67 GiB ![][gh-rockerBOO]
✓ ![int4][badge-int4] Mixed INT4/INT8 ConvRot 18.92 GiB ![][gh-rockerBOO]
✓ ![int4][badge-int4] INT4 ConvRot 16.67 GiB ![][gh-Merserk]
✓ ![int4][badge-int4] INT4 ConvRot (pruned) 10.56 GiB ![][gh-Merserk]
✓ ![int4][badge-int4] W4A8 ConvRot 11.68 GiB ![][gh-AX1Y2JP] ┊ ![][gh-Kijai] ┊ ![][gh-Winnougan]

GGUF quants — see the GGUF section.

Community Ref2VA conversions
Pruned Precision Method Size Download
![bf16][badge-bf16] BF16 61.73 GiB ![][gh-Comfy--Org]
![int8][badge-int8] ConvRot ┊ (patchin HF 1.02) 31.70 GiB ![][gh-Comfy--Org] ┊ ![][gh-t8star]
![int8][badge-int8] ConvRot Lean (HQ) 21.91 GiB ![][gh-DmitryDB]
![int8][badge-int8] ConvRot 20.94 GiB ![][gh-DmitryDB]
![int8][badge-int8] ConvRot Lite 20.33 GiB ![][gh-DmitryDB]
![nvfp4][badge-nvfp4] NVFP4 (HQ) 13.60 GiB ![][gh-DmitryDB]
![nvfp4][badge-nvfp4] NVFP4 10.86 GiB ![][gh-DmitryDB]
![nvfp4][badge-nvfp4] NVFP4 32.05 GiB ![][gh-rockerBOO]
![nvfp4][badge-nvfp4] NVFP4 mixed 22.76 GiB ![][gh-Abiray]
![int4][badge-int4] NF4 (DiffSynth) 15.98 GiB ![][gh-DiffSynth-Studio]
OrbitQuant W4A4 17.03 GiB ![][gh-WaveCut]
![nvfp4][badge-nvfp4] Hybrid NVFP4, FFN-only (Blackwell) 16.38 GiB ![][gh-abakanai]
![nvfp4][badge-nvfp4] Hybrid NVFP4, QKV+FFN (Blackwell) 14.03 GiB ![][gh-abakanai]
![int8][badge-int8] ⚠️ DT-sQKV ConvRot 21.00 GiB ![][gh-DmitryDB]
![int8][badge-int8] ⚠️ DT-sQKV ConvRot Lean 27.99 GiB ![][gh-DmitryDB]
✓ ![bf16][badge-bf16] BF16 37.46 GiB ![][gh-Comfy--Org]
✓ ![fp8][badge-fp8] FP8 scaled 19.52 GiB ![][gh-Comfy--Org]
✓ ![int8][badge-int8] ConvRot 19.53 GiB ![][gh-Comfy--Org] ┊ ![][gh-Abiray]
✓ ![nvfp4][badge-nvfp4]

awesome-minimax-h3-integration (MiniMax-AI/awesome-minimax-h3-integration)

Searching Database…

这是什么?一个由社区维护的 MiniMax H3 权重、工具和工作流索引。MiniMax H3 是一个视频生成模型,能从文本、图像、视频和音频输入生成带原生立体声的短片。
这是一个工具,还是只是一份链接列表?它是一份精心整理的参考列表(awesome-list),指向 MiniMax H3 的官方和社区权重、量化版本、ComfyUI 节点和工作流——它本身不运行任何东西。
这份列表是怎么组织的?开头是一张按目标分类的表(本地运行、使用音频、在 ComfyUI 中搭建、编写提示词、加速、微调、通过 API 提供服务、在 Apple Silicon 上运行),然后是一张 VRAM 表,把显存大小对应到推荐的权重和文本编码器组合。
运行 MiniMax H3 需要多少显存?从 8 GB(DiffSynth-Studio NF4 路径,大量卸载)到标准剪枝 int8 ConvRot 权重的 24 GB+ 不等;README 里的 VRAM 表把具体的显卡档次对应到权重/文本编码器组合。
这些量化权重是 MiniMax 官方发布的吗?不是。只有 MiniMaxAI 的原始 BF16 diffusers 权重是官方的;列表里所有 INT8/FP8/NVFP4/GGUF 转换版本都是社区重新打包的,下载前应先确认与你的运行环境兼容。
它只支持 NVIDIA 显卡,还是也能在 Mac 上用?大多数方案面向从 Turing(sm75)到 Blackwell/RTX 50 系的 NVIDIA 显卡,但也有通过 antirez/h3.c 提供的 Apple Silicon 原生 Metal 路径,M3/M5 Max 优化仍在进行中。
FL2VA 和 Ref2VA 有什么区别?FL2VA(首尾帧模式)接受零张、一张或两张输入图像,通常原始画质更好;Ref2VA(全参考模式)最多接受 9 张图像、3 个视频片段和 3 个音频片段,条件信息更丰富但基础画质略低,Ref Patch 能部分弥补这一点。
MiniMax H3 能在生成视频的同时生成音频吗?能,它可以直接从文本、图像、视频和音频输入生成带原生立体声的视频,为此还有专门的 comfyui-minimax-h3-audio-T8 节点和一个音频 VAE。
生成的短片有多长、分辨率多高?短片最高可达 2K 分辨率、15 秒时长,ComfyUI-H3-Motion-Context 之类的工具可以把多个块串联起来,生成更长、多镜头的视频。
有没有一条命令就能在本地运行 MiniMax H3 的办法?有,open-video-ai/open-video 被称为“视频模型界的 Ollama”——安装后拉取模型,一条命令即可运行。
Creator
MiniMax-AI
Size
0.2 MB
Stars
446
Updated
CommitOct 17
Age
2mo
Commits · 90 Days