Workflows/Generate MiniMax H3 video with audio in ComfyUI

Video / WORKFLOW

Generate MiniMax H3 video with audio in ComfyUI

Generate a short local MiniMax H3 video with synchronized sound in ComfyUI using the supported NVIDIA or Apple Silicon package for your machine.

Verified sourceSource checked 9/29/2026

LOCALRENTED GPU

Freshness 75 out of 100. Release age: 73 days old, Aug 3, 2026. Measures age, not quality.

Popularity 90 out of 100. 3,827,783 downloads in 30 days; 6,149 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.

What do these scores mean?

Two ways to compare models. Neither score measures answer quality, benchmark performance, or whether a model fits your machine.

Freshness · release age

A new release starts at 100. The score halves every 180 days: 100 × 2^(−age in days / 180). When the release date is unavailable, a repository date is a fallback—not proof of when the model was released. Missing dates stay unscored.

Check the date source

Popularity · adoption

Downloads over the past 30 days carry 80% of the score; lifetime likes carry 20%. Both use a logarithmic scale, capped at 10 million downloads and 100,000 likes. If likes are unavailable, downloads carry all the weight. Missing download statistics stay unscored.

round(80 × min(1, log10(1 + downloads) / 7) + 20 × min(1, log10(1 + likes) / 5))

Check the statistics source

Scores as of Oct 15, 2026. Statistics checked Oct 15, 2026. Statistics are marked for refresh after 72 hours.Full scoring method

Before you begin

Hardware

Compare the exact ComfyUI package for your machine. The NVIDIA numbers are memory estimates; the MacMax timing comes from its maintainer’s 48 GB M5 Pro test.

Machine	H3 pruned INT8 + 8-step	H3 MacMax Q5 GGUF + Parasyte
NVIDIA GPU · 12 GB + 64 GB RAM	Runs with heavy CPU offload; expect a long render	Mac-only package; use the NVIDIA column on this machine.
NVIDIA GPU · 16–24 GB + 64 GB RAM	Runs with CPU offload; more GPU headroom	Mac-only package; use the NVIDIA column on this machine.
NVIDIA GPU · 32 GB + 64 GB RAM	Recommended single-GPU starting point	Mac-only package; use the NVIDIA column on this machine.
Rented NVIDIA GPU · 48 GB+	More room for the full pipeline; check price	Mac-only package; use the NVIDIA column on this machine.
AMD Radeon · 24 GB	No documented ROCm route for this exact INT8 ConvRot package; use or rent a supported NVIDIA GPU.	Mac-only package; use the NVIDIA column on this machine.
Intel Arc · 16 GB	No documented Intel GPU route for this exact INT8 ConvRot package; use or rent a supported NVIDIA GPU.	Mac-only package; use the NVIDIA column on this machine.
Apple Silicon · 48 GB unified memory	Official NVIDIA INT8 template has a reported Mac MPS failure; choose MacMax instead.	Community-tested on 48 GB M5 Pro; ~7 min for a 5-second 352×608 clip
CPU only · 64 GB system RAM	No documented CPU-only route for this 42.5 GB video pipeline; use a supported NVIDIA GPU or rented GPU instead.	Mac-only package; use the NVIDIA column on this machine.

The MacMax performance is a third-party single-run measurement, not a speed promise. Its author has not validated smaller-memory Macs.

VIDEO + AUDIO / COMFYUISOURCE-LINKED SETUP

Make a MiniMax H3 clip on NVIDIA or Apple Silicon.

Choose the official NVIDIA ComfyUI template or the separately documented MacMax Apple Silicon canvas. Each route lists its exact model files, importable JSON, first-render settings and a playable video-with-audio success check.

Choose a package Load the visual workflow Check the result

BEFORE YOU BEGIN

Choose your package

Select the weights you want. The instructions, downloads and settings change with your choice.

Save your machine in My Hardware to get an automatic starting choice. You can always choose any package yourself.

CURRENT SETUP

MiniMax H3 · pruned INT8 + 8-step LoRA · official

ComfyUI reports about 42.5 GB total memory footprint for its smallest H3 variants and shows dynamic GPU offload down to an RTX 3060. On a 32 GB GPU, allow system RAM for offloaded components; 64 GB RAM is a cautious starting recommendation. These are not measured speed tiers.

Freshness 74 out of 100. Repository age: 77 days old, Jul 30, 2026. Measures age, not quality.

Popularity 93 out of 100. 24,344,675 downloads in 30 days; 2,292 lifetime likes. Checked Oct 15, 2026. Measures adoption, not quality.

What do these scores mean?

Two ways to compare models. Neither score measures answer quality, benchmark performance, or whether a model fits your machine.

Freshness · release age

Check the date source

Popularity · adoption

round(80 × min(1, log10(1 + downloads) / 7) + 20 × min(1, log10(1 + likes) / 5))

Check the statistics source

Scores as of Oct 15, 2026. Statistics checked Oct 15, 2026. Statistics are marked for refresh after 72 hours.Full scoring method

View weight source

THE WORKFLOW

Set up MiniMax H3 · pruned INT8 + 8-step LoRA on your machine

Choose your computer first, then its GPU path. Follow the visual setup and import the linked workflow.

CHOOSE YOUR SETUPWindows

Use current ComfyUI with a working NVIDIA driver. The preferred INT8 convrot files require a compatible PyTorch/CUDA build; follow ComfyUI’s version guidance if a kernel fails.

Other operating systems

Mac: The exact official INT8 ConvRot and NVFP4 ComfyUI template has a reported Apple MPS integer-matrix kernel failure. Use the separate Mac-specific H3 ComfyUI workflow on a 48 GB Apple Silicon machine instead; do not load this NVIDIA package as though it were the Mac path. Compatibility source

Install and update ComfyUI

Install ComfyUI for Windows from the official download page and update it to version 0.30.0 or later. Launch it and confirm the NVIDIA GPU is detected.

Official ComfyUI installation MiniMax H3 ComfyUI requirements

Download the exact H3 files

Place these five files in the matching folders under ComfyUI/models. They include two VAEs and the eight-step LoRA; the diffusion checkpoint alone is not a runnable pipeline. Restart ComfyUI so its loader nodes find them.

Put these files in ComfyUI/models/5 required files

diffusion_models/

minimax_h3_fl2va_pruned_int8_convrot.safetensorsPruned, quantized H3 video modelGet file

text_encoders/

qwen3vl_32b_minimax_h3_nvfp4_awq.safetensorsPrompt and multimodal encoderGet file

vae/

minimax_h3_video_vae_int8_convrot.safetensorsVideo decoderGet file minimax_h3_audio_vae_fp32.safetensorsAudio decoderGet file

loras/

minimax_h3_fl2v_turbo_8step_v1.0_comfyui_bf16.safetensorsEight-step acceleration adapterGet file

Import the official text-to-video template

Download the JSON and drag it into the ComfyUI canvas, or open Template Library → Video → MiniMax H3 Text to Video. Check each loader against the filenames above. If a node is red, update ComfyUI before changing the graph.

Download the official H3 text-to-video workflow Read the official node walkthrough

Set a simple first shot

In the official MiniMax H3 text-to-video template, select the pruned INT8 diffusion model. For a first preview, choose 9:16, set the Resolution Selector to 0.2 megapixels with a multiple of 32, and choose the shortest available duration. Enable turbo_mode and set turbo_steps to 8 (8 steps) to use the downloaded Lightning LoRA. Enter one subject, one movement, and one sound; raise size or length only after the short clip succeeds.

Queue and inspect the clip

Click Queue Prompt, wait for the graph to finish, then play the saved video and listen for the synchronized audio. If memory runs out, close other GPU apps and reduce output size or duration before trying the full-size canvas.

SUCCESS CHECK · MiniMax H3 · pruned INT8 + 8-step LoRA

A video with synchronized sound

ComfyUI finishes the queue and saves a playable local video. Watch it and listen for the audio described in your prompt; only then raise the duration or resolution.

This guide has two different ComfyUI packages. NVIDIA uses the official INT8 canvas; Apple Silicon uses the MacMax community GGUF canvas. Do not mix their model files or node graphs. Review MiniMax H3 model and community-node licenses before commercial use.

Back to setup steps

When it doesn’t go to plan

A red node or missing model file appears when the template loads.

Update ComfyUI, restart it, and match every loader to the exact filenames and model folders shown in the file tree.

Generation runs out of GPU memory or stalls during loading.

Close other GPU-heavy apps, keep the 3-second preview, lower resolution, and ensure 64 GB system RAM is available for offload; use a larger NVIDIA rental GPU if needed.

The result has no usable sound or the video is black.

Check both video and audio VAE loaders and the eight-step LoRA, then rerun the short preview without changing the template graph.

The Mac canvas shows missing nodes or a model loader cannot find its file.

Install the MacMax node packs from Terminal, restart ComfyUI, and check the five exact Mac-specific filenames under ComfyUI/models. Do not substitute the NVIDIA INT8/NVFP4 files.

REFERENCE LIBRARY

Sources and files

Original instructions, model files, and compatibility notes behind this setup.

6 SOURCES

01Official ComfyUI H3 guidedocs.comfy.org 02Official H3 text-to-video templateraw.githubusercontent.com 03ComfyUI model fileshuggingface.co 04ComfyUI local-memory statementblog.comfy.org 05MacMax Apple Silicon guide, downloads and measurementsgithub.com 06MacMax ComfyUI JSONraw.githubusercontent.com

The model behind this workflow

MiniMax H3

Keep exploring

Text

SETUP GUIDE

Qwen3.5 · 9B private chat setup

Source-backed package and platform-route plan for private local chat with Qwen3.5 · 9B. Preserves the official Ollama 9B package, a smaller 4B family fallback, and a clearly labeled community uncensored GGUF option.

Freshness 41 out of 100. Release age: 230 days old, Feb 27, 2026. Measures age, not quality.

Build a private coding assistant with Qwen3.8 27B

Run Qwen3.8 27B with Ollama, connect it to Cline in VS Code, and make one reviewable code change in a tiny local project.

Freshness 76 out of 100. Repository age: 71 days old, Aug 5, 2026. Measures age, not quality.

Arena Open Source Web coding: rank 10 of 55. Ranked model: qwen3.8-27b. Original model ranking, not a benchmark of this package. Export dated Oct 13, 2026. Source checked Oct 15, 2026. Human preference ranking, not hardware suitability.+3

SOFTWAREOllama + Visual Studio Code + Cline + Python

Create images locally with FLUX.2 Klein

Generate images from text prompts in ComfyUI with the fast distilled or more adjustable base 4B model. Save your results and reuse the visual workflow.

Freshness 35 out of 100. Release age: 273 days old, Jan 15, 2026. Measures age, not quality.

SOFTWAREComfyUI

LOCALRENTED GPU

Verified source