ComfyUI Node

MiniMax H3

Native-audio video without the 42GB geofenced download

By Runware·Created 2 years ago·Updated a day ago· 140
MiniMax H3
  • referenceImages
  • frameImages
  • referenceVideos
  • video
positivePrompt
size768p (~16:9)
seed0
duration5
numberResults1
safetyfalse
safety.checkContentfalse
safety.modefast
ttlfalse
ttl_value60
outputFormatMP4
outputQuality95
advanced_json

MiniMax H3 is the model that broke the video-audio dam: an omni-modal 33B model that treats text, image, video, and audio as one input and generates video with native stereo sound - the first serious open-weights answer to Veo's audio advantage. The reception said it plainly: "Minimax H3 is very, VERY good. Unbelievably so," with day-zero ComfyUI support. And then the other shoe dropped, and it's why you're reading about a cloud node: H3's community license geofences the West - the US, EU, UK, and South Korea are excluded territories, so users there aren't licensed to run the local weights - and even where it's licensed, it's a 42.5GB download demanding serious VRAM.

This node sidesteps both walls. It reaches H3 through Runware's cloud, so you don't download anything, and you're using MiniMax's hosted H3 service rather than the territory-restricted local weights. That is genuinely the cleanest way most Western users will ever touch H3, and it's worth saying plainly: this is a case where the API isn't the lazy option, it's the legally unambiguous one.

The input set reflects what made H3 famous. positivePrompt is required and sentence-style, with Image 1/Video 1/Audio 1 asset references supported. referenceImages and frameImages give you I2V and frame conditioning; referenceVideos opens video-to-video motion transfer, the "like Kling" capability reviewers called out. Notably, there's no audio toggle to enable - H3 generates synchronized stereo audio natively, that's the model. size starts at 768p ~16:9 with presets up to 1440p, duration runs 5–15 seconds (default 5), and seed plus numberResults (1–4) round out the creative controls. safety with fast/full moderation modes, outputFormat (MP4/WEBM/MOV), outputQuality, and TTL settings fill the rest. Output is VIDEO.

Install

The pack-level setup applies as always: ComfyUI Manager → search Runware → install → restart, or git clone https://github.com/Runware/ComfyUI-Runware into custom_nodes plus pip install -r requirements.txt. Runware API key from the dashboard via Settings → Runware API key, RUNWARE_API_KEY, or runware auth login.

The decision, laid out

Three doors to H3-class output: the geofenced local weights (not an option for much of the West), this hosted node, and MiniMax's own API. The node wins when you want H3 inside a ComfyUI workflow with your normal graph around it - refs, upscaling, masking - without licensing or VRAM math. The price is per-call metering on video, which is real; watch the cost on the node's title bar and iterate at the small size/duration defaults before committing. Your prompt and media go through Runware and MiniMax's moderation. For native-audio, 2K-class video quality, that's the honest trade - and for many users it's the only compliant one.

CategoryRunware/Video/minimax

Inputs (16)

NameTypeDefaultDescription
positivePromptSTRINGText prompt describing elements to include in the generated output.
referenceImagesoptIMAGE
frameImagesoptIMAGE
referenceVideosoptRUNWARE_REFERENCEVIDEOS
sizeoptCOMBO768p (~16:9)Output size. Only one of resolution or dimensions is sent.
seedoptINT00–4294967295Random seed for reproducible generation. When not provided, a random seed is generated in the unsigned 32-bit range.
durationoptINT55–15Length of the generated video in seconds. The total number of frames produced is determined by duration multiplied by the model's frame rate (fps).
numberResultsoptINT11–4Number of results to generate. Each result uses a different seed, producing variations of the same parameters.
safetyoptBOOLEANfalseEnable to set safety. Off uses the model's default.
safety.checkContentoptBOOLEANfalseEnable or disable content safety checking. Increases total generation time.
safety.modeoptCOMBOfastSafety checking mode for video generation.
ttloptBOOLEANfalseEnable to set ttl. Off uses the model's default.
ttl_valueoptINT60Time-to-live (TTL) in seconds for generated content. Only applies when `outputType` is `URL`.
outputFormatoptCOMBOMP4File format for the generated video.
outputQualityoptINT9520–99Compression quality of the output. Higher values preserve quality but increase file size.
advanced_jsonoptSTRINGOptional JSON merged into the request. For: inputs.referenceAudios

Outputs (1)

NameTypeDescription
videoVIDEO