MiniMax H3 Open Video Model Puts Omni Reference on Local PCs
An open model with Seedance 2.0-class quality and Omni Reference, previously a cloud-only selling point, now runs locally on consumer hardware.
Reporting from 1 source: ASCII.jp.
Chinese AI firm MiniMax released MiniMax H3, an open video generation model that matches ByteDance's Seedance 2.0 in cloud performance. Its Omni Reference feature accepts up to nine images, three videos, and three audio tracks, and it runs on a local PC with an RTX 4090. ASCII.jp tested it with live-action and anime-style prompts.
MiniMax H3, released by the Chinese company MiniMax, is an open video generation model that ASCII.jp tested extensively. The cloud version performs on par with ByteDance's Seedance 2.0, and even the compressed ComfyUI version retains most of its capability.
The standout feature is Omni Reference, which lets the model reference up to nine images, three videos, and three audio tracks while keeping output consistent with them. Closed cloud video AIs already use this feature as a selling point, but this is the first time an open model has included it.
ASCII.jp generated a 15-second clip from a Japanese summer scene prompt, in both live-action and anime styles, on a PC with an RTX 4090. The render took about 20 minutes. The model handles Japanese scenery and motion naturally, and on a small screen the result is hard to distinguish from real footage.
Synthesized by Yomimono from the 1 cited source below, including Japanese-language reporting where cited, then editorially reviewed before publishing.