Skip to main content
Back to News
MiniMax H3 open-weight video generation model launches, free to test with local execution and Omni Reference
Product
2 min read
CN

MiniMax H3 open-weight video generation model launches, free to test with local execution and Omni Reference

The AMW Read

Open-weight video model with Omni Reference and local execution updates the multimodal media player map and advances the open-weight strategy, but builds on known MiniMax trajectory.
NoveltySignificance
Multimodal Β· Player MapScaling Laws
MiniMax
MiniMax

Foundation Models / LLMs

View Company Profile

MiniMax H3 open-weight video generation model launches, free to test with local execution and Omni Reference

Chinese AI lab MiniMax has released MiniMax H3, an open-weight video generation model that the company claims matches ByteDance's Seedance 2.0 in cloud service performance while remaining locally executable on consumer hardware. Testers report generating 15-second clips at 1248x832 resolution on an RTX 4090-powered PC in approximately 20 minutes per clip, with both photorealistic and anime-style output. The model's signature feature is Omni Reference, which accepts up to nine images, three videos, and three audio tracks as reference inputs to maintain consistency across generated content β€” a capability previously limited to closed cloud services.

The release represents a meaningful shift in the video generation market's competitive structure. MiniMax has positioned itself as the first open-weight provider to offer multi-modal reference capabilities that rival top closed models like Seedance 2.0 and OpenAI's Sora 2. The H3 release follows MiniMax's broader open-sourcing strategy, including the M3 multimodal flagship model in June, and arrives roughly one month after the company closed a $2.04 billion funding round amid a surge in investor interest in Chinese AI labs. The ability to run frontier-adjacent video generation locally also reduces dependence on cloud inference costs for developers and creators.

For builders, the practical implication is direct: teams can now integrate video generation into production workflows without per-token cloud fees, provided they have adequate local GPU resources. The Omni Reference capability also opens new product categories in character-consistent storytelling and brand-consistent marketing content. For investors, the release reinforces that open-weight Chinese models continue compressing the commercial moat around closed video generation APIs, pressuring pricing across the segment. The company's mainland China listing preparation, disclosed after its Hong Kong shares surged 400% per the AI Market Watch index tracking roughly 5,000 companies, suggests capital markets are rewarding this open-ecosystem approach.

#MiniMax #VideoGeneration #OpenWeights #AIGenerativeMedia #ChinaAI #LocalAI

#MiniMax H3#video generation#open-weight model#Omni Reference#local AI inference#Seedance 2.0

How This Connects

Based on Multimodal Β· Player Map

  1. 4h agoKling AI Secures Record $3B in Independent Funding for Video GenerationKling AI
  2. 1d agoAlibaba has released the beta of its Wan3.0 video generation model, enabling creation of clips up to...Alibaba Wan3.0
  3. 2d agoMiniMax H3 open-weight video generation model launches, free to test with local execution and Omni Reference Β· THIS ARTICLE
  4. 3w agoNetflix Paid $587 Million Cash for Ben Affleck's AI Startup InterPositiveInterPositive
  5. 1mo agoKling AI spins out from Kuaishou, raises $3B at $15B pre-money valuation in record multimodal funding roundεŒ—δΊ¬ε―η΅

Related News

More news from MiniMax

Stay updated with the latest news and announcements from MiniMax.

View all MiniMax news

Discover AI Startups

Explore 5,000+ AI companies with VC-grade analysis, funding data, and investment insights.

Explore Dashboard