MiniMax just raised the bar on AI video generation. Their new model, MiniMax H3, unifies text, images, video, and audio as one input and outputs 2K video clips with native stereo sound.
Launched on July 31, 2026, MiniMax H3 delivers clips between 4 and 15 seconds long, strictly in integer durations. It can process up to three reference videos, images, and audio clips in a single request. The output respects tight file size limits—videos top out at 50 MB, images at 30 MB, and audio at 15 MB.
MiniMax insists H3 is not just a text-to-video model with extras. Instead, it treats all input types as a single context. “Reference the camera movement from Video 1, have the character in Image 2 sing, match the vocals to Audio 3,” says Asif Razzaq, the article’s author. This is a leap in multimodal AI complexity.
Pricing is aggressive. At 2K resolution, the pay-as-you-go rate is $0.13 per second, making a 15-second clip about $1.95. MiniMax claims this is less than a third of mainstream model costs at 2K and less than half at 768p compared to 720p alternatives. The model is accessible via API and consumer app Hailuo AI, with weights set to drop soon under the MiniMax Community License.
Artificial Analysis ranks H3 as the most powerful AI for video editing, besting most rivals except Google’s Gemini Omni Flash in text-to-video tasks. In image-to-video, it trails only Seedance 2.0 and Gemini Omni Flash.
Other AI and Drone Innovations
Meshy AI announced its 3D Agent Beta on July 29, 2026. It claims to be the world’s first AI 3D agent, guiding users from early ideas to downloadable 3D models in one conversation. Features include creative brainstorming, batch concept generation, chat-to-3D creation, and a 3D knowledge base. Meshy AI has over 10 million users and has generated more than 100 million 3D models. The beta is live now.
Hi3D revealed an AI-powered 3D content platform aimed at manufacturing automation. It automates modeling, repair, splitting, color prep, and print optimization to turn digital concepts into physical products. New users get 300 free credits to test its AI-to-3D-print workflow. Hi3D integrates with hardware makers like FlashForge and Formlabs.
HoverAir launched Versa on July 31, 2026—a hybrid gimbal camera and drone for vloggers and filmmakers. Versa switches from handheld to hover mode in seconds with no learning curve or pilot skills. HoverAir calls it “the first 3-axis gimbal camera that can go from hand to hover in a matter of seconds.” Handheld mode resembles DJI’s Osmo Pocket 4 or Insta360 Luna but promises better dynamic range and low-light footage.
To fly, slide a back plate onto the camera and snap it onto the drone carrier. The drone unfolds and takes off. Versa will feature an AI follow-me mode and a 3D Worlds photo mode that does 360-degree flight and reconstructs photos in 3D. Details on camera sensor size, video resolution, lenses, and pricing remain under wraps.
Based on
- MiniMax Releases MiniMax H3: An Omni-Modal Video Model That Generates 15-Second 2K Clips With Native Stereo Audio — marktechpost.com
- Video AI: MiniMax challenges ByteDance with low price, open weights for new H3 model | South China Morning Post — scmp.com
- Meshy AI Announces Launch of 3D Agent Beta to Streamline Concept-to-Model Workflows | Currency News | Financial and Business News | Markets Insider — markets.businessinsider.com
- How Hi3D is turning AI creators into manufacturing entrepreneurs | VentureBeat — venturebeat.com
- HoverAir’s Versa Is A Pocket Gimbal Camera And Drone Mashup — engadget.com
