AI Video Generation
MiniMax's new video model does anime better, and that's the whole point
MiniMax's Hailuo 2.3 video model improves motion and facial expressions, expands style support for anime and CG, and ships with a full-modal Media Agent. The bet is that Asian visual culture, not just resolution, is the differentiator.
Emmanuel Fabrice Omgbwa Yasse AI-assisted
2026-07-07 · Last updated: 2026-07-15 · 2 min read
Chinese AI startup MiniMax has released Hailuo 2.3, an upgraded video model that goes after motion quality, stylistic range, and facial animation. The update builds on Hailuo 02 from earlier this year and keeps the same pricing for both enterprise and consumer tiers.
What got better
MiniMax says Hailuo 2.3 handles complex body movements more smoothly and obeys motion prompts more reliably. Lighting, shadow transitions, and color tones across dynamic camera shots now edge closer to real-world footage. The company also claims it sets a new global benchmark for cost-effectiveness, though that kind of framing is standard in every launch announcement. See also: Anthropic's approach to value.
The real shift is in style support. Hailuo 2.3 covers anime, illustrations, ink wash paintings, and game CGI, visual languages that matter in Asian markets. Earlier Hailuo 01 Live was already popular with anime creators, but this version broadens the palette without sacrificing stability.
Facial expressions and object motion
MiniMax emphasized that Hailuo 2.3 produces more natural facial performances and micro-expression changes. Those micro-expressions were the weak spot in earlier versions, faces that looked slightly off, tugging at a viewer's sense that something was wrong. The new version closes that gap. The model also responds better to object motion instructions, a capability tested during China's Singles' Day shopping festival, where beta users generated e-commerce ads with reportedly higher success rates. See also: MiniMax's work on making AI outputs feel less artificial.
Pricing and performance
Pricing stays unchanged from Hailuo 02. A faster variant, Hailuo 2.3 Fast, offers quicker generation at a lower price point, potentially cutting costs by up to 50% for batch production work. Free daily trial quotas run during the launch period.
The model is available across MiniMax's Hailuo AI web platform, mobile app, and open API.
Media Agent
Alongside the model, MiniMax introduced Media Agent, an evolution of the Hailuo Video Agent launched this summer. It accepts text prompts, images, video clips, and audio, then matches them into a coherent final piece without manual editing. A 'one-click to video' feature generates complete videos from user inputs. For professional creators, Media Agent supports step-by-step composition with free-form uploads of images, video, or audio, with customization on demand.
MiniMax demonstrated the feature by generating a 30-second advertisement for a fictional corn chip brand, 'Casa Nacho,' from a brief describing desired scenes, color tones, camera styles, and music. Future updates will let users adjust any part of the output pipeline via a canvas interface, a workflow MiniMax calls 'conversation as creation.'
Market context
Video generation has gotten crowded: OpenAI's Sora, Runway, Pika, and Kuaishou's Kling are all competing for creators' attention. MiniMax's bet is that pricing plus breadth of visual styles, especially those popular in Asian markets like anime and ink painting, is a differentiating wedge. Whether that wedge holds depends on execution quality, and MiniMax has shown incremental improvement rather than a leap. See also: MiniMax's earlier quiet achievement in coding and the broader platform play from Chinese AI companies.
Get the tech essentials in 3 minutes every morning
One email, every weekday, with what actually matters in AI and tech.