Generation video Hailuo 2.3

Hailuo 2.3 is MiniMax's flagship video model and the state-of-the-art on instruction following — the precision with which the output matches the prompt's specific requests. Where other models tend to ignore or paraphrase fine-grained directions ('zoom in slowly while panning left'), Hailuo executes them.

The family ships in two variants. Hailuo 2.3 is the full model with both text-to-video and image-to-video modes; pricing is per-clip rather than per-second — 18 credits for a 6-second clip and 30 credits for 10 seconds. Hailuo 2.3 Fast is image-to-video only at lower cost (11 credits for 6s, 18 for 10s) — use it when you have a hero image and just want it animated, no other knobs.

Both variants support two resolutions: 768p (the default, available at 6 or 10 seconds) and 1080p (locked to 6 seconds). Hailuo uses the source image's native aspect ratio rather than offering a ratio picker — what you upload is what you get.

The standout capability is the built-in camera command vocabulary. Hailuo recognises 15 named camera moves (zoom in, zoom out, pan left, pan right, tilt up, tilt down, dolly forward, dolly back, push in, pull out, tracking, orbit, handheld, static, crash zoom). Drop the name into the prompt and Hailuo executes it precisely. This makes Hailuo unusually strong for storyboard-driven work, ad shoots and any context where the director's intent is non-negotiable.

Hailuo is weaker than Sora 2 Pro for photorealistic faces and weaker than Seedance for multi-reference composition. Pick Hailuo when prompt precision and director-level camera control matter more than ultimate visual fidelity.