MiniMax-Hailuo-2.3

MiniMax Hailuo 2.3 is a video-generation model with text-to-video and image-to-video modes. It builds on Hailuo 02 with a focus on physical action, visual style, character micro-expressions and response to motion instructions. Supply a written scene to create a shot from scratch, or use a starting image to establish the opening view and describe the movement that follows.

Physical performance is one of the edition’s main areas of improvement. Hailuo 2.3 aims to make body movements more fluid and controlled, including scenes with a moving camera. Plan the important action before adding decorative detail. A character lifting an object, turning towards another person or moving through a room gives the shot a visible event to evaluate. Decide which movement carries the story, then keep the surrounding instructions consistent with it.

Stylization is another focus. The model supports work in anime and illustration as well as styles such as ink-wash painting and game CG. Describe the visual treatment you want, and keep the appearance coherent with the scene. If a starting image already establishes a drawing style, use the text to guide the animation rather than introducing a competing description of the character’s design. Review the whole clip for whether that treatment remains readable during movement.

The model also focuses on facial performance and subtle expression changes. Prepare a character shot around the reaction that matters: noticing an object, becoming curious or responding with a small smile. A close view can make that change easier to assess, while a large sequence of gestures may distract from it. Check the face before, during and after the reaction rather than judging only a representative still frame.

Object motion is useful for product-centred scenes. An original brief might show a small ceramic vase turning in warm side light while the camera gradually approaches it. Identify which properties must remain recognisable, such as the silhouette, rim and surface finish. Review those details through the movement, because a convincing atmosphere alone does not establish a faithful product view.

For text-to-video, the required description can contain up to 2,000 characters. State the subject, action, setting and visual style in a coherent order. Image-to-video requires a starting-frame image, with an optional description of up to 2,000 characters. The image fixes the opening composition, so prepare the subject and framing before requesting motion. Leave visible space where an arm, object or character needs to move.

The starting image accepts JPG, JPEG, PNG or WebP, must be smaller than 20 MB, needs a short edge greater than 300 pixels and must have an aspect ratio between 2:5 and 5:2. It can be supplied as a public image URL or a Base64 data URL. Choose a clear opening picture rather than relying on the animation to repair an obscured face or an already cropped action.

Hailuo 2.3 generates at 24 frames per second. A six-second shot can use 768P or 1080P; a ten-second shot uses 768P. The defaults are six seconds and 768P. Choose the length for the action you want to see: a brief reaction may fit a short shot, while a slower movement needs more room. Read the completed result’s actual width and height when placing the clip in an edit, since a resolution label alone is not the full framing decision.

Camera direction can use bracketed commands. The available controls cover trucking, panning, pushing, pedestal movement, tilting, zooming, shaking, tracking and a static shot. A command such as [Push in] directs the approach; commands grouped inside one bracket act together, while commands placed in order describe a sequence. MiniMax recommends at most three simultaneous commands. Keep the camera movement compatible with the subject’s action so the main event remains easy to follow.

Automatic prompt optimisation is enabled by default and can be disabled for closer control over the supplied wording. Optional fast preprocessing reduces the optimisation time when that optimisation is enabled; it is a separate setting from choosing the Hailuo 2.3 Fast model. Standard Hailuo 2.3 supports both creation modes, while Fast is the distinct image-to-video edition. Choose the edition based on whether the project begins with text alone or an existing frame.

The generation workflow returns an asynchronous task identifier, then reports preparation, queueing, processing, success or failure. On success, the result supplies a file identifier and the actual video dimensions, and the completed video is retrieved for download. Save the clip while its download address is valid. The maker’s current video-download reference gives the returned download URL a one-hour lifetime. For an edited sequence, retain each shot with its prompt and starting image so that the chosen creative direction remains traceable during revision.

Review action, physical continuity, facial or product details, visual style and camera behaviour in the saved clip. Watch the whole shot, then inspect its beginning and end where it will join other material. If the request contains too many competing movements, simplify the main action; if the opening design is unclear, improve the starting image. Select a shot that communicates the intended event, then add narration, titles and transitions in the editing workflow.