MiniMax H3 on Visual Maker AI: Build Videos from Text and References
Use MiniMax H3 for short, high-resolution video generation from prompts, images, first and last frames, and multimodal references.
MiniMax H3 is a flexible video model for projects that need a prompt plus visual direction. In Visual Maker AI, it supports text-to-video, image-to-video, first/last-frame animation, and reference-to-video workflows.
Its available output settings are 768p and 2K, with durations from 4 to 15 seconds. It also supports wide, square, landscape, and vertical aspect ratios, which makes it practical for both campaign assets and social placements.
Four ways to direct a clip
Text to video is the fastest way to explore a concept. Include a clear subject, a single action, camera behavior, and an intentional setting. It works well for a rough commercial idea, a cinematic establishing shot, or a motion graphic direction.
Image to video starts from one uploaded image. Use it for an animated poster, a product reveal, or an old photograph. The image establishes what must stay recognizable; the prompt should focus on the motion you want added.
First and last frame gives a sequence a stronger destination. Use two supplied frames for a controlled product transformation, title transition, or scene reveal. Describe the path between them rather than repeating every visual detail already visible in the images.
Reference to video is for a richer brief. Combine image, video, and audio references when identity, movement, and sound each come from different source material. Every asset should have a role. A reference set works best when it is small and deliberate, not when it contains every available file.
Bring old photos to life without over-animating them
For archival portraits, subtlety is the goal. Upload the photo in Image to video mode and ask for one or two restrained motions: a blink, a small head movement, slight fabric motion, or gentle background activity. Explicitly ask the model to preserve face shape, clothing, period details, and the original composition.
Avoid asking for a strong camera move and multiple character actions at the same time. The simpler direction better protects the identity and feeling of the source image.
Prompt example
A vintage family portrait in a sunlit garden. Preserve every face, outfit, and the original framing. Add a soft breeze moving the leaves, a subtle blink from the seated woman, and gentle film grain. Locked-off camera, quiet and respectful mood.
This prompt defines what should remain unchanged before it asks for motion.
Start with MiniMax H3
Use the image-to-video guide and generator for an image-led animation workflow, or choose MiniMax H3 in the Video generator for a different workflow.