Standard preview
$0.52
audio incl.
AI video · MiniMax
Turn more ideas into moving scenes.
MiniMax H3 Max creates 5–15-second videos from text, frames or image, video and audio references in 480P, 768P or 1080P. Use it on the website or through MCP. Generate native audio or supply a soundtrack for text and image workflows; choose MiniMax H3 for 2K or 4K creation. Recast and Relight provide separate source operations; only Relight adds 2K. New Extend, Insert and Reframe remain pending.

Reviewed launch examples
Contemporary dance on a mirrored stage
Choose your output resolution
Render at 480P, 768P or 1080P and choose balanced or quality prompt expansion to develop your scene.
Direct with frames or references
Use a start image, an end image or both. In reference mode, guide identity, movement and sound with images, videos and audio clips.
5–15-second clips
Choose any whole-second duration and a landscape, square or vertical format for text-to-video.
Native audio or your soundtrack
Generate dialogue, ambience, music and effects, or supply a soundtrack in text and image modes. Audio-only references are also supported in reference mode.
Recast / Relight
Video references guide a newly generated 5–15-second clip; they do not directly edit those reference clips. Recast and Relight are selected separately and inherit source timing.
Extend / Insert / Reframe
H3 Max Extend, Insert and Reframe remain pending and cannot be activated in this release. Recast and Relight are specialized source tools, separate from mixed-reference creation.
The exact price is calculated before rendering from duration, mode, and resolution.
$0.52
audio incl.
$0.84
Most popularaudio incl.
Generation example: 5 s · 768P · Text to video · Audio included — $0.52
The total shown at generation time remains authoritative.
Two distinct launch renders show the model on people, scenes, or products.
Check the requested subject, action, framing, and final beat.
Review physical contact, camera movement, and temporal stability.
Choose H3 Max for text, frames and mixed references up to 1080P. Compare MiniMax H3 when your delivery needs 2K or 4K.
MiniMax H3 adds 2K and 4K output while keeping text, frame and multimodal reference workflows. Compare both models with the same brief and current quote.
Lead with the visible action, then add camera, continuity, and sound.
Describe the subject, action, camera and sound. Add an optional soundtrack to guide the audio.
Use a start image, an end image or both, with an optional soundtrack. Keep frame inputs separate from mixed references.
Use up to 9 images, 3 videos and 3 audio clips within 12 unique references. Audio-only references work too; video and audio each have a 15-second combined limit.
Use a soundtrack in text or image mode, or audio references in reference mode. Describe dialogue, ambience and synchronized sound cues in the prompt.
Recast replaces people using 1–4 ordered portraits, mapped left to right within each shot. Use a 5–30-second source, with each shot at most 15 seconds, and 768P or 1080P output. Source motion, timing, cuts and audio guide the replacement; review the result rather than assuming identity preservation.
Relight uses a lighting-sphere image without a prompt. Prepare a constant 24 fps source with 17n+5 decoded frames, at least 56 frames and at most 15 seconds. Review any ending removed by preparation. Relight retains prepared timing and audio and offers 480P, 768P, 1080P or 2K.
H3 Max Extend, Insert and Reframe remain pending and cannot be activated in this release. Recast and Relight are specialized source tools, separate from mixed-reference creation.
Define the subject, visible change, camera path, and sound response.
A sculptural ceramic lamp in a dark design studio unfolds from a compact form into a warm glowing arch. One slow orbit, stable materials, precise joints, soft room tone, and a clean final hero frame. No text or logo.
Use balanced expansion for direct prompts and quality expansion when the brief benefits from added cinematic detail.
A sculptural ceramic lamp in a dark design studio unfolds from a compact form into a warm glowing arch. One slow orbit, stable materials, precise joints, soft room tone, and a clean final hero frame. No text or logo.
12-second ORIGINAL PHOTOREALISTIC LIVE-ACTION disaster micro-film, 16:9. EXACTLY THREE SHOTS, with only two clean hard cuts at approximately 5 seconds and 9 seconds. Do not blend shots, do not dissolve, do not morph locations, do not create a montage. Strict continuity of the same two people, clothes, faces, weather and direction of movement. FICTIONAL SETTING: an alternate-history European city in 1978, filmed on real streets with real actors, real vintage cars, wet asphalt, concrete architecture, practical smoke, dust and sparks. Natural skin texture, realistic hair and cloth, physically plausible gravity and movement, subtle 35mm anamorphic grain, grounded handheld photography. Original large-scale disaster-film tension. NO CGI look, NO 3D, NO game-engine lighting, NO plastic skin, NO holograms, NO impossible flying debris, NO recognizable celebrity, brand or real city. CHARACTER CONTINUITY: a real-looking mother in her early 30s, short dark curly hair, burnt-orange leather coat, cream shirt and dark flared trousers; her same 9-year-old child in a yellow raincoat and dark wool cap. The child stays beside the mother, is never hurt and helps find safety. SHOT 1 — 0:00–0:05. One uninterrupted street-level tracking shot moving backward at their running speed. The mother and child run hand-in-hand toward camera on a rain-slick avenue. Far behind them, an elevated tram station erupts in one believable practical orange flash; metal sparks and a heavy rolling dust cloud spread through the intersection while period cars brake. Nothing flies toward the actors and no one is struck. The mother sees a recessed red-steel emergency shelter under a concrete arcade, changes direction once and pulls the child inside. Cut on their movement as they cross the shelter threshold. SHOT 2 — 0:05–0:09. Static intimate medium close-up from inside the same shelter, matching their position and clothing exactly. Warm tungsten emergency light, realistic tired breathing, disaster noise muffled outside. The child looks at the mother and quietly says in natural synchronized English: “Mom, I’m scared.” The mother kneels to eye level, holds the child gently by the shoulders and replies softly with natural synchronized lips: “I’m here. Breathe with me.” She takes one slow breath; the child copies her and begins to calm. At the end, the child notices a large mechanical safety lever beside them, marked only by a small green lamp, and points to it. No subtitles and no other speech. SHOT 3 — 0:09–0:12. Stable medium-wide shot from the same shelter interior. The child leads the mother to the lever and they pull it together once. A heavy red steel shutter closes smoothly across the entrance just before the exterior dust shockwave passes. The camera and characters remain physically stable. End on one clear warm frame: the mother embraces the child; the child gives a small relieved smile; both are safe. The child’s observation completes the escape. AUDIO: precise location sound, distant sirens, rain, tram metal and low blast; muffled ambience and clear close dialogue in shot 2; heavy mechanical lever and shutter in shot 3; only a restrained low analogue-synth note after they are safe. No loud music over dialogue. No gore, injury, death, panic close-up, text, captions or logos.
Best practices, common fixes, and important limitations to help you get the strongest results with H3 Max.
Compare prices, formats, and capabilities before choosing your model.
Create 5–15 second character-led videos from text, images, video, and audio references with MiniMax H3 in MaxVideoAI.
Compare H3 Max vs MiniMax H3Seedance 2.5 video generation workflows on MaxVideoAI.
Compare H3 Max vs Seedance 2.5Generate fast, stylized motion from prompts or animate uploaded stills with MiniMax Hailuo 02. Ideal for testing motion concepts, loops, and storyboard passes—silent and budget-friendly via MaxVideoAI.
Formats, settings and limits to check before creating.
Built-in safeguards and best practices for responsible creation with H3 Max.
Inspect anatomy, identity, clothing, and natural movement.
Inspect detail, lighting, materials, and resolution.
Reject unwanted text, marks, unsafe content, or unapproved likenesses.
