MiniMax H3 Max Reference to Video API

MiniMax H3 Max Reference to Video

MiniMax H3 Max Reference to Video is a multimodal video model built for complex narratives and continuity across shots. Supply multiple reference images, videos, and audio clips to guide facial features, clothing materials, camera placement, and sound cues. Adaptive aspect ratios and output from 480p to 1080p support high-throughput production for digital-character dramas, animation prototypes, and advertising series.

Why choose the MiniMax H3 Max Reference to Video API?

• Combined guidance across media typesUse still images, video clips, and ambient audio together to control subject design, movement rhythm, and acoustic texture within a single generation.

• Consistent character features across shotsMultiple reference images anchor facial contours, hairstyles, hair color, clothing, and accessories, reducing character distortion across scenes in longer narratives.

• Transfer motion rhythm and camera movementExtract camera paths, push-in and pullback timing, and subject movement range from reference videos, then apply them to newly prompted scenes.

• Adaptive aspect ratios and super-resolution outputAutomatically infer an appropriate aspect ratio from reference images with adaptive, then deliver 480p, 768p, or 1080p output as needed.

• Transparent billing with a shared free allowanceOutput video is billed by the requested duration parameter in seconds. Reference media receives a free allowance of 4,096 tokens, with usage beyond that billed separately for flexible cost control.

Parameters

How to Use

1. 01

   Prepare multimodal referencesPrepare images with clear subjects, videos with steady camera work, or clean audio files. Ensure each has a publicly accessible HTTP(S) URL.

2. 02

   Assign a role to each referenceExplain each reference's purpose in the prompt, such as which image supplies the character's appearance, which video guides camera rhythm, and which audio clip provides ambient sound.

3. 03

   Write the scene progression promptSpecify the subject's actions, emotional expression, and lighting in the new setting. Avoid instructions that conflict with existing reference features.

4. 04

   Choose aspect ratio and resolutionKeep adaptive to follow the first image's ratio, or choose a fixed ratio such as 16:9 or 9:16. Set a 5–15 second duration and 480p/768p/1080p resolution as needed.

5. 05

   Submit the task and poll for resultsSend a request to receive a task_id, then poll the status endpoint for the generated video's public playback URL. For longer tasks, a callback_url is recommended.

Usage and billing

The MiniMax H3 Max Reference to Video API has two cost components: the base output fee, billed by the requested duration parameter in seconds, and token usage for reference media.

Best Use Cases

• Serialized dramas and recurring charactersUse multiple portrait references to maintain consistent facial features and body proportions across scenes and wardrobe changes.

• Brand visuals and consistent style across productsCombine brand color palettes, product photography, and specified camera movements to keep commercials and promotional materials visually consistent.

• Motion capture and shot reuseReference camera work and body movement from existing film clips to recreate high-quality fight or dance shots at a lower cost.

• Integrated audiovisual concept previewsCombine concept art, storyboard clips, and audio samples to present an immersive audiovisual preview early in a project's development.

Pro Tips

• Point to specific references in the prompt: identify which image should guide the face and clothing and which video should guide camera rhythm to direct the model's attention.
• Focus each reference on one core feature: use an image to emphasize a face, clothing style, or architectural background. Cluttered references can confuse the model.
• At least one reference image is required. Video and audio are optional additions and cannot be submitted without a reference image.
• Match reference dimensions for adaptive framing: when using adaptive, keep all reference images at the same aspect ratio for more stable, predictable composition.
• Check reference video specifications: use constant 24fps and square pixels. Rotation metadata or nonstandard encoding can cause validation failures.

Notes

• The combined number of image, video, and audio references must not exceed 12. Each media type must also stay within its individual limit.
• Supports 1–9 images (JPEG/PNG/WebP, ≤ 30MB each), up to 3 videos (MP4/MOV, ≤ 50MB each), and up to 3 audio clips (MP3/WAV, ≤ 15MB each).
• A request must include at least 1 reference image. Video and audio are optional supporting references and cannot be used alone.
• A successful submission returns a task_id and runs asynchronously. Poll the status endpoint or configure callback_url to receive the final result asynchronously.