minimax/h3

Generate videos with MiniMax H3, a multimodal video model from MiniMax that supports text-to-video, first- and last-frame image-to-video, and reference-based video generation.

13 runs
Playground Examples README

MiniMax H3

MiniMax H3 is a multimodal video generation model from MiniMax. It can generate videos from text prompts, animate a first or last frame, and use reference images, videos, or audio to guide the result.

What it does

  • Text-to-video: describe a scene and generate a video from scratch.
  • Image-to-video: provide a first frame, a last frame, or both to control the start and end of the video.
  • Reference generation: provide reference images, videos, or audio to guide character, motion, camera, style, voice, or editing rhythm.

The model supports 4–15 second videos at 768P or 2K resolution.

Inputs

  • prompt: the text prompt describing the video.
  • first_frame_image: optional first frame image. When provided, the output uses the input image aspect ratio.
  • last_frame_image: optional last frame image.
  • reference_image_urls: optional public URLs for reference images.
  • reference_video_urls: optional public URLs for reference videos.
  • reference_audio_urls: optional public URLs for reference audio clips.
  • duration: output duration in seconds, from 4 to 15.
  • resolution: 768P or 2K.
  • ratio: output aspect ratio for text-to-video. Use a concrete ratio like 16:9; image-to-video uses adaptive.

Billing

MiniMax H3 is billed per second of generated video. 768P outputs cost $0.08 per second, and 2K outputs cost $0.13 per second.

Model created