MiniMax is an AI company, founded in early 2022, that develops its own multimodal foundation models: models that understand and generate several kinds of data, here text, audio, images, video and music. Its model lines include MiniMax M3 for language, MiniMax H3 for video and MiniMax Speech 2.8. Its products include MiniMax Code, MiniMax Design, MiniMax Audio, Talkie, and an open platform with an API for developers.
The video model MiniMax H3 follows the earlier Hailuo 01 and Hailuo 02 models. It takes a text prompt plus optional images, video clips and audio, and generates a video of 4 to 15 seconds at 768p or 2K resolution, with native stereo sound. Its modes are text-to-video, image-to-video from a first and a last frame, and reference generation, which copies a character, motion, camera movement or voice from the inputs. A program creates a generation task through the API, polls its status with the task ID, and downloads the finished video.
API use is billed per second of output video for video models and per million tokens for language models; audio and video packages are also sold.