> For clean Markdown of any page, append .md to the page URL. > For a complete documentation index, see https://docs.empiriolabs.ai/llms.txt. > For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.empiriolabs.ai/_mcp/server. # MiniMax H3 > Generates 4 to 15 second clips at up to 2K with native stereo audio, following text, image, video, and audio references in one request. ![MiniMax H3](https://media.empiriolabs.ai/model-logos/minimax.png) [MiniMax](/providers/minimax) · Video Generation `POST /v1/videos/generations` Generates 4 to 15 second clips at up to 2K with native stereo audio, following text, image, video, and audio references in one request. ## At a glance | Field | Value | | ------------------- | ----------------------------------------------------------------- | | Model id | `minimax-h3` | | Model release date | 2026-07-31 | | Input modalities | Text, Image, Video, Audio | | Output modalities | Video | | Context window | - | | Weight precision | - | | Region | Singapore | | Features | video_generation, image_to_video, audio_out, audio_in, multimodal | | Native inference | No | | New | No | | Supported endpoints | `POST /v1/videos/generations` | | Alternate model ids | `minimax/minimax-h3`, `hailuo-3`, `minimax/hailuo-3` | ## Pricing | Charge | Spec | Rate | | ----------------------- | --------------------------------- | ------ | | 768p | per second | \$0.18 | | 2K | per second | \$0.26 | | 2K regeneration | per second | \$0.10 | | Reference image input | per input image after the first 5 | \$0.08 | | Prompt expansion input | per 1M prompt tokens | \$1.80 | | Prompt expansion output | per 1M generated tokens | \$7.20 | ## Example request ```bash curl https://api.empiriolabs.ai/v1/videos/generations \ -H 'Authorization: Bearer $EMPIRIOLABS_API_KEY' \ -H 'Content-Type: application/json' \ -d '{"model": "minimax-h3", "prompt": "sunrise over the ocean", "duration": 6}' ``` ## Parameters | Parameter | Type | Required | Default | Description | | ---------------- | ------- | -------- | ------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `prompt` | string | yes | - | Scene description, up to 7000 characters. Can describe action, camera work, dialogue, sound effects, and on-screen text. | | `mode` | enum | no | `"auto"` | auto: detect from the inputs. t2v: text-to-video. i2v: animate one image as the opening frame. flf: transition from a first frame to a last frame. r2v: reference-to-video from images, videos, and audio. regenerate: re-render a 768p clip from this model at 2K. · Allowed: `auto`, `t2v`, `i2v`, `flf`, `r2v`, `regenerate` | | `resolution` | enum | no | `"768p"` | Output resolution. 2K is billed at a higher per-second rate than 768p. · Allowed: `768p`, `2k` | | `aspect_ratio` | enum | no | `"adaptive"` | Output aspect ratio. Adaptive follows the supplied input media. · Allowed: `adaptive`, `21:9`, `16:9`, `4:3`, `1:1`, `3:4`, `9:16` | | `duration` | number | no | `6` | Clip length in seconds, whole numbers only. · Range: 4 – 15 | | `image` | string | no | - | Image URL. One opening frame for image-to-video, or up to 9 reference images for reference-to-video. | | `last_frame` | string | no | - | Closing frame image URL. Supplying it selects first and last frame mode. | | `video` | string | no | - | Reference video URL, up to 3 clips of 2 to 15 seconds each. Reference video is billed for its own duration in addition to the generated clip. | | `audio` | string | no | - | Reference audio URL, up to 3 clips of 2 to 15 seconds each. Requires an accompanying reference image or video. | | `base_video` | string | no | - | A 768p clip previously generated by this model, to re-render at 2K. Supplying it selects regeneration, which is billed at its own lower per-second rate on the source clip and the output. | | `watermark` | boolean | no | false | Add the AI-generated content watermark to the output. | | `enhance_prompt` | boolean | no | false | Expand your brief into a full prompt before rendering, adding shot structure, camera work and a soundscape. Any reference media you attach informs the expansion. Adds a per-token charge on top of the per-second video cost. Not used by regenerate. | ## Notes MiniMax H3 generates 4 to 15 second clips at 24 fps in 768p or 2K, each with a native stereo audio track. **Modes** * Text to video: prompt only. * Image to video: one image becomes the opening frame. * First and last frame: supply an opening and a closing image. * Reference to video: up to 9 images, 3 videos and 3 audio clips, 12 reference files in total. * Regeneration: re-render a 768p clip you generated earlier at 2K. Send the original prompt plus the source clip as base_video. **Prompt expansion** Set `enhance_prompt` to true and the model first rewrites your brief into a full prompt, adding shot-by-shot structure, camera moves, lighting and a soundscape, then renders from that. Any reference media you attach informs the expansion. It adds a per-token charge on top of the per-second video cost, and is skipped for regeneration because that path reproduces the original request. **Reference media** * Images 256 to 5760 px per side, aspect ratio between 0.4 and 2.5, up to 30 MB each. * Videos 2 to 15 seconds, H.264 or H.265, up to 50 MB each. * Audio 2 to 15 seconds, WAV or MP3, up to 15 MB each. Reference audio requires an accompanying image or video reference. * Reference video and audio apply to reference-to-video only. * Prompts can be up to 7000 characters and may describe dialogue, sound effects, on-screen text, and camera work. **Billing** * Billed per second of the generated clip at the resolution you request. * A reference video is billed for its own duration in addition to the generated clip, so a 5 second reference used to make a 6 second clip bills 11 seconds. * Regeneration is billed at its own lower per-second rate, on the source clip duration plus the regenerated output. Reference images carried over from the original request are not charged again. * The first 5 reference images are included. Each additional reference image is charged separately. * Prompt expansion, when enabled, is billed per token. --- *Machine-readable schema:* `GET https://api.empiriolabs.ai/v1/models/minimax-h3`.