# MiniMax H3 - Core.Today AI API

> MiniMax H3 multimodal video generation via Replicate — one endpoint for text-to-video, first/last-frame image-to-video, and reference-based generation with images, videos, and audio cited in the prompt. 4-15 second duration at 768P or 2K.

- **Provider**: MiniMax
- **Model ID**: minimax/h3
- **Category**: Video Generation
- **Credits**: 1510 per 5s 2K video — billed per second (768P 186/s, 2K 302/s)
- **Speed**: Medium
- **Quality**: Ultra

## Features
- One endpoint for text-to-video, image-to-video, and reference-based generation
- First and/or last frame control for precise start and end framing
- Up to 9 reference images, 3 reference videos, and 3 reference audio clips
- 4-15 second duration at 768P or 2K
- Seven aspect ratios including adaptive

## Use Cases
- First-to-last keyframe transitions between two designed shots
- Character-consistent scenes driven by up to 9 reference photos
- Motion or camera-style transfer from short reference clips
- Voice- or rhythm-guided takes from reference audio
- Cinematic text-to-video clips up to 15 seconds

## API Endpoint

Base URL: https://api.core.today/v1
Create Prediction: POST /predictions
Get Status: GET /predictions/{job_id}

## Authentication
Header: X-API-Key: YOUR_API_KEY

## Input Parameters

### Required
- **prompt**: string - Text prompt for generation

### Optional
- **first_frame_image**: string - First frame image. The output video will have the same aspect ratio as this image
- **last_frame_image**: string - Last frame image. The final frame of the output video will match this image
- **reference_image_urls**: array - Public URLs for up to 9 reference images
- **reference_video_urls**: array - Public URLs for up to 3 reference videos
- **reference_audio_urls**: array - Public URLs for up to 3 reference audio clips
- **duration**: integer (default: 5) - Output duration in seconds (4-15). Billed per second
  Options: 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15
- **resolution**: string (default: 2K) - Output resolution. 768P bills 186 credits/sec, 2K bills 302 credits/sec
  Options: 768P, 2K
- **ratio**: string (default: 16:9) - Output aspect ratio. Required for text-to-video; adaptive for image-to-video
  Options: adaptive, 21:9, 16:9, 4:3, 1:1, 3:4, 9:16

## Examples

### Image-to-Video from a First Frame
Animate a still image with a prompt-guided motion
```json
{
  "model": "minimax/h3",
  "input": {
    "prompt": "the car revs through the desert",
    "first_frame_image": "https://example.com/car.png",
    "duration": 5,
    "resolution": "2K",
    "ratio": "16:9"
  }
}
```

### Reference-Based Generation
Keep a subject consistent using reference images
```json
{
  "model": "minimax/h3",
  "input": {
    "prompt": "The woman from the reference images walks through a sunlit market, cinematic tracking shot",
    "reference_image_urls": [
      "https://example.com/subject-1.jpg"
    ],
    "duration": 8,
    "resolution": "768P",
    "ratio": "adaptive"
  }
}
```

## Response Format
```json
{
  "job_id": "abc123",
  "status": "pending | processing | completed | failed",
  "result": "URL or data (when completed)"
}
```

## Usage Flow
1. POST /predictions with model and input -> receive job_id
2. GET /predictions/{job_id} -> poll until status is completed or failed
3. Result contains output URL(s)

## Tips
- Billing is per second of output — iterate at 768P (about 60% of the 2K rate) and short durations, then render the final at 2K
- Provide first_frame_image and last_frame_image together for controlled start-to-end transitions; the output follows the first frame's aspect ratio
- Set an explicit ratio for text-to-video; with image inputs use adaptive and let the frame follow the image
- Reference videos and audio are capped at 3 each — trim clips to the motion or rhythm segment that matters

## Documentation
https://replicate.com/minimax/h3