# FLUX 3 - Core.Today AI API

> Black Forest Labs' multimodal FLUX 3 generates 5-20 second video with synchronized audio from a text prompt, from one or more images (first frame, first+last frame, or a storyboard of up to 10), or by continuing an existing clip — at 720p or 1080p, with a low-cost draft mode.

- **Provider**: Black Forest Labs
- **Model ID**: black-forest-labs/flux-3
- **Category**: Video Generation
- **Credits**: 1980 per second — 720p 395/s, 1080p 674/s, draft (720p) 140/s; video continuation (start_video) 720p 953/s, 1080p 1232/s, draft 279/s; 5-20s (default 5s at 720p = 1980; 'auto' reserves 20s and settles on the actual length)
- **Speed**: Slow
- **Quality**: High

## Features
- Synchronized audio generated with the video (can be turned off)
- Text-to-video, image-to-video, first+last frame and storyboards of up to 10 images
- Video continuation from an existing clip via start_video
- 720p or 1080p output, 5-20 seconds or 'auto' length
- Draft mode for fast, low-cost 720p previews

## Use Cases
- Storyboard-driven shots where keyframes are fixed in advance
- Extending a shot or chaining clips into a longer sequence
- Cheap draft passes before a full-quality 1080p render
- Animating a still image with matching sound

## 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 description of the video. Describe the scene, action, camera moves, and any audio you want — the prompt is interpreted and expanded before generation

### Optional
- **images**: array (default: []) - Optional images placed pixel for pixel: 1 = first frame, 2 = first and last frame, 3-10 = storyboard (needs an explicit duration). PNG, JPEG or WebP. Can't be combined with start_video
- **start_video**: string - A video to continue from its final frames (mp4, ≤50MB, ≤15s). Billed at the video-continuation rate. Can't be combined with images
- **aspect_ratio**: string (default: auto) - Aspect ratio of the generated video. 'auto' picks a ratio from your prompt and inputs
  Options: auto, 21:9, 2:1, 16:9, 4:3, 1:1, 3:4, 9:16
- **resolution**: string (default: 720p) - Output resolution. Ignored in draft mode (drafts are 720p)
  Options: 720p, 1080p
- **duration**: string (default: 5) - Clip length in seconds ("5"-"20"). Defaults to "5" when omitted. 'auto' lets the model choose (5-20s) and is billed on the actual output length (20s is reserved up front)
  Options: auto, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20
- **generate_audio**: boolean (default: true) - Generate synchronized audio (ambient sound, speech, effects). Set to false for a silent clip
- **draft**: boolean (default: false) - Generate a fast, low-cost 720p draft preview instead of a full-quality clip
- **safety_tolerance**: integer (default: 2) - Moderation tolerance, 0 (strictest) to 4 (most permissive). Requests with image or video inputs are limited to 2

## Examples

### Text-to-Video (720p)
A 5-second 720p clip with audio — 5s at 395/s = 1,980 credits
```json
{
  "model": "black-forest-labs/flux-3",
  "input": {
    "prompt": "a home video of a futuristic san francisco with people riding hoverboards",
    "images": [],
    "duration": "5",
    "resolution": "720p",
    "aspect_ratio": "auto",
    "generate_audio": true,
    "draft": false,
    "safety_tolerance": 2
  }
}
```

### Draft Preview
Fast 720p draft to check the idea — 8s at 140/s
```json
{
  "model": "black-forest-labs/flux-3",
  "input": {
    "prompt": "A red panda walks along a mossy log in a sunlit forest, one continuous shot, birdsong",
    "duration": "8",
    "draft": true
  }
}
```

## 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
- Iterate with draft: true (140/s) and only render the final at full quality
- 'auto' duration reserves the 20-second maximum up front and refunds the difference once the actual length is known — set an explicit duration if your balance is tight
- start_video (video continuation) costs more than twice the text/image rate — use it for extending shots, not as a general input
- Three or more images form a storyboard and need an explicit duration
- Requests with image or video inputs are limited to safety_tolerance 2

## Documentation
https://replicate.com/black-forest-labs/flux-3