# Wan 3.0 Prime Reference to Video - Core.Today AI API

> Alibaba Wan 3.0 Prime reference-to-video via Fal.AI — the higher-fidelity Wan 3.0 tier. Combines up to 10 reference images, 5 reference video clips, and 5 reference audio clips — each cited in the prompt by position (Image 1, Video 1, Audio 1) — into one coherent video that keeps character identity, style, motion, and sound cues. 2-30 second output at 480p/720p/1080p with optional generated audio.

- **Provider**: Alibaba
- **Model ID**: alibaba/wan-3.0-prime/reference-to-video
- **Category**: Video Generation
- **Credits**: 3260 per 5s 1080p video — billed per second of output plus reference-video seconds (480p 158/s, 720p 326/s, 1080p 651/s)
- **Speed**: Slow
- **Quality**: Ultra

## Features
- Prime tier: higher fidelity and stronger multimodal coherence than Wan 3.0
- Positional citation in the prompt — Image 1, Video 1, Audio 1
- Keeps character identity, visual style, movement, and sound cues consistent
- 2-30 second output at 480p, 720p, or 1080p with adaptive or fixed aspect ratio
- Optional generated audio, prompt expansion, and enhanced reasoning (thinking) with document/web sources

## Use Cases
- Character-consistent scenes driven by reference photos
- Motion or camera transfer from a short reference clip to a new subject
- Voice- and music-following takes from reference audio
- Brand-consistent ad variations reusing the same product and talent references
- Explainer videos grounded in a document or public webpage (with thinking)

## 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 directing how the reference media is used. Address references positionally, e.g. 'the subject in Image 1 walks past Video 1'. Up to 20000 characters

### Optional
- **reference_image_urls**: array - Reference image URLs (character, object, or style), cited as Image 1, Image 2, ... Up to 10 images
- **reference_video_urls**: array - Reference video URLs (motion, camera, scene), cited as Video 1, ... Up to 5 clips totaling at most 15 seconds, each at least 16 fps. Reference video seconds are billed in addition to the output duration
- **reference_audio_urls**: array - Reference audio URLs (voice, music, sound cues), cited as Audio 1, ... Up to 5 clips totaling at most 15 seconds
- **resolution**: string (default: 1080p) - Output video resolution tier
  Options: 480p, 720p, 1080p
- **aspect_ratio**: string (default: adaptive) - Output aspect ratio, or 'adaptive' to let the model choose
  Options: adaptive, 16:9, 4:3, 1:1, 3:4, 9:16
- **duration**: integer (default: 5) - Output duration in seconds (2-30). Billed per second of output plus any reference video seconds
- **audio**: boolean (default: true) - Include generated audio. Costs the same either way
- **enable_prompt_expansion**: boolean (default: true) - Enable intelligent prompt rewriting. Disabling it saves roughly 20-60 seconds of latency but is likely to degrade quality
- **enable_thinking**: boolean (default: false) - Enable enhanced reasoning before generation. Turned on automatically when file_url or web_url is provided
- **file_url**: string - Document URL to base the video on (enables thinking automatically)
- **web_url**: string - Public webpage URL to base the video on; pages behind a login cannot be read (enables thinking automatically)
- **enable_safety_checker**: boolean (default: true) - Enable content moderation for input and output
- **seed**: integer - Random seed for reproducible generation (0-2147483647). Random when omitted

## Examples

### Character-Consistent Scene
Keep a character and a prop consistent with their reference images
```json
{
  "model": "alibaba/wan-3.0-prime/reference-to-video",
  "input": {
    "prompt": "The woman in Image 1 walks along a rainy neon-lit street at night, holding the umbrella from Image 2, cinematic tracking shot.",
    "reference_image_urls": [
      "https://example.com/woman.jpg",
      "https://example.com/umbrella.jpg"
    ],
    "duration": 5,
    "resolution": "1080p",
    "aspect_ratio": "adaptive"
  }
}
```

### Motion and Voice Transfer
Follow the motion of a reference clip and the voice of a reference audio
```json
{
  "model": "alibaba/wan-3.0-prime/reference-to-video",
  "input": {
    "prompt": "The character in Image 1 performs the dance from Video 1 while singing with the voice in Audio 1.",
    "reference_image_urls": [
      "https://example.com/character.png"
    ],
    "reference_video_urls": [
      "https://example.com/dance.mp4"
    ],
    "reference_audio_urls": [
      "https://example.com/voice.mp3"
    ],
    "duration": 8,
    "resolution": "720p"
  }
}
```

## 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
- Cite every reference explicitly in the prompt (Image 1, Video 1, Audio 1) — uncited references contribute weakly
- Reference video seconds are billed on top of the output duration — trim clips to the part that matters
- Iterate at 480p, then render the final at 1080p
- Disable enable_prompt_expansion to cut 20-60 seconds of latency when your prompt is already detailed
- Use Wan 3.0 (non-Prime) reference-to-video for drafts at about 70% of the Prime price

## Documentation
https://fal.ai/models/alibaba/wan-3.0-prime/reference-to-video