Video Models

Available video generation models in the Decart provider

Model IDs

type DecartVideoModelId =
  | "lucy-pro-t2v"
  | "lucy-pro-i2v"
  | "lucy-dev-i2v"
  | "lucy-motion";

The Decart provider supports four video generation models with different capabilities.

Available Models

lucy-pro-t2v

Text-to-video generation model powered by Lucy Pro.

Capabilities:

• Text-to-video generation
• Seed-based reproducibility
• Resolution control (720p, 480p)
• Aspect ratio control (16:9, 9:16)

Use Cases:

• Generating videos from text descriptions
• Creating video content without source images

import { createDecart } from '@decartai/ai-sdk-provider';
import { experimental_generateVideo } from 'ai';

const decart = createDecart();

const { video } = await experimental_generateVideo({
  model: decart.video('lucy-pro-t2v'),
  prompt: 'Waves crashing on a beach at sunset'
});

lucy-pro-i2v

Image-to-video generation model powered by Lucy Pro.

Capabilities:

• Image-to-video generation
• Seed-based reproducibility
• Resolution control (720p, 480p)
• Aspect ratio control (16:9, 9:16)
• Optional text prompt for guidance

Use Cases:

• Animating static images
• Creating videos from existing artwork
• Adding motion to photographs

import { createDecart } from '@decartai/ai-sdk-provider';
import { experimental_generateVideo } from 'ai';
import { readFileSync } from 'fs';

const decart = createDecart();

const imageData = readFileSync('input.jpg');

const { video } = await experimental_generateVideo({
  model: decart.video('lucy-pro-i2v'),
  image: {
    type: 'binary',
    data: new Uint8Array(imageData),
    mediaType: 'image/jpeg'
  },
  prompt: 'Gentle camera pan across the scene'
});

lucy-dev-i2v

Development version of the image-to-video model.

Capabilities:

• Same as lucy-pro-i2v
• May include experimental features
• May have different performance characteristics

Use Cases:

• Testing new features
• Development and experimentation

import { createDecart } from '@decartai/ai-sdk-provider';
import { experimental_generateVideo } from 'ai';

const decart = createDecart();

const { video } = await experimental_generateVideo({
  model: decart.video('lucy-dev-i2v'),
  image: {
    type: 'url',
    url: 'https://example.com/image.jpg'
  }
});

lucy-motion

Motion-controlled video generation model.

Capabilities:

• Image-to-video with trajectory control
• Define specific motion paths
• Frame-by-frame motion control
• Resolution and aspect ratio control

Use Cases:

• Creating videos with specific camera movements
• Precise control over motion paths
• Complex animation sequences

Special Settings:

• Requires trajectory parameter in providerOptions.decart
• Trajectory defines motion across frames

import { createDecart } from '@decartai/ai-sdk-provider';
import { experimental_generateVideo } from 'ai';

const decart = createDecart();

const { video } = await experimental_generateVideo({
  model: decart.video('lucy-motion'),
  image: {
    type: 'url',
    url: 'https://example.com/image.jpg'
  },
  providerOptions: {
    decart: {
      trajectory: [
        { frame: 0, x: 0.5, y: 0.5 },
        { frame: 15, x: 0.6, y: 0.4 },
        { frame: 30, x: 0.7, y: 0.5 }
      ]
    }
  }
});

Model Specifications

specificationVersion · string · default: "v3"

All Decart video models implement the AI SDK v3 experimental specification.

maxVideosPerCall · number · default: "1"

Each generation call produces exactly one video.

provider · string · default: "decart.video"

Provider identifier for the video models.

Comparison Matrix

Feature	lucy-pro-t2v	lucy-pro-i2v	lucy-dev-i2v	lucy-motion
Text-to-video	✓	—	—	—
Image-to-video	—	✓	✓	✓
Trajectory control	—	—	—	✓
Resolution control	✓	✓	✓	✓
Aspect ratio control	✓	✓	✓	✓
Seed support	✓	✓	✓	✓

Generation Process

All video models use an asynchronous job-based generation process:

1. Job Submission - Request is submitted and returns a job ID
2. Polling - Status is polled until completion
3. Download - Video content is downloaded when ready

The polling behavior can be customized using provider options:

const { video } = await experimental_generateVideo({
  model: decart.video('lucy-pro-t2v'),
  prompt: 'A video',
  providerOptions: {
    decart: {
      pollIntervalMs: 2000,  // Poll every 2 seconds
      pollTimeoutMs: 600000  // Timeout after 10 minutes
    }
  }
});

Response Format

All video models return the same response structure:

{
  videos: [{
    type: 'binary',
    data: Uint8Array,
    mediaType: 'video/mp4'
  }],
  warnings: Array<{
    type: 'unsupported',
    feature: string,
    details?: string
  }>,
  response: {
    timestamp: Date,
    modelId: string,
    headers: Record<string, string>
  }
}