Text-to-Video

Generate videos from text prompts using lucy-pro-t2v

The lucy-pro-t2v model generates videos directly from text descriptions without requiring an input image. This is ideal for creating videos from scratch based on natural language prompts.

Basic Usage

1. Import Dependencies

   Import the Decart provider and the experimental video generation function:

   import { decart } from '@decartai/ai-sdk-provider';
   import { experimental_generateVideo as generateVideo } from 'ai';
   import fs from 'fs';
   
2. Generate Video

   Call generateVideo with the lucy-pro-t2v model and your text prompt:

   const { videos } = await generateVideo({
     model: decart.video('lucy-pro-t2v'),
     prompt: 'A man is riding a horse in a field',
   });
   

   Source Reference: examples/tasks/video-generation-t2v.ts:7-10

3. Save the Video

   Write the generated video to a file:

   fs.writeFileSync('video.mp4', videos[0].uint8Array);
   

Complete Example

import { decart } from '@decartai/ai-sdk-provider';
import { experimental_generateVideo as generateVideo } from 'ai';
import fs from 'fs';

export async function generateTextToVideo() {
  const result = await generateVideo({
    model: decart.video('lucy-pro-t2v'),
    prompt: 'A man is riding a horse in a field',
  });
  
  const filename = `video-${Date.now()}.mp4`;
  fs.writeFileSync(filename, result.videos[0].uint8Array);
  console.log(`Video saved to ${filename}`);
  
  return result;
}

Customizing Output

Aspect Ratio

Control video orientation with the aspectRatio parameter:

Landscape (16:9)

const { videos } = await generateVideo({
  model: decart.video('lucy-pro-t2v'),
  prompt: 'A sunset over the ocean',
  aspectRatio: '16:9',
});

The aspectRatio parameter is converted to an orientation value (landscape or portrait) in the API request.

Source Reference: src/decart-video-model.ts:110-122

Resolution

Set video resolution using the resolution parameter:

720p (High Quality)

const { videos } = await generateVideo({
  model: decart.video('lucy-pro-t2v'),
  prompt: 'A city skyline at night',
  resolution: '1280x720',
});

Only 1280x720 (720p) and 854x480 (480p) resolutions are supported. Other values will generate a warning.

Source Reference: src/decart-video-model.ts:125-135

Reproducible Results

Use the seed parameter to generate reproducible videos:

const { videos } = await generateVideo({
  model: decart.video('lucy-pro-t2v'),
  prompt: 'A tranquil lake surrounded by mountains',
  seed: 42,
});

Using the same prompt, seed, and settings will produce the same video output.

Source Reference: src/decart-video-model.ts:152-154

Advanced Configuration

Custom Polling Behavior

Video generation is asynchronous. Customize the polling behavior:

const { videos } = await generateVideo({
  model: decart.video('lucy-pro-t2v'),
  prompt: 'A spaceship flying through space',
  providerOptions: {
    decart: {
      pollIntervalMs: 2000,  // Poll every 2 seconds (default: 1500)
      pollTimeoutMs: 600000, // 10 minute timeout (default: 300000)
    },
  },
});

pollIntervalMs · number · default: "1500"

Time in milliseconds between status checks during video generation.

Source Reference: src/decart-video-model.ts:196

pollTimeoutMs · number · default: "300000"

Maximum time in milliseconds to wait for video generation (default: 5 minutes).

Source Reference: src/decart-video-model.ts:197

Abort Signal

Cancel long-running video generation requests:

const controller = new AbortController();

// Cancel after 2 minutes
setTimeout(() => controller.abort(), 120000);

const { videos } = await generateVideo({
  model: decart.video('lucy-pro-t2v'),
  prompt: 'A long animation sequence',
  abortSignal: controller.signal,
});

Source Reference: src/decart-video-model.ts:204-209

Writing Effective Prompts

Prompt Best Practices

• Be specific about the subject, action, and environment
• Include motion or camera movement descriptions
• Specify lighting, weather, or atmosphere when relevant
• Keep prompts concise but descriptive (1-2 sentences)

Good Examples:

• “A golden retriever running through a sunny meadow with wildflowers”
• “Aerial view of a city at sunset, camera slowly panning across skyscrapers”
• “Close-up of raindrops falling on a window with a blurred cityscape behind”

Less Effective:

• “A dog” (too vague)
• “A detailed scene with a dog running in a field with flowers and trees and mountains in the background on a sunny day…” (too long)

Response Structure

The generateVideo function returns a result object:

const result = await generateVideo({
  model: decart.video('lucy-pro-t2v'),
  prompt: 'A peaceful forest scene',
});

// Access the video
const video = result.videos[0];
console.log(video.mimeType);      // "video/mp4"
console.log(video.uint8Array);    // Uint8Array containing video data

// Access response metadata
console.log(result.response.modelId);    // "lucy-pro-t2v"
console.log(result.response.timestamp);  // Date object

Source Reference: src/decart-video-model.ts:247-261

Error Handling

try {
  const { videos } = await generateVideo({
    model: decart.video('lucy-pro-t2v'),
    prompt: 'A dramatic storm approaching',
  });
  
  fs.writeFileSync('storm.mp4', videos[0].uint8Array);
} catch (error) {
  if (error.name === 'AI_APICallError') {
    console.error('Video generation failed:', error.message);
  }
  throw error;
}

Common errors include timeout (exceeding 5 minutes), job failure, or network issues.

Source References:

• Timeout error: src/decart-video-model.ts:211-216
• Job failure: src/decart-video-model.ts:230-235

Next Steps

Image-to-Video

Animate existing images into videos

Motion Control

Control camera and subject movement

Settings Reference

View all configuration options

Examples

Browse complete working examples