Synthome Docs
Guides

Multi-Scene Videos

Create complex videos with multiple scenes, transitions, and media types

Multi-Scene Videos

Learn how to create professional multi-scene videos by combining generated content, existing media, and various operations.

Basic Multi-Scene Video

Use merge() to combine multiple scenes sequentially:

import { compose, generateVideo, merge, videoModel } from "@synthome/sdk";

const execution = await compose(
  merge([
    generateVideo({
      model: videoModel("bytedance/seedance-1-pro", "replicate"),
      prompt: "Scene 1: A sunrise over mountains, golden light",
      duration: 5,
    }),
    generateVideo({
      model: videoModel("bytedance/seedance-1-pro", "replicate"),
      prompt: "Scene 2: Birds flying across the sky",
      duration: 5,
    }),
    generateVideo({
      model: videoModel("bytedance/seedance-1-pro", "replicate"),
      prompt: "Scene 3: A peaceful lake reflecting mountains",
      duration: 5,
    }),
  ]),
).execute();

console.log("Multi-scene video:", execution.result?.url);

All scenes generate in parallel, then merge sequentially. Total video: 15 seconds.

Mixed Media Scenes

Combine generated videos with existing media:

import {
  compose,
  generateVideo,
  generateImage,
  merge,
  videoModel,
  imageModel,
} from "@synthome/sdk";

const execution = await compose(
  merge([
    // Existing intro video
    "https://your-cdn.com/intro.mp4",

    // Generated scene
    generateVideo({
      model: videoModel("bytedance/seedance-1-pro", "replicate"),
      prompt: "Product showcase, sleek design, rotating view",
      duration: 5,
    }),

    // Image as a scene (displayed for specified duration)
    {
      media: generateImage({
        model: imageModel("google/nano-banana", "fal"),
        prompt: "Product features infographic",
      }),
      duration: 3,
    },

    // Existing outro
    "https://your-cdn.com/outro.mp4",
  ]),
).execute();

Adding Audio to Scenes

Background Music

Add background music by including an audio file in your merge. The audio plays from the start across the entire video duration:

import { compose, generateVideo, merge, videoModel } from "@synthome/sdk";

const execution = await compose(
  merge([
    // Video scenes
    generateVideo({
      model: videoModel("bytedance/seedance-1-pro", "replicate"),
      prompt: "Scene 1: Ocean waves",
      duration: 5,
    }),
    generateVideo({
      model: videoModel("bytedance/seedance-1-pro", "replicate"),
      prompt: "Scene 2: Beach sunset",
      duration: 5,
    }),
    // Background music - plays from start across full duration
    "https://your-cdn.com/background-music.mp3",
  ]),
).execute();

Generated Voiceover

Add AI-generated narration the same way:

import {
  compose,
  generateVideo,
  generateAudio,
  merge,
  videoModel,
  audioModel,
} from "@synthome/sdk";

const execution = await compose(
  merge([
    // Video scenes
    generateVideo({
      model: videoModel("bytedance/seedance-1-pro", "replicate"),
      prompt: "Mountain landscape, cinematic",
      duration: 5,
    }),
    generateVideo({
      model: videoModel("bytedance/seedance-1-pro", "replicate"),
      prompt: "Forest trail, morning mist",
      duration: 5,
    }),
    // AI voiceover - plays from start
    generateAudio({
      model: audioModel("elevenlabs/turbo-v2.5", "elevenlabs"),
      text: "Discover the beauty of nature. From majestic mountains to serene forest trails, adventure awaits.",
      voiceId: "EXAVITQu4vr4xnSDxMaL",
    }),
  ]),
).execute();

Scene with Captions

Add auto-generated captions to your video:

import {
  compose,
  generateVideo,
  generateAudio,
  merge,
  captions,
  videoModel,
  audioModel,
} from "@synthome/sdk";

// First, create the video with audio
const videoWithAudio = merge([
  generateVideo({
    model: videoModel("bytedance/seedance-1-pro", "replicate"),
    prompt: "Scene 1: Product introduction",
    duration: 5,
  }),
  generateVideo({
    model: videoModel("bytedance/seedance-1-pro", "replicate"),
    prompt: "Scene 2: Product features",
    duration: 5,
  }),
  // Voiceover
  generateAudio({
    model: audioModel("elevenlabs/turbo-v2.5", "elevenlabs"),
    text: "Introducing our new product. It features cutting-edge technology and sleek design.",
    voiceId: "EXAVITQu4vr4xnSDxMaL",
  }),
]);

// Then add auto-generated captions
const execution = await compose(
  captions({
    video: videoWithAudio,
    transcribe: {
      model: "openai/whisper",
      provider: "replicate",
    },
    style: {
      position: "bottom",
      fontSize: 24,
      fontColor: "#FFFFFF",
      backgroundColor: "rgba(0,0,0,0.7)",
    },
  }),
).execute();

Picture-in-Picture Scenes

Create videos with overlay content:

import { compose, generateVideo, layers, videoModel } from "@synthome/sdk";

const execution = await compose(
  layers({
    layers: [
      // Main video (full screen)
      {
        media: generateVideo({
          model: videoModel("bytedance/seedance-1-pro", "replicate"),
          prompt: "Conference presentation, speaker on stage",
          duration: 10,
        }),
        placement: "full",
      },
      // Picture-in-picture (slide content)
      {
        media: "https://your-cdn.com/slides.mp4",
        placement: "picture-in-picture",
      },
    ],
  }),
).execute();

Speaking Head with Custom Background

Create a fully AI-generated talking head video with a custom background. This example:

  1. Generates a portrait image on a green screen
  2. Generates speech audio from text
  3. Creates a lip-synced video using Fabric
  4. Removes the green screen and composites onto a custom background
import {
  compose,
  generateVideo,
  generateImage,
  generateAudio,
  layers,
  videoModel,
  imageModel,
  audioModel,
} from "@synthome/sdk";

const execution = await compose(
  layers({
    layers: [
      // Background layer
      {
        media: generateImage({
          model: imageModel("google/nano-banana", "fal"),
          prompt: "Modern office interior, blurred background, professional",
        }),
        placement: "full",
      },
      // Speaking head with green screen removed
      {
        media: generateVideo({
          model: videoModel("veed/fabric-1.0", "fal"),
          // Portrait on green screen
          image: generateImage({
            model: imageModel("google/nano-banana", "fal"),
            prompt:
              "Professional woman, business attire, neutral expression, green screen background",
          }),
          // Generated speech
          audio: generateAudio({
            model: audioModel("elevenlabs/turbo-v2.5", "elevenlabs"),
            text: "Welcome to our company. Let me tell you about our latest innovations.",
            voiceId: "EXAVITQu4vr4xnSDxMaL",
          }),
        }),
        placement: "full",
        chromaKey: true,
        chromaKeyColor: "#00FF00",
      },
    ],
  }),
).execute();

This pipeline runs in parallel where possible:

  • The background image and portrait image generate simultaneously
  • The speech audio generates in parallel
  • Fabric creates the lip-synced video once the portrait and audio are ready
  • Finally, the green screen is removed and composited onto the background

Variations

With existing portrait:

generateVideo({
  model: videoModel("veed/fabric-1.0", "fal"),
  image: "https://your-cdn.com/spokesperson-greenscreen.png",
  audio: generateAudio({
    model: audioModel("elevenlabs/turbo-v2.5", "elevenlabs"),
    text: "Your message here.",
    voiceId: "EXAVITQu4vr4xnSDxMaL",
  }),
});

With existing background:

layers({
  layers: [
    {
      media: "https://your-cdn.com/office-background.jpg",
      placement: "full",
    },
    {
      media: generateVideo({ ... }),
      placement: "full",
      chromaKey: true,
      chromaKeyColor: "#00FF00",
    },
  ],
});

Dynamic Scene Generation

Generate scenes programmatically from data:

import { compose, generateVideo, merge, videoModel } from "@synthome/sdk";

interface SceneConfig {
  prompt: string;
  duration: number;
}

const scenes: SceneConfig[] = [
  { prompt: "Dawn breaking over a city skyline", duration: 4 },
  { prompt: "Morning commuters in a busy street", duration: 3 },
  { prompt: "Coffee shop interior, cozy atmosphere", duration: 3 },
  { prompt: "Sunset over the same city skyline", duration: 4 },
];

const execution = await compose(
  merge(
    scenes.map((scene) =>
      generateVideo({
        model: videoModel("bytedance/seedance-1-pro", "replicate"),
        prompt: scene.prompt,
        duration: scene.duration,
      }),
    ),
  ),
).execute();

Parallel Scene Generation

Synthome automatically parallelizes independent operations. In this example, all three scenes generate simultaneously:

const execution = await compose(
  merge([
    // These run in parallel
    generateVideo({ ... }),  // Scene 1
    generateVideo({ ... }),  // Scene 2
    generateVideo({ ... }),  // Scene 3
  ])
).execute();

// Total time ≈ longest scene generation time + merge time
// NOT: scene1 + scene2 + scene3 + merge

Complex Multi-Layer Video

Combine multiple techniques for a professional result:

import {
  compose,
  generateVideo,
  generateAudio,
  merge,
  layers,
  captions,
  videoModel,
  audioModel,
} from "@synthome/sdk";

// Build a complete video with:
// - Multiple scenes
// - Logo overlay
// - Background music
// - Voiceover
// - Auto-generated captions

const videoWithAudio = merge([
  // Intro with logo overlay
  layers({
    layers: [
      {
        media: generateVideo({
          model: videoModel("bytedance/seedance-1-pro", "replicate"),
          prompt: "Abstract flowing particles, brand intro",
          duration: 3,
        }),
        placement: "full",
      },
      {
        media: "https://your-cdn.com/logo.png",
        placement: "center",
      },
    ],
  }),

  // Main scene
  generateVideo({
    model: videoModel("bytedance/seedance-1-pro", "replicate"),
    prompt: "Product reveal, dramatic lighting",
    duration: 5,
  }),

  // Feature highlight with text overlay
  layers({
    layers: [
      {
        media: generateVideo({
          model: videoModel("bytedance/seedance-1-pro", "replicate"),
          prompt: "Product features demonstration",
          duration: 5,
        }),
        placement: "full",
      },
      {
        media: "https://your-cdn.com/feature-text.png",
        placement: "bottom-center",
      },
    ],
  }),

  // Voiceover - plays from start
  generateAudio({
    model: audioModel("elevenlabs/turbo-v2.5", "elevenlabs"),
    text: "Welcome to our brand. Discover innovation. Experience excellence.",
    voiceId: "EXAVITQu4vr4xnSDxMaL",
  }),

  // Background music - plays from start
  "https://your-cdn.com/background-music.mp3",
]);

// Add auto-generated captions
const execution = await compose(
  captions({
    video: videoWithAudio,
    transcribe: {
      model: "openai/whisper",
      provider: "replicate",
    },
    style: {
      position: "bottom",
      fontSize: 20,
    },
  }),
).execute();

Best Practices

1. Plan Your Scenes

Sketch out your video structure before coding:

1. Intro (3s) - Logo animation
2. Scene A (5s) - Product overview
3. Scene B (5s) - Feature 1
4. Scene C (5s) - Feature 2
5. Outro (3s) - Call to action

2. Keep Scenes Consistent

Use consistent prompting for visual coherence:

const style = "cinematic, 4K, professional lighting";

const scenes = [
  `Product on white background, ${style}`,
  `Product in use, ${style}`,
  `Product close-up detail, ${style}`,
];

3. Optimize Duration

  • Keep individual scenes 3-7 seconds for engagement
  • Total video length depends on platform (15s for ads, 60s for content)
  • Match audio duration to video duration

4. Use Existing Assets

Mix generated content with existing branded assets for consistency:

merge([
  "https://cdn.brand.com/intro.mp4", // Existing brand intro
  generateVideo({ ... }),             // Generated content
  "https://cdn.brand.com/outro.mp4", // Existing brand outro
])

Next Steps

How is this guide?

AI Agents

Use AI agents to dynamically generate and execute media pipelines

Webhooks

Handle long-running pipelines with async webhook notifications

On this page

Multi-Scene VideosBasic Multi-Scene VideoMixed Media ScenesAdding Audio to ScenesBackground MusicGenerated VoiceoverScene with CaptionsPicture-in-Picture ScenesSpeaking Head with Custom BackgroundVariationsDynamic Scene GenerationParallel Scene GenerationComplex Multi-Layer VideoBest Practices1. Plan Your Scenes2. Keep Scenes Consistent3. Optimize Duration4. Use Existing AssetsNext Steps