Video Models Overview

The video model lineup

Skytells gives you access to 9 video models through a single API:

All video models use the same POST /v1/predictions endpoint — just change the model field.

Choosing a video model

For photorealistic content

Google Veo 3.1 produces the most photorealistic video outputs, especially for nature, human subjects, and architectural scenes.

{
  "model": "veo-3.1",
  "input": {
    "prompt": "A timelapse of storm clouds forming over the ocean, cinematic, 4K",
    "duration_seconds": 5,
    "aspect_ratio": "16:9"
  }
}

For creative and stylized content

Sora 2 excels at imaginative, surreal, and stylized scenarios:

{
  "model": "sora-2",
  "input": {
    "prompt": "A tiny wizard casting spells in a library made entirely of books, whimsical, soft light",
    "duration_seconds": 6
  }
}

For Skytells-native production use

TrueFusion Video Pro is the recommended default for production applications — it balances quality, speed, and cost:

{
  "model": "truefusion-video-pro",
  "input": {
    "prompt": "Corporate explainer animation, clean design, text overlays",
    "duration_seconds": 8,
    "aspect_ratio": "16:9",
    "fps": 24
  }
}

For lip-sync

LipFusion takes a source image (face) and an audio file, and produces a video where the face mouth movements match the audio:

{
  "model": "lipfusion",
  "input": {
    "face_image_url": "https://example.com/headshot.jpg",
    "audio_url": "https://example.com/speech.mp3"
  }
}

LipFusion is ideal for:

• AI avatar presenters
• Dubbed video localization
• Synthetic spokesperson generation

Common video parameters

Video generation is asynchronous

Unlike image models (which often return in seconds), video generation typically takes 30 seconds to several minutes. Always use async polling or webhooks for video predictions.

Status flow: queued → processing → succeeded
Typical time: 30s (fast models) to 5min (high-quality long videos)

You'll implement polling and webhook handling in the next module.

Summary

• 9 video models available: 5 Skytells-native + 2 Google Veo + 2 OpenAI Sora
• Use truefusion-video-pro as your production default
• Use veo-3.1 for photorealism, sora-2 for creative work
• Use lipfusion for lip-sync and AI avatars
• Video generation is always async — plan for polling or webhooks