Skip to main content

​
Overview

Generate videos directly in chat using AI models. Users describe what they want and the model produces a short video clip, which is then playable inline.

​
Quick Start

Toggle in chat.config.ts: 
features: {
  videoGeneration: true, // Requires BLOB_READ_WRITE_TOKEN
}
Video generation requires BLOB_READ_WRITE_TOKEN since generated videos are uploaded to Vercel Blob storage.
Configure the default video model: 
models: {
  defaults: {
    video: "xai/grok-imagine-video",
  },
}

​
Parameters

The tool accepts the following inputs:
ParameterTypeDefaultDescription
promptstring—Descriptive text for the video
aspectRatio"16:9" | "9:16" | "1:1""16:9"Output aspect ratio
durationSecondsnumber (1–10)5Length of the generated video
The LLM fills in these parameters automatically based on the user’s request.

​
Architecture

Follows the Tool Part pattern: 
lib/ai/tools/generate-video.ts → components/part/generate-video.tsx

​
Tool Output

return { videoUrl: result.url, prompt };
The generated video is uploaded to blob storage and the URL returned. Supported container formats are mp4, webm, and mov, resolved from the response media type.

​
UI States

StateShows
input-availablePulsing placeholder + “Generating video: “
output-availableInline video player with controls, autoplay, loop, and mute
failedError message with suggestion to try a different prompt

​
Configuration

​
Video Model

chat.config.ts
models: {
  defaults: {
    video: "xai/grok-imagine-video",
  },
}

​
Model Selection Logic

When the tool runs, it resolves the model to use:
  1. If the user’s selected chat model supports video output (per the app model registry, model.output.video), use it.
  2. Otherwise, fall back to config.models.defaults.video.

​
Google Model Handling

For Google (Gemini) models, aspectRatio is passed in provider-specific options rather than as a top-level parameter.