xAI Grok Provider

The xAI Grok provider contains language model support for the xAI API.

Setup

The xAI Grok provider is available via the @ai-sdk/xai module. You can install it with

pnpm add @ai-sdk/xai

Provider Instance

You can import the default provider instance xai from @ai-sdk/xai:

1
import { xai } from '@ai-sdk/xai';

If you need a customized setup, you can import createXai from @ai-sdk/xai and create a provider instance with your settings:

1
import { createXai } from '@ai-sdk/xai';
2

3
const xai = createXai({
4
  apiKey: 'your-api-key',
5
});

You can use the following optional settings to customize the xAI provider instance:

• baseURL string

  Use a different URL prefix for API calls, e.g. to use proxy servers. The default prefix is https://api.x.ai/v1.

• apiKey string

  API key that is being sent using the Authorization header. It defaults to the XAI_API_KEY environment variable.

• headers Record<string,string>

  Custom headers to include in the requests.

• fetch (input: RequestInfo, init?: RequestInit) => Promise<Response>

  Custom fetch implementation. Defaults to the global fetch function. You can use it as a middleware to intercept requests, or to provide a custom fetch implementation for e.g. testing.

Language Models

You can create xAI models using a provider instance. The first argument is the model id, e.g. grok-4.6.

1
const model = xai('grok-4.6');

Example

You can use xAI language models to generate text with the generateText function:

1
import { xai } from '@ai-sdk/xai';
2
import { generateText } from 'ai';
3

4
const { text } = await generateText({
5
  model: xai('grok-4.6'),
6
  prompt: 'Write a vegetarian lasagna recipe for 4 people.',
7
});

xAI language models can also be used in the streamText function and support structured data generation with Output (see AI SDK Core).

Reasoning Effort

For models with configurable reasoning, you can control how much effort the model spends thinking before responding via providerOptions.xai.reasoningEffort. This works for both the Responses API (default) and the Chat Completions API (xai.chat()).

1
import { xai } from '@ai-sdk/xai';
2
import { generateText } from 'ai';
3

4
const { text } = await generateText({
5
  model: xai('grok-4.3'),
6
  prompt: 'Explain quantum entanglement.',
7
  providerOptions: {
8
    xai: { reasoningEffort: 'medium' },
9
  },
10
});

The AI SDK option accepts these values, but each xAI model supports a subset:

• 'none' — Disables reasoning entirely; no thinking tokens are used. Best for simple use cases that need a near-instant response.
• 'low' — Uses some reasoning tokens, but still fast. Good for general agentic use and tool calling.
• 'medium' — More thinking for less-latency-sensitive applications such as complex data analysis and long-context reasoning.
• 'high' — More reasoning tokens for deeper thinking. Suited for very challenging problems, complex math, multi-step logic, and competition-level tasks.
• 'xhigh' — Uses the most reasoning tokens for the most challenging tasks. This level is only supported by grok-4.6.

Priority Processing

providerOptions.xai.serviceTier requests higher scheduling priority, which typically lowers time-to-first-token and speeds up inter-token latency. This works for both the Responses API (default) and the Chat Completions API (xai.chat()).

1
import { xai } from '@ai-sdk/xai';
2
import { generateText } from 'ai';
3

4
const { providerMetadata } = await generateText({
5
  model: xai('grok-4.6'),
6
  prompt: 'Explain quantum entanglement.',
7
  providerOptions: {
8
    xai: { serviceTier: 'priority' },
9
  },
10
});
11

12
// 'priority' when the request was served at the priority tier,
13
// 'default' when priority capacity was unavailable.
14
console.log(providerMetadata?.xai?.serviceTier);

Priority requests are billed at a premium per-token rate, and xAI only charges that rate when the response confirms the priority tier — so read the applied tier back from providerMetadata.xai.serviceTier rather than assuming the request you sent is the tier you got. Omitting the option is equivalent to 'default'. See xAI's priority processing docs.

Realtime Models

You can create models that call the xAI Realtime API using the .experimental_realtime() factory method.

1
import { xai } from '@ai-sdk/xai';
2

3
const model = xai.experimental_realtime('grok-voice-latest');

Realtime sessions run in the browser and require a short-lived token created on your server with xai.experimental_realtime.getToken():

1
const token = await xai.experimental_realtime.getToken({
2
  model: 'grok-voice-latest',
3
});

See Realtime for the complete setup and tool calling pattern.

Responses API (Agentic Tools)

The xAI Responses API is the default when using xai(modelId) (since AI SDK 7). You can also use xai.responses(modelId) explicitly. This enables the model to autonomously orchestrate tool calls and research on xAI's servers.

1
const model = xai.responses('grok-4.6');

The Responses API provides server-side tools that the model can autonomously execute during its reasoning process:

• web_search: Real-time web search and page browsing
• x_search: Search X (Twitter) posts, users, and threads
• code_execution: Execute Python code for calculations and data analysis
• view_image: View and analyze images
• view_x_video: View and analyze videos from X posts
• mcp_server: Connect to remote MCP servers and use their tools
• file_search: Search through documents in vector stores (collections)

Vision

The Responses API supports image input with vision models:

1
import { xai } from '@ai-sdk/xai';
2
import { generateText } from 'ai';
3

4
const { text } = await generateText({
5
  model: xai.responses('grok-4.6'),
6
  messages: [
7
    {
8
      role: 'user',
9
      content: [
10
        { type: 'text', text: 'What do you see in this image?' },
11
        {
12
          type: 'file',
13
          mediaType: 'image',
14
          data: fs.readFileSync('./image.png'),
15
        },
16
      ],
17
    },
18
  ],
19
});

You can control the resolution at which the model processes an image with the imageDetail provider option on the image part:

1
import { xai } from '@ai-sdk/xai';
2
import { generateText } from 'ai';
3

4
const { text } = await generateText({
5
  model: xai('grok-4.6'),
6
  messages: [
7
    {
8
      role: 'user',
9
      content: [
10
        { type: 'text', text: 'What do you see in this image?' },
11
        {
12
          type: 'file',
13
          mediaType: 'image/png',
14
          data: fs.readFileSync('./image.png'),
15
          providerOptions: {
16
            xai: { imageDetail: 'low' },
17
          },
18
        },
19
      ],
20
    },
21
  ],
22
});

The following image detail values are supported:

• low: processes the image at reduced resolution and consumes fewer input tokens.
• high: processes the image at full resolution.
• auto: lets the xAI API decide.

When not set, the image is processed at full resolution.

Web Search Tool

The web search tool enables autonomous web research with optional domain filtering and image understanding:

1
import { xai } from '@ai-sdk/xai';
2
import { generateText } from 'ai';
3

4
const { text, sources } = await generateText({
5
  model: xai.responses('grok-4.6'),
6
  prompt: 'What are the latest developments in AI?',
7
  tools: {
8
    web_search: xai.tools.webSearch({
9
      allowedDomains: ['arxiv.org', 'openai.com'],
10
      enableImageUnderstanding: true,
11
    }),
12
  },
13
});
14

15
console.log(text);
16
console.log('Citations:', sources);

Web Search Parameters

• allowedDomains string[]

  Only search within specified domains (max 5). Cannot be used with excludedDomains.

• excludedDomains string[]

  Exclude specified domains from search (max 5). Cannot be used with allowedDomains.

• enableImageSearch boolean

  Allow the model to perform image search as a separate Web Search mode when images are useful for the answer. The model can choose between regular web search and image search; responses can include Markdown image embeds when relevant.

• enableImageUnderstanding boolean

  Enable the model to view and analyze images found during search. Increases token usage. To allow explicitly searching for images, use enableImageSearch.

X Search Tool

The X search tool enables searching X (Twitter) for posts, with filtering by handles and date ranges:

1
const { text, sources } = await generateText({
2
  model: xai.responses('grok-4.6'),
3
  prompt: 'What are people saying about AI on X this week?',
4
  tools: {
5
    x_search: xai.tools.xSearch({
6
      allowedXHandles: ['elonmusk', 'xai'],
7
      fromDate: '2025-10-23',
8
      toDate: '2025-10-30',
9
      enableImageUnderstanding: true,
10
      enableVideoUnderstanding: true,
11
    }),
12
  },
13
});

X Search Parameters

• allowedXHandles string[]

  Only search posts from specified X handles (max 10). Cannot be used with excludedXHandles.

• excludedXHandles string[]

  Exclude posts from specified X handles (max 10). Cannot be used with allowedXHandles.

• fromDate string

  Start date for posts in ISO8601 format (YYYY-MM-DD).

• toDate string

  End date for posts in ISO8601 format (YYYY-MM-DD).

• enableImageUnderstanding boolean

  Enable the model to view and analyze images in X posts.

• enableVideoUnderstanding boolean

  Enable the model to view and analyze videos in X posts.

Code Execution Tool

The code execution tool enables the model to write and execute Python code for calculations and data analysis:

1
const { text } = await generateText({
2
  model: xai.responses('grok-4.6'),
3
  prompt:
4
    'Calculate the compound interest for $10,000 at 5% annually for 10 years',
5
  tools: {
6
    code_execution: xai.tools.codeExecution(),
7
  },
8
});

View Image Tool

The view image tool enables the model to view and analyze images:

1
const { text } = await generateText({
2
  model: xai.responses('grok-4.6'),
3
  prompt: 'Describe what you see in the image',
4
  tools: {
5
    view_image: xai.tools.viewImage(),
6
  },
7
});

View X Video Tool

The view X video tool enables the model to view and analyze videos from X (Twitter) posts:

1
const { text } = await generateText({
2
  model: xai.responses('grok-4.6'),
3
  prompt: 'Summarize the content of this X video',
4
  tools: {
5
    view_x_video: xai.tools.viewXVideo(),
6
  },
7
});

Image Generation Tool

The image generation tool lets the model create and edit images with Grok Imagine as part of a conversation. The model decides when to call the tool, writes the image prompt, and returns the finished image alongside its text response:

1
import { xai } from '@ai-sdk/xai';
2
import { generateText } from 'ai';
3

4
const result = await generateText({
5
  model: xai.responses('grok-4.6'),
6
  prompt:
7
    'Generate an image of a corgi surfing a big wave, in the style of a Japanese woodblock print',
8
  tools: {
9
    image_generation: xai.tools.imageGeneration(),
10
  },
11
});
12

13
for (const toolResult of result.staticToolResults) {
14
  if (toolResult.toolName === 'image_generation') {
15
    const base64Image = toolResult.output.result;
16
  }
17
}

The tool result also contains the prompt that the model wrote for the image model, which is useful for understanding and debugging what was generated.

Image Generation Parameters

• action 'auto' | 'generate' | 'edit'

  Restricts what the tool can do. Defaults to 'auto'.

  • 'auto': the model can generate and edit images.
  • 'generate': text-to-image generation only.
  • 'edit': image editing only (edits images already in the conversation, including images the model generated earlier).

MCP Server Tool

The MCP server tool enables the model to connect to remote Model Context Protocol (MCP) servers and use their tools:

1
const { text } = await generateText({
2
  model: xai.responses('grok-4.6'),
3
  prompt: 'Use the weather tool to check conditions in San Francisco',
4
  tools: {
5
    weather_server: xai.tools.mcpServer({
6
      serverUrl: 'https://example.com/mcp',
7
      serverLabel: 'weather-service',
8
      serverDescription: 'Weather data provider',
9
      allowedTools: ['get_weather', 'get_forecast'],
10
    }),
11
  },
12
});

MCP Server Parameters

• serverUrl string (required)

  The URL of the remote MCP server.

• serverLabel string

  A label to identify the MCP server.

• serverDescription string

  A description of what the MCP server provides.

• allowedTools string[]

  List of tool names that the model is allowed to use from the MCP server. If not specified, all tools are allowed.

• headers Record<string, string>

  Custom headers to include when connecting to the MCP server.

• authorization string

  Authorization header value for authenticating with the MCP server (e.g., 'Bearer token123').

File Search Tool

The file search tool enables searching through documents stored in xAI vector stores (collections):

1
import { xai, type XaiLanguageModelResponsesOptions } from '@ai-sdk/xai';
2
import { streamText } from 'ai';
3

4
const result = streamText({
5
  model: xai.responses('grok-4.6'),
6
  prompt: 'What documents do you have access to?',
7
  tools: {
8
    file_search: xai.tools.fileSearch({
9
      vectorStoreIds: ['collection_your-collection-id'],
10
      maxNumResults: 10,
11
    }),
12
  },
13
  providerOptions: {
14
    xai: {
15
      include: ['file_search_call.results'],
16
    } satisfies XaiLanguageModelResponsesOptions,
17
  },
18
});

File Search Parameters

• vectorStoreIds string[] (required)

  The IDs of the vector stores (collections) to search.

• maxNumResults number

  The maximum number of results to return from the search.

Provider Options for File Search

• include Array<'file_search_call.results'>

  Include file search results in the response. When set to ['file_search_call.results'], the response will contain the actual search results with file content and scores.

Multiple Tools

You can combine multiple server-side tools for comprehensive research:

1
import { xai } from '@ai-sdk/xai';
2
import { streamText } from 'ai';
3

4
const { stream } = streamText({
5
  model: xai.responses('grok-4.6'),
6
  prompt: 'Research AI safety developments and calculate risk metrics',
7
  tools: {
8
    web_search: xai.tools.webSearch(),
9
    x_search: xai.tools.xSearch(),
10
    code_execution: xai.tools.codeExecution(),
11
    file_search: xai.tools.fileSearch({
12
      vectorStoreIds: ['collection_your-documents'],
13
    }),
14
    data_service: xai.tools.mcpServer({
15
      serverUrl: 'https://data.example.com/mcp',
16
      serverLabel: 'data-service',
17
    }),
18
  },
19
});
20

21
for await (const part of stream) {
22
  if (part.type === 'text-delta') {
23
    process.stdout.write(part.text);
24
  } else if (part.type === 'source' && part.sourceType === 'url') {
25
    console.log('\nSource:', part.url);
26
  }
27
}

Provider Options

The Responses API supports the following provider options:

1
import { xai, type XaiLanguageModelResponsesOptions } from '@ai-sdk/xai';
2
import { generateText } from 'ai';
3

4
const result = await generateText({
5
  model: xai.responses('grok-4.6'),
6
  providerOptions: {
7
    xai: {
8
      reasoningEffort: 'high',
9
    } satisfies XaiLanguageModelResponsesOptions,
10
  },
11
  // ...
12
});

The following provider options are available:

• reasoningEffort 'none' | 'low' | 'medium' | 'high'

  Control the reasoning effort for supported models. See Reasoning Effort for model-specific values and defaults.

• logprobs boolean

  Return log probabilities for output tokens.

• topLogprobs number

  Number of most likely tokens to return per token position (0-8). When set, logprobs is automatically enabled.

• include Array<'file_search_call.results'>

  Specify additional output data to include in the model response. Use ['file_search_call.results'] to include file search results with scores and content.

• store boolean

  Whether to store the input message(s) and model response for later retrieval. Defaults to true.

• previousResponseId string

  The ID of the previous response from the model. You can use it to continue a conversation.

• serviceTier 'default' | 'priority'

  Scheduling priority for the request. 'priority' buys lower time-to-first-token and faster inter-token latency at a premium per-token price. The tier xAI actually applied comes back on providerMetadata.xai.serviceTier, and is 'default' when priority capacity was unavailable. See Priority Processing.

Model Capabilities

Speech Models

You can create models that call the xAI Text to Speech API using the .speech() factory method. xAI's text-to-speech endpoint does not require a model identifier.

1
const model = xai.speech();

Use xAI speech models with the generateSpeech function:

1
import { xai } from '@ai-sdk/xai';
2
import { generateSpeech } from 'ai';
3

4
const result = await generateSpeech({
5
  model: xai.speech(),
6
  text: 'Hello from the AI SDK!',
7
  voice: 'ara',
8
  language: 'en',
9
  outputFormat: 'mp3',
10
  speed: 1.1,
11
});

Supported Parameters

• text string (required)

  Text to synthesize. You can include xAI speech tags such as [pause], [laugh], or <whisper>...</whisper> directly in the text.

• voice string

  The xAI voice ID. Defaults to 'eve'. Built-in voice IDs are 'eve', 'ara', 'rex', 'sal', and 'leo'; custom voice IDs are also accepted.

• language string

  A BCP-47 language code or 'auto' for automatic detection. Defaults to 'auto' when omitted.

• speed number

  Speech rate multiplier from 0.7 to 1.5.

• outputFormat string

  The audio codec to generate. Supported values are 'mp3', 'wav', 'pcm', 'mulaw', and 'alaw'. Defaults to 'mp3'.

Provider Options

You can pass xAI-specific controls using providerOptions.xai:

1
import { xai, type XaiSpeechModelOptions } from '@ai-sdk/xai';
2
import { generateSpeech } from 'ai';
3

4
const result = await generateSpeech({
5
  model: xai.speech(),
6
  text: 'A high fidelity narration sample.',
7
  outputFormat: 'mp3',
8
  providerOptions: {
9
    xai: {
10
      sampleRate: 44100,
11
      bitRate: 192000,
12
      optimizeStreamingLatency: 1,
13
      textNormalization: true,
14
    } satisfies XaiSpeechModelOptions,
15
  },
16
});

• sampleRate 8000 | 16000 | 22050 | 24000 | 44100 | 48000

  Sample rate of the output audio in Hz.

• bitRate 32000 | 64000 | 96000 | 128000 | 192000

  MP3 bit rate in bits per second. Only applies when outputFormat is 'mp3'.

• optimizeStreamingLatency 0 | 1 | 2

  Latency optimization level. Higher values reduce time to first audio with a quality tradeoff at chunk boundaries.

• textNormalization boolean

  Whether to normalize written-form input text before synthesizing speech.

Model Capabilities

Transcription Models

You can create models that call the xAI Speech to Text API using the .transcription() factory method. xAI's batch transcription endpoint does not require a model identifier.

1
const model = xai.transcription();

Use xAI transcription models with the transcribe function:

1
import { xai } from '@ai-sdk/xai';
2
import { transcribe } from 'ai';
3
import { readFile } from 'fs/promises';
4

5
const result = await transcribe({
6
  model: xai.transcription(),
7
  audio: await readFile('meeting.mp3'),
8
});

Provider Options

You can pass xAI-specific controls using providerOptions.xai:

1
import { xai, type XaiTranscriptionModelOptions } from '@ai-sdk/xai';
2
import { transcribe } from 'ai';
3
import { readFile } from 'fs/promises';
4

5
const result = await transcribe({
6
  model: xai.transcription(),
7
  audio: await readFile('meeting.mp3'),
8
  providerOptions: {
9
    xai: {
10
      language: 'en',
11
      format: true,
12
      keyterm: ['AI SDK', 'Grok'],
13
      diarize: true,
14
    } satisfies XaiTranscriptionModelOptions,
15
  },
16
});

• audioFormat pcm | mulaw | alaw

  Audio encoding for raw, headerless input audio.

• sampleRate 8000 | 16000 | 22050 | 24000 | 44100 | 48000

  Sample rate of the input audio in Hz.

• language string

  Language code used for inverse text normalization.

• format boolean

  Enables inverse text normalization. Requires language.

• multichannel boolean

  Enables per-channel transcription for interleaved multichannel audio.

• channels 2 | 3 | 4 | 5 | 6 | 7 | 8

  Number of interleaved audio channels.

• diarize boolean

  Enables speaker diarization.

• keyterm string | string[]

  One or more terms to bias transcription toward.

• fillerWords boolean

  Includes filler words such as uh and um in the transcript.

• streaming object

  Options for streaming speech-to-text over WebSocket. Use with experimental_streamTranscribe.

  • interimResults boolean

    Emits partial transcripts while speech is being processed.

  • endpointing number

    Silence duration in milliseconds before an utterance-final event. Range: 0-5000.

  • smartTurn number

    End-of-turn detection threshold. When set, enables Smart Turn. Range: 0.0-1.0.

  • smartTurnTimeout number

    Maximum silence duration in milliseconds before forcing speech_final. Range: 1-5000.

Model Capabilities

Image Models

You can create xAI image models using the .image() factory method. For more on image generation with the AI SDK see generateImage().

1
import { xai } from '@ai-sdk/xai';
2
import { generateImage } from 'ai';
3

4
const { image } = await generateImage({
5
  model: xai.image('grok-imagine-image'),
6
  prompt: 'A futuristic cityscape at sunset',
7
});

Image Editing

xAI supports image editing through the grok-imagine-image model. Pass input images via prompt.images to transform or edit existing images.

Basic Image Editing

Transform an existing image using text prompts:

1
import { xai } from '@ai-sdk/xai';
2
import { generateImage } from 'ai';
3
import { readFileSync } from 'fs';
4

5
const imageBuffer = readFileSync('./input-image.png');
6

7
const { images } = await generateImage({
8
  model: xai.image('grok-imagine-image'),
9
  prompt: {
10
    text: 'Turn the cat into a golden retriever dog',
11
    images: [imageBuffer],
12
  },
13
});

Multi-Image Editing

Combine or reference multiple input images in the prompt:

1
import { xai } from '@ai-sdk/xai';
2
import { generateImage } from 'ai';
3
import { readFileSync } from 'fs';
4

5
const cat = readFileSync('./cat.png');
6
const dog = readFileSync('./dog.png');
7

8
const { images } = await generateImage({
9
  model: xai.image('grok-imagine-image'),
10
  prompt: {
11
    text: 'Combine these two animals into a group photo',
12
    images: [cat, dog],
13
  },
14
});

Style Transfer

Apply artistic styles to an image:

1
const imageBuffer = readFileSync('./input-image.png');
2

3
const { images } = await generateImage({
4
  model: xai.image('grok-imagine-image'),
5
  prompt: {
6
    text: 'Transform this into a watercolor painting style',
7
    images: [imageBuffer],
8
  },
9
  aspectRatio: '1:1',
10
});

Image Provider Options

You can customize the image generation behavior with provider-specific settings via providerOptions.xai:

1
import { xai, type XaiImageModelOptions } from '@ai-sdk/xai';
2
import { generateImage } from 'ai';
3

4
const { images } = await generateImage({
5
  model: xai.image('grok-imagine-image-pro'),
6
  prompt: 'A futuristic cityscape at sunset',
7
  aspectRatio: '16:9',
8
  providerOptions: {
9
    xai: {
10
      resolution: '2k',
11
      quality: 'high',
12
    } satisfies XaiImageModelOptions,
13
  },
14
});

• resolution '1k' | '2k'

  Output resolution. 1k produces ~1024×1024 images, 2k produces ~2048×2048 images (actual dimensions vary based on aspect ratio). Available for grok-imagine-image-pro.

• quality 'low' | 'medium' | 'high'

  Image quality level. Higher quality may increase generation time.

Image Model Capabilities

Video Models

You can create xAI video models using the .video() factory method. For more on video generation with the AI SDK see generateVideo().

This provider supports standard video generation from text prompts or image input, plus explicit video editing, video extension, and reference-to-video (R2V) operations.

Text-to-Video

Generate videos from text prompts:

1
import { xai, type XaiVideoModelOptions } from '@ai-sdk/xai';
2
import { experimental_generateVideo as generateVideo } from 'ai';
3

4
const { video } = await generateVideo({
5
  model: xai.video('grok-imagine-video'),
6
  prompt: 'A chicken flying into the sunset in the style of 90s anime.',
7
  aspectRatio: '16:9',
8
  duration: 5,
9
  providerOptions: {
10
    xai: {
11
      user: 'user-123',
12
      pollTimeoutMs: 600000, // 10 minutes
13
    } satisfies XaiVideoModelOptions,
14
  },
15
});

Generation with Image Input

Generate videos using an image as the starting frame with an optional text prompt. This uses the standard generation path rather than a separate provider mode:

1
import { xai, type XaiVideoModelOptions } from '@ai-sdk/xai';
2
import { experimental_generateVideo as generateVideo } from 'ai';
3

4
const { video } = await generateVideo({
5
  model: xai.video('grok-imagine-video'),
6
  prompt: {
7
    image: 'https://example.com/start-frame.png',
8
    text: 'The cat slowly turns its head and blinks',
9
  },
10
  duration: 5,
11
  providerOptions: {
12
    xai: {
13
      pollTimeoutMs: 600000, // 10 minutes
14
    } satisfies XaiVideoModelOptions,
15
  },
16
});

You can also pass the starting frame via the top-level frameImages option using the first_frame role. This is provider-agnostic and accepts file data (e.g. a Buffer) or a { type: 'url', url } object:

1
import { xai, type XaiVideoModelOptions } from '@ai-sdk/xai';
2
import { experimental_generateVideo as generateVideo } from 'ai';
3
import fs from 'node:fs';
4

5
const { video } = await generateVideo({
6
  model: xai.video('grok-imagine-video'),
7
  prompt: 'The cat slowly turns its head and blinks',
8
  frameImages: [
9
    {
10
      image: fs.readFileSync('./start-frame.png'),
11
      frameType: 'first_frame',
12
    },
13
  ],
14
  duration: 5,
15
  providerOptions: {
16
    xai: {
17
      pollTimeoutMs: 600000, // 10 minutes
18
    } satisfies XaiVideoModelOptions,
19
  },
20
});

Video Editing

Edit an existing video using a text prompt by providing a source video URL via provider options:

1
import { xai, type XaiVideoModelOptions } from '@ai-sdk/xai';
2
import { experimental_generateVideo as generateVideo } from 'ai';
3

4
const { video } = await generateVideo({
5
  model: xai.video('grok-imagine-video'),
6
  prompt: 'Give the person sunglasses and a hat',
7
  providerOptions: {
8
    xai: {
9
      mode: 'edit-video',
10
      videoUrl: 'https://example.com/source-video.mp4',
11
      pollTimeoutMs: 600000, // 10 minutes
12
    } satisfies XaiVideoModelOptions,
13
  },
14
});

Chaining and Concurrent Edits

The xAI-hosted video URL is available in providerMetadata.xai.videoUrl. You can use it to chain sequential edits or branch into concurrent edits using Promise.all:

1
import { xai, type XaiVideoModelOptions } from '@ai-sdk/xai';
2
import { experimental_generateVideo as generateVideo } from 'ai';
3

4
const providerOptions = {
5
  xai: {
6
    mode: 'edit-video',
7
    videoUrl: 'https://example.com/source-video.mp4',
8
    pollTimeoutMs: 600000,
9
  } satisfies XaiVideoModelOptions,
10
};
11

12
// Step 1: Apply an initial edit
13
const step1 = await generateVideo({
14
  model: xai.video('grok-imagine-video'),
15
  prompt: 'Add a party hat to the person',
16
  providerOptions,
17
});
18

19
// Get the xAI-hosted URL from provider metadata
20
const step1VideoUrl = step1.providerMetadata?.xai?.videoUrl as string;
21

22
// Step 2: Apply two more edits concurrently, building on step 1
23
const [withSunglasses, withScarf] = await Promise.all([
24
  generateVideo({
25
    model: xai.video('grok-imagine-video'),
26
    prompt: 'Add sunglasses',
27
    providerOptions: {
28
      xai: {
29
        mode: 'edit-video',
30
        videoUrl: step1VideoUrl,
31
        pollTimeoutMs: 600000,
32
      },
33
    },
34
  }),
35
  generateVideo({
36
    model: xai.video('grok-imagine-video'),
37
    prompt: 'Add a scarf',
38
    providerOptions: {
39
      xai: {
40
        mode: 'edit-video',
41
        videoUrl: step1VideoUrl,
42
        pollTimeoutMs: 600000,
43
      },
44
    },
45
  }),
46
]);

Video Extension

Extend an existing video from its last frame. The duration controls the length of the extension only, not the total output. The output inherits aspectRatio and resolution from the source video.

1
import { xai, type XaiVideoModelOptions } from '@ai-sdk/xai';
2
import { experimental_generateVideo as generateVideo } from 'ai';
3

4
// Step 1: Generate a source video
5
const source = await generateVideo({
6
  model: xai.video('grok-imagine-video'),
7
  prompt: 'A cat sitting on a sunlit windowsill, tail gently swishing.',
8
  duration: 5,
9
  aspectRatio: '16:9',
10
  providerOptions: {
11
    xai: {
12
      pollTimeoutMs: 600000,
13
    } satisfies XaiVideoModelOptions,
14
  },
15
});
16

17
const sourceUrl = source.providerMetadata?.xai?.videoUrl as string;
18

19
// Step 2: Extend the video with a new scene
20
const extended = await generateVideo({
21
  model: xai.video('grok-imagine-video'),
22
  prompt: 'The cat turns its head, notices a butterfly, and leaps off.',
23
  duration: 6,
24
  providerOptions: {
25
    xai: {
26
      mode: 'extend-video',
27
      videoUrl: sourceUrl,
28
      pollTimeoutMs: 600000,
29
    } satisfies XaiVideoModelOptions,
30
  },
31
});

Reference-to-Video (R2V)

Provide reference images to guide the video's style and content. Unlike image-to-video, reference images are not used as the first frame — the model incorporates their visual elements into the generated video. Each reference image can be a public HTTPS URL or a base64 data URI.

1
import { xai, type XaiVideoModelOptions } from '@ai-sdk/xai';
2
import { experimental_generateVideo as generateVideo } from 'ai';
3

4
const { video } = await generateVideo({
5
  model: xai.video('grok-imagine-video'),
6
  prompt:
7
    'The comic cat from <IMAGE_1> and the comic dog from <IMAGE_2> ' +
8
    'are having a playful chase through a sunlit park. ' +
9
    'Cinematic slow-motion, warm afternoon light.',
10
  duration: 8,
11
  aspectRatio: '16:9',
12
  providerOptions: {
13
    xai: {
14
      mode: 'reference-to-video',
15
      referenceImageUrls: [
16
        'https://example.com/comic-cat.png',
17
        'https://example.com/comic-dog.png',
18
      ],
19
      pollTimeoutMs: 600000,
20
    } satisfies XaiVideoModelOptions,
21
  },
22
});

Use <IMAGE_1>, <IMAGE_2>, etc. in your prompt to reference specific images. Up to 7 reference images are supported per request.

Alternatively, use the provider-agnostic top-level inputReferences option. Providing it automatically selects reference-to-video mode, and it accepts file data (e.g. a Buffer) or { type: 'url', url } objects:

1
import { xai, type XaiVideoModelOptions } from '@ai-sdk/xai';
2
import { experimental_generateVideo as generateVideo } from 'ai';
3
import fs from 'node:fs';
4

5
const { video } = await generateVideo({
6
  model: xai.video('grok-imagine-video'),
7
  prompt:
8
    'The comic cat and the comic dog are having a playful chase ' +
9
    'through a sunlit park. Cinematic slow-motion, warm afternoon light.',
10
  inputReferences: [
11
    fs.readFileSync('./comic-cat.png'),
12
    fs.readFileSync('./comic-dog.png'),
13
  ],
14
  duration: 8,
15
  aspectRatio: '16:9',
16
  providerOptions: {
17
    xai: {
18
      pollTimeoutMs: 600000,
19
    } satisfies XaiVideoModelOptions,
20
  },
21
});

inputReferences accepts image references only. A reference with a non-image media type (for example a video or audio clip) is ignored with a warning, and a reference with no media type is treated as an image. If no image reference remains, reference-to-video is not selected and no reference images are sent.

Reference Audio

Reference-to-video can also give the subject a voice. referenceVoiceIds takes up to 3 xAI preset voice ids — you cannot upload your own audio clips. Reference the voices from the prompt with <AUDIO_0>, <AUDIO_1>, and <AUDIO_2>, in the order the voices are passed.

1
import { xai, type XaiVideoModelOptions } from '@ai-sdk/xai';
2
import { experimental_generateVideo as generateVideo } from 'ai';
3

4
const { video } = await generateVideo({
5
  model: xai.video('grok-imagine-video-1.5'),
6
  prompt:
7
    'The person from <IMAGE_0> stands in the room from <IMAGE_1> and speaks ' +
8
    'to the camera with the voice from <AUDIO_0>.',
9
  aspectRatio: '9:16',
10
  duration: 10,
11
  providerOptions: {
12
    xai: {
13
      mode: 'reference-to-video',
14
      referenceImageUrls: [
15
        'https://example.com/person.png',
16
        'https://example.com/room.png',
17
      ],
18
      referenceVoiceIds: ['eve'],
19
      resolution: '720p',
20
      pollTimeoutMs: 600000,
21
    } satisfies XaiVideoModelOptions,
22
  },
23
});

Valid voice ids come from the xAI text-to-speech voice roster. Ids are case-insensitive, and an unknown id returns a 400 response listing the available voices. referenceVoiceIds is ignored with a warning when the resolved operation is not reference-to-video.

Video Provider Options

The following provider options are available via providerOptions.xai. You can validate the provider options using the XaiVideoModelOptions type.

• pollIntervalMs number

  Polling interval in milliseconds for checking task status. Defaults to 5000.

• pollTimeoutMs number

  Maximum wait time in milliseconds for video generation. Defaults to 600000 (10 minutes).

• resolution '480p' | '720p' | '1080p'

  Video resolution. When using the SDK's standard resolution parameter, 1920x1080 maps to 1080p, 1280x720 maps to 720p, and 854x480 maps to 480p. Use this provider option to pass the native format directly. 1080p requires the grok-imagine-video-1.5 model and is available for text-to-video and image-to-video; reference-to-video is capped at 720p (a 1080p request is downgraded with a warning).

• user string

  Optional identifier for the end user responsible for the request. xAI uses this value for abuse monitoring. It is sent unchanged for video generation (including reference-to-video) and video editing requests, and is omitted when not configured. It is not sent for video extension requests. Use an opaque stable identifier and avoid sending unnecessary personal information.

• mode 'edit-video' | 'extend-video' | 'reference-to-video'

  Selects the explicit video operation. Each mode is mutually exclusive:

  • 'edit-video' — edit an existing video (requires videoUrl)
  • 'extend-video' — extend a video from its last frame (requires videoUrl)
  • 'reference-to-video' — generate from reference images (requires referenceImageUrls)

  When omitted, standard generation is used. Legacy inputs are still auto-detected from fields for backward compatibility.

• videoUrl string

  URL of a source video. Used with mode: 'edit-video' for video editing and mode: 'extend-video' for video extension.

• referenceImageUrls string[]

  Array of reference image URLs (1–7 images) or base64 data URIs for reference-to-video (R2V) generation. The model incorporates visual elements from these images without using them as the first frame. Use <IMAGE_1>, <IMAGE_2>, etc. in the prompt to reference specific images. Used with mode: 'reference-to-video'.

• referenceVoiceIds string[]

  Up to 3 xAI preset voice ids that give the subject a voice in reference-to-video (R2V) generation. Preset voices only — audio clips cannot be uploaded. Ids are case-insensitive and come from the text-to-speech voice roster; an unknown id returns a 400 listing the available voices. Use <AUDIO_0>, <AUDIO_1>, and <AUDIO_2> tags in the prompt to reference the voices. Ignored with a warning outside reference-to-video. Reference audio is documented as US-only and limited to trusted partners.

Aspect Ratio and Resolution

For text-to-video, you can specify both aspectRatio and resolution. The default aspect ratio is 16:9 and the default resolution is 480p. The grok-imagine-video-1.5 model additionally supports native 1080p for text-to-video and image-to-video.

For image-to-video, the output defaults to the input image's aspect ratio. If you specify aspectRatio, it will override this and stretch the image to the desired ratio.

For video editing, the output matches the input video's aspect ratio and resolution. Custom duration, aspectRatio, and resolution are not supported — the output resolution is capped at 720p (e.g., a 1080p input will be downsized to 720p).

For video extension, the output inherits aspectRatio and resolution from the source video. duration is supported and controls only the extension length.

For reference-to-video (R2V), you can specify duration, aspectRatio, and resolution. Unlike text-to-video, R2V is capped at 720p — a 1080p request is downgraded to 720p with a warning.

Video Model Capabilities

* Native 1080p applies to text-to-video and image-to-video. Reference-to-video is capped at 720p — a 1080p request is downgraded with a warning.