WaveSpeedAI Infinitetalk Talking Avatar Video Generation

WaveSpeedAI InfiniteTalk

Generate talking head videos from a portrait image and audio using WaveSpeed AI's InfiniteTalk model. Produces lip-synced video up to 10 minutes long with natural facial animations.

Authentication

export WAVESPEED_API_KEY="your-api-key"

Get your API key at wavespeed.ai/accesskey.

Quick Start

import wavespeed from 'wavespeed';
// Upload local image and audio files
const imageUrl = await wavespeed.upload("/path/to/portrait.png");
const audioUrl = await wavespeed.upload("/path/to/speech.mp3");
const output_url = (await wavespeed.run(
  "wavespeed-ai/infinitetalk",
  {
    image: imageUrl,
    audio: audioUrl
  }
))["outputs"][0];

You can also pass existing URLs directly:

const output_url = (await wavespeed.run(
  "wavespeed-ai/infinitetalk",
  {
    image: "https://example.com/portrait.jpg",
    audio: "https://example.com/speech.mp3"
  }
))["outputs"][0];

API Endpoint

Model ID: wavespeed-ai/infinitetalk

Animate a portrait image with lip-synced audio to produce a talking head video.

Parameters

| Parameter | Type | Required | Default | Description | |-----------|------|----------|---------|-------------| | image | string | Yes | -- | URL of the portrait image to animate | | audio | string | Yes | -- | URL of the audio to drive the animation | | mask_image | string | No | -- | URL of a mask image to specify which person to animate. Warning: The mask should only cover the regions to animate — do not upload the full image as mask_image, or the result may render as fully black. | | prompt | string | No | -- | Text prompt for additional guidance. Keep it short; English recommended to avoid noisy results. | | resolution | string | No | 480p | Output resolution. One of: 480p, 720p | | seed | integer | No | -1 | Random seed (-1 for random). Range: -1 to 2147483647 |

Example

import wavespeed from 'wavespeed';
const imageUrl = await wavespeed.upload("/path/to/portrait.png");
const audioUrl = await wavespeed.upload("/path/to/speech.mp3");
const output_url = (await wavespeed.run(
  "wavespeed-ai/infinitetalk",
  {
    image: imageUrl,
    audio: audioUrl,
    resolution: "720p",
    seed: 42
  }
))["outputs"][0];

Using a Mask Image

When multiple people are in the image, use a mask to specify which face to animate:

const imageUrl = await wavespeed.upload("/path/to/group-photo.png");
const audioUrl = await wavespeed.upload("/path/to/speech.mp3");
const maskUrl = await wavespeed.upload("/path/to/mask.png");
const output_url = (await wavespeed.run(
  "wavespeed-ai/infinitetalk",
  {
    image: imageUrl,
    audio: audioUrl,
    mask_image: maskUrl,
    resolution: "720p"
  }
))["outputs"][0];

> Important: The mask should only highlight the face region to animate. Using the full image as a mask will produce a fully black output.

With Prompt Guidance

const output_url = (await wavespeed.run(
  "wavespeed-ai/infinitetalk",
  {
    image: imageUrl,
    audio: audioUrl,
    prompt: "natural head movements, subtle expressions"
  }
))["outputs"][0];

Advanced Usage

Custom Client with Retry Configuration

import { Client } from 'wavespeed';
const client = new Client("your-api-key", {
  maxRetries: 2,
  maxConnectionRetries: 5,
  retryInterval: 1.0,
});
const imageUrl = await client.upload("/path/to/portrait.png");
const audioUrl = await client.upload("/path/to/speech.mp3");
const output_url = (await client.run(
  "wavespeed-ai/infinitetalk",
  {
    image: imageUrl,
    audio: audioUrl,
    resolution: "720p"
  }
))["outputs"][0];

Error Handling with runNoThrow

import { Client, WavespeedTimeoutException, WavespeedPredictionException } from 'wavespeed';
const client = new Client();
const result = await client.runNoThrow(
  "wavespeed-ai/infinitetalk",
  {
    image: imageUrl,
    audio: audioUrl
  }
);
if (result.outputs) {
  console.log("Video URL:", result.outputs[0]);
  console.log("Task ID:", result.detail.taskId);
} else {
  console.log("Failed:", result.detail.error.message);
  if (result.detail.error instanceof WavespeedTimeoutException) {
    console.log("Request timed out - try increasing timeout");
  } else if (result.detail.error instanceof WavespeedPredictionException) {
    console.log("Prediction failed");
  }
}

Resolution and Pricing

| Resolution | Cost per 5 seconds | Rate per second | Max length | |------------|--------------------|-----------------| -----------| | 480p | $0.15 | $0.03/s | 10 minutes | | 720p | $0.30 | $0.06/s | 10 minutes |

Minimum charge is 5 seconds. Video length is determined by the audio duration (up to 10 minutes).

Tips

Security Constraints

import wavespeed from 'wavespeed';
const imageUrl = await wavespeed.upload("/path/to/portrait.png");
const audioUrl = await wavespeed.upload("/path/to/speech.mp3");
const output_url = (await wavespeed.run(
  "wavespeed-ai/infinitetalk",
  {
    image: imageUrl,
    audio: audioUrl,
    resolution: "720p",
    seed: 42
  }
))["outputs"][0];

Using a Mask Image

When multiple people are in the image, use a mask to specify which face to animate:

const imageUrl = await wavespeed.upload("/path/to/group-photo.png");
const audioUrl = await wavespeed.upload("/path/to/speech.mp3");
const maskUrl = await wavespeed.upload("/path/to/mask.png");
const output_url = (await wavespeed.run(
  "wavespeed-ai/infinitetalk",
  {
    image: imageUrl,
    audio: audioUrl,
    mask_image: maskUrl,
    resolution: "720p"
  }
))["outputs"][0];

> Important: The mask should only highlight the face region to animate. Using the full image as a mask will produce a fully black output.

With Prompt Guidance

const output_url = (await wavespeed.run(
  "wavespeed-ai/infinitetalk",
  {
    image: imageUrl,
    audio: audioUrl,
    prompt: "natural head movements, subtle expressions"
  }
))["outputs"][0];