Now in Claude: generate images & videos from chat
Connect

InfiniteTalk API Guide: Talking Avatar Videos from Audio

Use the InfiniteTalk API to turn a character image and uploaded audio into a talking-avatar video up to 10 minutes, with duration-based credit preview and polling.

Guide4 min read

InfiniteTalk is an audio-driven talking-avatar model. Instead of asking a video model to invent speech, you supply the exact character image and audio track. The audio controls the timing while the model animates the portrait into a speaking video.

The MakeInfluencer integration supports audio-driven videos up to 600 seconds, making it useful for explainers, sales messages, training clips, localized content, and AI influencer monologues.

Required Inputs

FieldRequirement
startFrameUrlRequired reference image for the speaker
audioUrlRequired uploaded or hosted audio file
durationSecondsOptional client estimate; server measurement controls credits
resolutionDefaults to 480p in the current public workflow

The live InfiniteTalk API page publishes the exact field contract and a duration-based sample quote.

If the image or audio is not already hosted, follow the POST /api/v1/assets/presign asset guide, upload the raw file to the returned URL, and use its assetUrl below.

Preview a 30-Second Talking Video

curl --request POST \
  https://www.makeinfluencer.ai/api/v1/models/infinitetalk/preview-credits \
  --header "Authorization: Bearer $MAKEINFLUENCER_API_KEY" \
  --header "Content-Type: application/json" \
  --data '{
    "input": {
      "startFrameUrl": "https://example.com/approved-spokesperson.png",
      "audioUrl": "https://example.com/voiceover.mp3",
      "resolution": "480p",
      "durationSeconds": 30
    }
  }'

Credit cost scales with the uploaded audio's measured duration. The preview endpoint downloads the owned audio asset and measures it server-side before returning the quote, so changing durationSeconds cannot reduce the final charge.

Source Image Checklist

The image should be prepared for speech animation:

  • One clearly visible face
  • Front-facing or modest three-quarter angle
  • Mouth unobstructed
  • Enough resolution for facial detail
  • Stable lighting across the face
  • No hands, microphones, or accessories covering the lips
  • Framing with room for subtle head and shoulder movement

A dramatic profile, tiny face, or heavily occluded mouth gives the model less information to animate.

Audio Checklist

  • Remove long silence at the beginning and end
  • Use clean speech with limited background noise
  • Upload the final audio before requesting a credit preview
  • Normalize volume before submission
  • Split very long scripts into logical chapters when separate review is useful
  • Verify commercial rights for the voice and script

Audio Duration Is Verified Server-Side

The client duration is only an estimate for forms. MakeInfluencer measures the uploaded audio for preview and generation billing, then charges from that verified duration.

Product Workflow

1

Approve the Character

Choose a consistent portrait with a clear face and brand-safe styling.

2

Create the Audio

Record or generate the final approved voice track and measure its duration.

3

Preview Cost

Send image URL, audio URL, resolution, and measured duration to preview-credits.

4

Generate

Queue the same payload with an idempotency key and store the returned video ID.

5

Review

Poll the owned job, then check lip timing, face stability, framing, and audio quality.

InfiniteTalk vs Generated-Audio Video

Use InfiniteTalk when the words and voice are fixed. Use a model with generated audio, such as Seedance 2 Fast, when ambient sound or model-created audio is acceptable and exact dialogue control is not the central requirement.

NeedBetter starting point
Exact approved script and voice trackInfiniteTalk
Long talking-avatar outputInfiniteTalk
General cinematic scene with ambient audioSeedance 2 Fast or another video model
Motion copied from a reference clipKling Motion Control

Money-Safe Long-Form Strategy

For a long video, generate a short representative segment before committing the full duration. Test the character image, voice recording style, and framing on 10–20 seconds. Once the result is acceptable, queue the complete audio.

That small approval gate prevents a user from spending several minutes of credits on an image that does not animate well.

API FAQ

How long can an InfiniteTalk video be?

The current public model contract accepts 1 to 600 seconds.

How is it priced?

Credits are calculated from server-measured audio duration. Upload the final audio, then call preview-credits; the server verifies the media before returning the quote.

Does InfiniteTalk create the voice?

No. You provide the audio URL. This gives the workflow exact control over words, pacing, language, and voice rights.

How do I get the result?

Creation returns HTTP 202 and a video generation ID. Poll /api/v1/generations/video/{generationId} until terminal is true. On success, ready is true and the response includes the video URL.

Ready to try it yourself?

Start creating AI influencers and generating content in minutes.