InfiniteTalk

Audio-Driven VideoTalking AvatarOpen Source

InfiniteTalk is MeiGen-AI’s audio-driven human video model for synchronizing speech with facial expression, head movement, and body posture—not only the mouth. APIXO provides a focused portrait-and-audio workflow with an optional mask, up to ten minutes of driving audio, 480p or 720p output, and asynchronous delivery.

APIXO Inputs

Portrait + audio

APIXO Price

From $0.03 / second480p, five-second minimum

APIXO Resolution

480p / 720p

Audio Limit

Up to 600 seconds

Released

Aug 19, 2025

Loading workspace...

Checking API access...

Integrate with AI

Copy a prompt for your AI assistant to install APIXO MCP, read the live schema, and wire this model into your app.

Endpoints

Create task

POSThttps://api.apixo.ai/api/v1/generateTask/infinitetalk

Poll status

GEThttps://api.apixo.ai/api/v1/statusTask/infinitetalk?taskId={taskId}

Request body

request_type

async

Use callback with callback_url for webhooks.

input

object

Mode, prompt, references, size, duration, and audio settings.

request_typerequired

Typeselect

Use async to poll statusTask, or callback to receive a webhook when the result is ready.async · callback

callback_urlconditional

Typetext

Public HTTPS URL for callback delivery.Only in request_type=callback

moderequired

Typeselect

Choose the video workflow supported by this model.image-to-video

promptoptional

Typestring

Optional guidance for expression, motion, or scene behavior. Remove it or provide a non-empty instruction.

resolutionrequired

Typeselect

Choose the output resolution tier.480p · 720p

seedoptional

Typetext

Use -1 for random output, or 0-2147483647 for reproducible generation.

Parameter	Type	Required	Description
`request_type`	select	required	Use async to poll statusTask, or callback to receive a webhook when the result is ready.async · callback
`callback_url`	text	conditional	Public HTTPS URL for callback delivery.Only in request_type=callback
`mode`	select	required	Choose the video workflow supported by this model.image-to-video
`prompt`	string	optional	Optional guidance for expression, motion, or scene behavior. Remove it or provide a non-empty instruction.
`resolution`	select	required	Choose the output resolution tier.480p · 720p
`seed`	text	optional	Use -1 for random output, or 0-2147483647 for reproducible generation.

File parameters (URL arrays)

image_urls[mode=image-to-video]Upload one source portrait plus an optional mask image.Only in mode=image-to-video

audio_urls[mode=image-to-video]Upload audio references for multimodal generation when supported.Only in mode=image-to-video

Example (cURL)

curl -X POST "https://api.apixo.ai/api/v1/generateTask/infinitetalk" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "request_type": "async",
    "input": {
      "mode": "image-to-video",
      "image_urls": [
        "https://example.com/portrait.png"
      ],
      "audio_urls": [
        "https://example.com/voice.mp3"
      ],
      "prompt": "make the portrait speak naturally with subtle head motion",
      "resolution": "480p",
      "seed": -1
    }
  }'

Full API documentation

Capabilities

Audio drives the wider human performance

Speech-Synchronized Motion

InfiniteTalk uses the driving audio to coordinate mouth movement with expression, head motion, and body posture. Lip synchronization is a core objective, although accuracy can still vary with speech speed, pose, and source quality.

Long-Sequence Continuity

InfiniteTalk’s upstream framework uses temporal context between generated segments to support longer audio-driven sequences. APIXO accepts driving audio up to 600 seconds, although long outputs should still be reviewed for identity drift and transition consistency.

Portrait-to-Performance

APIXO’s implementation starts from one portrait image rather than a text-only character description. The source establishes the subject’s visible identity, clothing, composition, and background before audio-driven movement is synthesized.

Spoken and Sung Performance

Drive the portrait with finished speech, dialogue, narration, or singing audio. Vocal pacing, emotion, pronunciation, and delivery come from the supplied recording rather than a text-to-speech system inside this endpoint.

Long Audio Input

APIXO accepts one audio file lasting up to 600 seconds. This makes the route suitable for material longer than a conventional short avatar clip, while processing time and consistency demands increase with length.

Reproducible Iteration

Choose 480p or 720p output and optionally supply an APIXO seed. A fixed seed supports controlled experimentation, but it does not guarantee identical results across changing inputs or provider revisions.

API Specifications

InfiniteTalk controls through APIXO

Public limits for the dedicated portrait-and-audio workflow.

1 portrait + optional mask

Image Inputs

1 audio URL

Audio Count

MP3 / WAV / M4A

Audio Formats

128 MB

Audio Size Limit

5 seconds

Minimum Billing

Async / Callback

Delivery

Production Uses

Create spoken content from prepared audio

Digital Presenters

Presenter and Spokesperson Videos

Pair a prepared portrait with recorded or synthesized speech to create presenter-style content. The endpoint animates the supplied performance audio; it does not write a script or generate speech from text.

Localization

Localized Marketing Versions

Reuse an approved portrait with separately produced voice tracks for regional campaign versions. InfiniteTalk responds to the acoustic performance rather than a declared language, but pronunciation and synchronization should be reviewed for every recording.

Education

Lessons and Training Modules

Convert narration into instructor, guide, or course-avatar videos at 480p or 720p. APIXO’s ten-minute audio limit supports longer explanations, while complex modules can still be divided into shorter segments for easier quality control.

Entertainment

Character Speech and Singing

Animate a suitable character portrait from dialogue, narration, or singing audio. Strongly stylized, non-human, full-body, or multi-person sources are not documented APIXO guarantees and should be tested before committing to a production format.

Production Workflow

Prepare the portrait, audio, and delivery path

Clean source assets reduce avoidable synchronization and identity problems.

Prepare a Clear Portrait

Choose one visible subject with a clear face, limited occlusion, and sufficient image quality. Keep the intended framing and background in the source because APIXO does not expose aspect-ratio or camera controls.

Upload Finished Speech Audio

Provide one public MP3, WAV, or M4A URL within the 128 MB and 600-second limits. Complete transcription, translation, voice generation, cleanup, and timing before submitting the track.

Generate and Review Motion

Select 480p or 720p, submit the asynchronous task, and retrieve the result through polling or callback delivery. Review the mouth, teeth, eyes, hands, pose, identity, and segment transitions before publishing.

Integration Notes

Understand what APIXO exposes

Important constraints

APIXO exposes one image-to-video mode requiring a portrait and a driving audio file.

APIXO’s published request schema marks the prompt as required; submit a clear motion instruction with every production request.

Billing uses the detected audio length or five seconds, whichever is greater.

Longer audio increases processing time; callback delivery is preferable for long production tasks.

APIXO does not publicly expose video input, aspect ratio, frame rate, or output-format controls.

Use portraits and voices only when you have the necessary rights, consent, and authorization.

InfiniteTalk questions

This page covers the original InfiniteTalk research release from MeiGen-AI, published with its technical report, code, and model weights on August 19, 2025. It does not describe LongCat Video Avatar, community checkpoints, or unrelated avatar systems.

Not through the publicly documented infinitetalk schema. The upstream research supports audio-driven video-to-video dubbing, but APIXO exposes a portrait-and-audio image-to-video workflow without a source-video field.

No. APIXO requires an uploaded audio file, and that recording drives the resulting performance. Use a separate recording or text-to-speech system to create the spoken or sung audio before calling InfiniteTalk.

The uploaded audio drives the performance, and APIXO probes its duration for billing, up to 600 seconds. There is no separate duration control. Allow for possible processing or encoding differences when validating the final video length.

Occluded faces, extreme poses, rapid speech, unclear audio, visible hands near the face, multiple people, and low-quality portraits can increase synchronization errors or visual drift. Long outputs should be reviewed for identity and temporal consistency throughout.