AI Avatar Video Generator API — Duix-Avatar & InfiniteTalk

Building an AI avatar video generator? KeyoAPI hosts Duix-Avatar (audio+video → talking face) and InfiniteTalk (image/audio → talking head) as async jobs — generate speech with IndexTTS-2 / CosyVoice3 / Qwen3-TTS on the same key, then render.

Digital human models

Indicative figures for planning. Confirm live Keyo sell rates on interactive Model Square or the static pricing list.

Productize the loop, not a single model

Typical product loop: draft script → TTS (optionally cloned) → avatar render → CDN cache. Cap max duration, show queue ETA, and moderate deepfake abuse. Try/buy each model from its /pricing/{id} page after /sign-up — no separate upload playground.

Rates and pipeline

This is the API path, not a consumer app: prepaid metering, OpenAI-style auth for adjacent chat and speech, async task polling for renders.

Indicative rates: Duix-Avatar ~$0.0068 per second of output; InfiniteTalk ~$0.342 per request. Confirm /pricing/Duix-Avatar and /pricing/InfiniteTalk before volume. Pick Duix when you have driving video; pick InfiniteTalk when you start from a still spokesperson photo.

How to call: async video endpoints with model=Duix-Avatar or InfiniteTalk, then GET /v1/task/{id}. Speech side: /tts-api and /voice-cloning-api. Model guides under /model/Duix-Avatar and /model/InfiniteTalk.

Likeness rights and abuse prevention are your product’s responsibility — rate-limit, log, and require consent flags before queueing.

curl https://www.keyoapi.xyz/v1/async/videos/image-to-video \
  -H "Authorization: Bearer $KEYO_API_KEY" \
  -F model=InfiniteTalk \
  -F image=@./portrait.png \
  -F audio=@./voice.wav
# then: GET /v1/task/{id}

FAQ

Also see /compare, /free-models, and model guides under /model/.