Request: High-Quality, Model-Selectable Video, TTS, and Talking Avatar APIs

Developers
1

Title

Request: High-Quality, Model-Selectable Video, TTS, and Talking Avatar APIs

Post

We are building Ad Studio as an Anna App. Text generation, image generation, storage, and job orchestration can already use Anna, but media generation still depends on external services:

  • Image editing: fal GPT Image 2
  • Video generation: fal Seedance 2.0 / Gemini Omni
  • TTS: ElevenLabs Multilingual v2
  • Video composition: FFmpeg

Anna already provides image.generate, image.edit, and tools.invokeAsync, so we are mainly requesting the following capabilities.

1. Video generation

A unified video.generate API supporting:

  • Reference images and audio
  • Native audio and lip-sync
  • 9:16, 16:9, and 1:1 aspect ratios
  • 720p and 1080p
  • Async status, cancellation, and recovery

The quality should be at least comparable to:

  • bytedance/seedance-2.0/reference-to-video
  • google/gemini-omni-flash/image-to-video

2. Text-to-speech

Please provide audio.speech and audio.listVoices.

Quality should be comparable to ElevenLabs eleven_multilingual_v2, with Chinese and English support, stable voice IDs, MP3/WAV output, speaking-rate controls, and audio-duration metadata.

3. Talking avatars and lip-sync

We would like Anna to support models comparable to:

  • fal-ai/heygen/avatar4/image-to-video
  • fal-ai/bytedance/omnihuman/v1.5
  • fal-ai/sync-lipsync/v3/image-to-video
  • For existing-video lip-sync: fal-ai/kling-video/lipsync/audio-to-video

Basic mouth-only models such as Wav2Lip or MuseTalk should not be the default implementation, as they often lack natural expressions, body motion, and reliable identity consistency.

4. Strict model selection

This is the most important requirement. Please support something like:

{
  "modelPreferences": {
    "hints": [{"name": "HeyGen Avatar 4"}],
    "strict": true
  },
  "fallbackPolicy": "error"
}

If the requested model is unavailable, the API should return an explicit error instead of silently falling back to a lower-quality model.

Responses should also report the actual provider, model, usage, and cost.

Our goal is to use Anna’s unified quota, credentials, storage, and auditing while still achieving media quality suitable for commercial advertising.