Who receives production content, for what, and how long they keep it, as configured now. This is the register the Privacy Notice points to. Where a provider's route-specific terms are not yet established on our side, the row says so.
| Provider | What it receives | Purpose | Location | Terms and retention |
|---|---|---|---|---|
| fal.ai (fal Features and Labels, Inc.) | approved synthetic portraits, generated voice audio, generated video; the prompts and parameters of each request | runs the generation models (speech, performance, face conformance, lip-sync, portrait and mask models) and stores inputs and outputs for the pipeline to fetch | United States (fal's cloud); per-model execution may route to the model's provider through fal | fal storage/retention, payload deletion and file access-control docs; X-Fal-Store-IO and CDN lifecycle to be set per route |
| MiniMax (speech-2.8-hd and H3 Max, through fal) | the script's words for each line and the chosen catalog voice id; for H3 Max, portrait and audio | text-to-speech; alternate performance renderer | through fal | API privacy and terms located; application to the H3 Max route via fal not established |
| ElevenLabs (through fal) | the script's words and a designed catalog voice id (voice cloning is not registered) | text-to-speech for voices cast from its catalog | through fal | privacy policy (training opt-out, prospective); Zero Retention Mode is enterprise-only and per request; DPA |
| VEED (Fabric 1.0, through fal) | approved portrait and the line's audio | primary performance renderer | through fal | general privacy policy (purpose-based retention); not a Fabric-through-fal contract |
See also Retention. Face embeddings are computed on our own machines and never sent to a provider.