# AI3D API v2 model: face-rig Face Studio beta — photographic eyes and facial blendshapes Product: https://api.ai3d.dev/v2/face-studio Endpoint: POST https://api.ai3d.dev/v2/face-rig Generic equivalent: POST /v2/jobs with model="face-rig". Authentication: customer API key, browser account session, or scoped mobile token. Fit 27 experimental blendshapes to an owned, completed TRELLIS 2 portrait mesh. Automatically finds the front, fits CaptureForge controls, cuts the mouth seam, reconstructs morphing lids and projects the original eye photographs onto one continuous surface per eye. No added teeth. This is a morph rig, not a skeleton, and is not a complete ARKit 52 rig. Output quality depends on generated topology. SUPPORTED INPUT - source_job_id: the UUID of a succeeded trellis-2 job owned by the caller. - That job must retain its archived input portrait and an embedded-resource GLB. - One textured head mesh, one UV map/material, at most 1,000,000 source vertices. - Frontal, upright head-and-shoulders portrait, neutral expression, eyes open, lips gently closed, no glasses or face coverings. PNG/JPEG/WebP up to 10 MiB. - Source mesh up to 100 MiB. Arbitrary external GLBs and multipart rigs are not accepted by this beta. Unsupported inputs fail with a useful job error. REQUEST POST /v2/face-rig Authorization: Bearer Content-Type: application/json Idempotency-Key: {"source_job_id":"","publish":false} Optional: publish (false), slug ("world"), viewer_style ("default" or "illustration"). Keep the same Idempotency-Key and body if a POST response is interrupted. A 202 receipt is acceptance, not a completed rig. Poll GET /v2/jobs/{id}; a running receipt may include progress.stage. Download face.glb and rig.json from its artifacts after status=succeeded. Existing job ownership and read/write scopes, quotas, publication, and authenticated artifact routes apply. PRICING / PRIVACY Rigging is free during the initial limited beta: 1/minute and 3/day per account (or operator tenant), shared across keys. Failed fits consume the beta attempt. Portrait and mesh generation keep their current separate prices; check GET /v2/billing/prices before submitting those stages. No guessed compute price. Rigs are private by default. Set publish=true for a permanent public result link; the final GLB includes its eye textures. The source portrait is not separately published by the rigging stage. TEXT OR IMAGE TO RIGGED FACE 1. For text, create a sana-sprint portrait. Choose a frontal neutral composition. For a photo, upload it through POST /v2/uploads. 2. Create a trellis-2 image job from that portrait and wait for success. 3. POST /v2/face-rig with its source_job_id and poll the normal job endpoint. Face Studio automates these steps with resumable, idempotent requests. POST /v2/face-studio/jobs is a private-by-default alias for the portrait, mesh and face-rig job models; all standard job parameters and billing remain in effect. OUTPUT / REVIEW face.glb: head, approximate mouth cavity, original-photo eyes and 27 controls. rig.json: control names, source/output hashes, blink/projection checks and limits. processing.log and third-party-notices.txt: bounded processing and license notes. Open Face controls in the viewer to test each shape. Review extreme combinations and side views before production animation. Eye texture outside the photographed opening is extrapolated; unseen mouth anatomy cannot be recovered from one image. TECHNOLOGY / LICENSES CaptureForge (GPL-3.0-or-later), pinned source: https://github.com/H0wZy/captureforge/tree/1f2183b47d2306c6ec370fb1ab925fd938055d0e Blender / official bpy 4.5.3 (GPL-3.0-or-later), MediaPipe Face Landmarker (Apache-2.0). The fixed pipeline runs in a separate bounded execution container, without the API's application secrets. No customer Python scripts are accepted.