This feature request proposes a verified “Synthetic Subject ID” system for OpenAI’s future video-generation capabilities.
It would let creators generate consistent digital actors across scenes without referencing real individuals, while remaining policy-compliant.
Context & Problem
The video-generation model Sora 2 currently prohibits uploading photorealistic human faces in reference images, even when those faces are fully AI-generated and non-real.
This restriction significantly limits creators who wish to maintain a consistent subject identity (actor, model) across multiple scenes (e.g., in brand ads, episodic narratives, product showcases).
Proposed Solution
Introduce an internal “Synthetic Subject ID” system within Sora:
-
Creators generate or select a synthetic human character within the platform (or upload a verified synthetic face) that is non-real and non-identifiable.
-
Each synthetic subject is assigned a unique identifier (e.g.,
@subject_001A). -
Reference images (front, 3/4, outfit, full-body) are stored as the subject’s canonical set.
-
In subsequent scene prompts, the creator uses the ID:
“Use subject
@subject_001A(same actor) walking through the city…” -
The model recognizes this ID, retains visual consistency (facial structure, outfit palette, body proportions) across scenes, yet remains policy-compliant because the subject is synthetic.
Benefits
-
For creators: Enables seamless multi-scene continuity of subject and product without re-uploading or specifying new references each time.
-
For OpenAI / Sora: Maintains safety by ensuring no real person likeness is used; provides traceability and compliance.
-
For production workflows: Simplifies storytelling, advertising and episodic content creation using a stable “actor” asset.
Implementation Outline-
Phase 1: In-app feature “Create Synthetic Subject” — generate or upload verified non-real human.
-
Phase 2: Assign persistent ID, store subject images + metadata.
-
Phase 3: Enable prompts referencing that ID; model conditions on the asset for consistency.
-
Phase 4: Open API for users to reference subject IDs in multi-scene pipelines.
Policy & Safety Considerations
All synthetic subjects are generated natively within the Sora interface or API, using OpenAI-controlled models.
-
Because creation happens in-app, Sora can guarantee provenance, confirm the subject is synthetic and non-real, and automatically mark it as safe for reuse across projects.
-
Each subject is internally tagged with metadata such as creation timestamp, generation parameters, and a unique Synthetic Subject ID.
-
Prompts referencing these IDs are automatically permitted, since the subject’s safety status is already verified by the system.
-
This eliminates the need for external uploads or manual verification pipelines — everything remains within OpenAI’s trusted ecosystem.
I ask OpenAI to evaluate adding a “Verified Synthetic Subject” feature into Sora’s upcoming roadmap. This capability would unlock a new class of consistent, high-quality visual narratives for creators, while aligning with OpenAI’s safety and ethical standards.
-
-