AI compliance for the post-production stack

Don’t get sued
for your AI content.

FaceStar AI scans every frame for celebrity likeness, IP infringement, and costume resemblance — before your distributor’s lawyer asks for paperwork.

Multi-jurisdiction legal reports (US · EU · China · Japan) · Bilingual EN/中文 · GPU-scanned full-length features in minutes.

No credit card required · Cancel anytime · SOC 2 in progress

Built for the people who get the cease-and-desist letter first

Short-Drama Platforms

Pre-distribution scan for vertical drama apps

Studios & VFX Houses

Audit AI shots before delivery

E&O Insurance Carriers

Risk reports backing entertainment policies

Entertainment Law Firms

Initial discovery for likeness claims

Three risks, one scan

Every AI-generated frame is assessed against three legal exposure vectors at once, with a per-match risk grade you can hand to counsel as-is.

Right of publicity

Celebrity Likeness

Face match against 32,000+ public figures using ArcFace 512-D embeddings + ethnicity-aware vector search. Calibrated multi-frame scoring catches resemblance even at low confidence.*

*Coverage based on current database; advisory and not legal advice.

IP infringement

Anime & Game Characters

Cross-domain CLIP similarity flags AI-generated characters that resemble protected anime, manga, or game IP — even when stylized differently from any single reference image.

IP holders sourced from major anime/game publishers; expanding quarterly.

Trade dress

Iconic Costume

OpenCLIP weighted-region matching against signature movie costumes, weapons, and silhouettes — the boundary between homage and infringement, scored numerically.

Includes franchise-recognizable looks; full list on request.

Coverage

Every frame. No exceptions.

We don’t sample. We don’t skip. A coarse pass covers the full feature, then the pipeline re-scans every flagged segment at full rate — because the frame your distributor’s lawyer screenshots isn’t the one you skipped.

172,800

Equivalent frames per feature*

24fps × 2 hours, full coverage

5 fps

Fine-pass rate on flagged segments

No sampling once risk is detected

0

Segments skipped per scan

Two-pass pipeline always re-scans

*FaceStar pipeline runs a 1fps coarse pass over the full feature plus an automatic 5fps deep-pass on any segment flagged for likeness, IP, or costume similarity. The 172,800 figure is equivalent-frame coverage at 24fps source rate; coarse-pass sampling rate adapts to input frame rate.

Infrastructure

Built on NVIDIA H200 SXM.

The same GPU that trains frontier models runs your compliance scan. A full-length feature finishes in minutes, not hours. Slate-volume scanning gets a dedicated pool.

141 GB

HBM3e per GPU

1.7× the memory of H100*

4.8 TB/s

Memory bandwidth

1.4× H100 throughput*

3,958

TFLOPS FP8

Per-GPU peak performance*

900 GB/s

NVLink interconnect

Multi-GPU scale-out fabric*

*NVIDIA H200 SXM5 published specifications. Comparisons are vs the prior-generation H100 SXM5. Production scans run on a shared GPU pool by default; Enterprise plans get dedicated reservation with SLA on queue time.

Security

Your unreleased footage stays yours.

Compliance scanning of pre-distribution material is only useful if the material doesn’t leak. Four contractual guarantees, in plain language.

AES-256 at rest, TLS 1.3 in transit

Industry-standard encryption from the moment a presigned upload starts to the moment the scan record is purged. Keys rotated on a published schedule.

12-hour zero-retention

Source video files purge automatically within 12 hours of scan completion. The PDF report and detection metadata stay; the raw footage doesn't.

No model training on your content

Your footage is never used to train, fine-tune, or improve any FaceStar detection model. Detection databases are built from licensed public-figure imagery only.

Isolated container per scan

Each job runs in a fresh worker container with no shared memory or filesystem between tenants. Container lifecycle ends with the scan.

DPA, SOC 2 progress letter, and an architecture review available on request from sales@facestar.ai.

Scanning a slate of AI films?

Studios, short-drama platforms, and E&O carriers run dozens of scans a week. We tier pricing by volume and offer dedicated jurisdictions, on-prem deployment, and SLAs on Enterprise.

From upload to court-ready PDF

Three steps. Same pipeline studios run on full-length features.

1

Upload your cut

MP4, MOV, JPG, or PNG — drag and drop or call our API. Files are encrypted at rest and deleted after 12 hours.

2

Every frame analyzed

Two-pass GPU scan (1 fps coarse → 5 fps on flagged segments) + cross-frame tracking dedupes the same face across scenes.

3

Get a graded PDF

Bilingual report with per-match risk grade, frame stamps, and cited statutes — a paper trail your distributor’s legal team will accept.

Frequently asked

What FaceStar AI detects, where the legal coverage applies, and what the reports actually represent.

What does FaceStar AI detect?
Three tracks in parallel: (A) celebrity face matches against a database of 32,000+ public figures using ArcFace embeddings, (B-1) anime and game IP characters via CLIP cross-domain similarity, and (B-2) costume and styling resemblance via OpenCLIP weighted-region matching.
Which jurisdictions do the compliance reports cover?
US right of publicity, EU GDPR personality rights, China Civil Code Article 1018-1019, and Japan personality rights case law. Reports cite the controlling statute or precedent and assess risk per detected match.
Are the reports legal advice?
No. They are advisory analysis intended to support legal review by qualified counsel. Risk levels and citations are computed from the detection results; binding opinions require a licensed attorney.
Can I scan video, not just images?
Yes. Videos are processed with a two-pass pipeline (1 fps coarse scan plus 5 fps fine scan on flagged segments) and SORT-style cross-frame tracking to deduplicate the same person across scenes.
Is the platform bilingual?
Yes — generated PDF reports are available in English, Simplified Chinese, or both side-by-side, and the platform UI supports the same languages.

Get updates from FaceStar AI

Product news, new detection tracks, and case studies from short-drama platforms and studios. No spam — unsubscribe anytime.