CineClear AI- E&O Legal Clearance & Remediation Dispatcher

Autonomous pre-distribution film & television clearance powered by Gemini Multimodal Vision, real-time Parallel Search grounding, and multi-agent production remediation.

Comment

🎬 Inspiration

In the entertainment industry, no feature film, television episode, or commercial can be licensed or distributed to platforms like Netflix, Apple TV+, or Disney+ without Errors & Omissions (E&O) insurance.

Securing this coverage requires clearance coordinators to spend weeks manually auditing thousands of background set props, fine art pieces, corporate logos, architectural facades, and screenplay dialogue. A single un-cleared background painting (Ringgold v. Black Entertainment Television) or active telephone number can trigger federal copyright claims (17 U.S.C. § 106), trademark dilution (15 U.S.C. § 1125), privacy torts, or emergency $40,000+ visual effects paint-outs after the physical set is struck.

We built CineClear AI to compress this multi-week manual bottleneck into an autonomous, 30-second legal audit and remediation dispatch engine that bridges legal counsel, visual effects, and post-production editorial.

⚡ What It Does

CineClear AI operates as an autonomous 3-Agent Triad running on Google Cloud and backed by a deterministic statutory invariant harness:

  1. Agent 1: Forensic Multimodal Extractor
    Ingests screenplay text, production stills, and video dailies via OpenCV temporal keyframe sampling. Employs the Google GenAI SDK (gemini-2.5-flash) to parse scene liabilities and calculate normalized 2D spatial bounding boxes ([ymin, xmin, ymax, xmax]).
  2. Agent 2: Senior Counsel Critic (Parallel Search Grounded)
    Executes real-time semantic queries against live USPTO trademark registries, US Copyright catalogs, and NANPA safe harbors via the Parallel Search API (https://api.parallel.ai/v1). It computes an objective 4-Factor Fair Use Scorecard (17 U.S.C. § 107), runs the 2-prong Rogers v. Grimaldi artistic relevance test, and evaluates multi-territorial exposure across the US, UK, EU, and Canada.
  3. Agent 3: Autonomous Remediation & Production Dispatcher
    Routes verified liabilities into execution-ready departmental deliverables:
    • Legal Counsel: Pre-populated Form-4A Artwork Release Agreements and Trademark Placement Contracts with mandatory State Bar UPL decision-support disclaimers.
    • Post-Production / VFX: Timecoded 2D Greeking & Logo Obscure Work Orders with normalized coordinate bounds.
    • Editorial: 1-Click CMX 3600 Edit Decision Lists (.EDL) color-coded by severity (Red = Critical, Orange = High, Yellow = Medium, Green = Low) for direct marker import into DaVinci Resolve, Adobe Premiere Pro, and Avid Media Composer.
    • Music Supervisors: Standardized ASCAP / BMI / SESAC Music Cue Sheets auditing master (17 U.S.C. § 114) and sync (17 U.S.C. § 106(4)) splits.
    • Underwriters: A court-ready, multi-page E&O Legal Clearance Binder (PDF) generated dynamically via ReportLab with formal policy sign-off blocks.

🛠️ How We Built It

  • Cloud Infrastructure & Hosting: Containerized via Docker and deployed serverless on Google Cloud Run, handling asynchronous media pipelines, temporal OpenCV video decode, and dynamic ReportLab PDF synthesis.
  • Multimodal Intelligence: Built with the official Google GenAI SDK (google-genai) utilizing Gemini multimodal vision for zero-shot script and visual asset entity extraction.
  • Agentic Web Grounding: The Parallel Search API performs synchronous semantic lookups directly against real-world trademark and copyright registries at runtime, streaming live citations, latency metrics, and verified source URLs into the legal critique loop.
  • Dynamic Model Cascade Ladder: A resilient multi-tier fallback ladder (gemini-2.5-flash $\rightarrow$ gemini-1.5-pro $\rightarrow$ deterministic bedrock) that catches 429 RESOURCE_EXHAUSTED rate limits to guarantee zero pipeline downtime.
  • Deterministic Legal Safety Bedrock: A hard-coded Python statutory harness (app/harness.py) enforcing 6 core statutory invariants and safe harbors that cannot be overridden by LLM drift:
    • UPL Shield: Auto-injects State Bar decision-support notices across all UI views and document footers.
    • 17 U.S.C. § 120(a) AWCPA: Clamps exterior street-view architecture and public skylines to Low/Exempt.
    • 18 U.S.C. §§ 701, 706, 712: Enforces criminal symbol bans, automatically flagging Red Cross emblems and Federal badges for mandatory prop replacement.
    • NANPA Safe Harbor: Clamps non-555-0100 through 555-0199 phone numbers to Critical PII risk.
    • IETF RFC 2606: Flags non-example.com domains on prop screens as High Risk cyber-trespass liabilities.
    • 18 U.S.C. § 504 / CDA: Flags prop cash to verify single-sided motion picture money standards.
  • Production Studio UI: Dark-mode studio dashboard built with FastAPI, TailwindCSS, dynamic SVG bounding box overlays, live partner API trace logging, and 1-click export actions.

🛡️ Challenges We Overcame

  • Eliminating Legal Hallucinations via Dual Grounding: Vision models frequently hallucinate that famous modern marks or public artwork are public domain. We engineered the CriticHarness invariant layer, which programmatically blocks post-1929 commercial marks from public domain classification and verifies active registration status via live Parallel Search API telemetry.
  • Ensuring Zero-Downtime Resilience: To guarantee smooth hackathon evaluation without quota interruptions, we built a non-blocking Dynamic Model Cascade ladder paired with an offline statutory knowledge fallback that gracefully handles network latency or quota limits.
  • Asynchronous Audit Parallelization: Sequential LLM and web queries resulted in 15+ second audit wait times. We refactored the grounding engine using asyncio.gather(), allowing concurrent frame sampling and multi-entity registry searches that reduced turnaround to under 3 seconds.

🏆 Accomplishments We're Proud Of

  • 33/33 Passing Automated Tests: Full test suite verifying statutory invariant clamps, cascade failover tiers, CMX 3600 timecode calculations, and ReportLab PDF compilation.
  • Full Statutory Invariant Compliance: Complete deterministic adherence to federal entertainment law standards (17 U.S.C. §§ 106, 107, 114, 120; 18 U.S.C. §§ 504, 701, 706; Lanham Act § 1125; Rogers v. Grimaldi).
  • Pre-Compiled Deliverable Artifacts: Shipped sample E&O binders, DaVinci Resolve EDLs, and Form-4A release contracts in the repository examples/ directory for direct in-browser inspection by judges.
  • 1-Command Evaluator Experience: Built doctor.py and platform launcher scripts (run.bat / run.sh) that automatically verify dependencies, API keys, and OpenCV video decoders with zero manual configuration.

🔮 What's Next for CineClear AI

  • Native NLE Timeline Plugins: Packaging CineClear AI as an OpenFX and Adobe CEP plugin to run real-time clearance scans directly inside DaVinci Resolve and Premiere Pro timelines.
  • Automated Rights Outreach: Autonomous agent dispatch to send pre-filled Form-4A e-signature links directly to identified visual artists and copyright registries.

Built With

Share this project:

Updates

Submission history