Newsrooms
Prioritize incoming clips for further investigation. Keep the score, file hash and sampled timestamps together so editors know what was actually checked.
Add a useful screening step before you publish.
Screen original video files, understand which frames were checked, and hand an advisory report to the person making the editorial decision.
Beta, not forensic certification. No authenticity guarantee or enterprise SLA. Always combine results with source checks and human review.
Fact-checking teams
Add a repeatable screening step alongside reverse searches, original-source verification and contextual reporting. An AI score is not a publishable finding by itself.
Content agencies
Review supplied video assets before a campaign or handoff. Route uncertain results to a reviewer rather than presenting an automated verdict as proof.
Available today
Original-file screening and reports
- 8, 16 or 24 sampled visual frames
- Verdict estimate, model confidence and explanation
- SHA-256 file identifier and sampled timestamps in API reports
- Explicit detection and provenance limitations
A practical integration path
- Google sign-in for developer accounts
- Server-side API keys, revocation and prepaid credits
- Idempotent requests and private report retrieval
- Original files deleted after processing; reports retained 30 days
Multi-user team administration, SSO, custom SLAs, dedicated deployment, certified accuracy and human analyst services are not included in the self-service beta. Discuss requirements before assuming they are available.
Start with a scoped pilot, not an open-ended contract.
Pilots from $299
Tell us your workflow and sample volume. We will propose a written scope covering evaluation clips, API credits, onboarding and success criteria before you pay.
- Agree on a representative evaluation set and human-review process.
- Assess false positives, false negatives and unhelpful results—not just a headline accuracy score.
- Decide whether screening saves reviewers time before discussing recurring service.
Starting price is indicative, not a fixed entitlement or automatic subscription. Higher-volume or custom requirements are quoted individually.
Request a pilot proposalWhere TrueSight fits
Different checks answer different questions. A missing watermark is not evidence that a clip is human-made.
| Approach | What it checks | How to use it |
|---|---|---|
| Google SynthID | Watermarks embedded by supported AI-generation tools. | Useful positive provenance evidence when a watermark is detected. Coverage is not universal. |
| Hive | Dedicated AI-image/video, deepfake and other modality classifiers; documented generation/source heads. | Consider when you need specialized model coverage or platform-scale moderation. Benchmark on your own content. |
| Reality Defender | Enterprise deepfake screening with APIs, SDKs and organization workflows. | Consider for broader enterprise fraud and multimodal requirements. |
| TrueSight beta | Sampled visual video frames, an explanation and an interpretable API report. | Use as an accessible advisory screening step. We do not claim benchmark superiority, audio verification or SynthID/C2PA validation. |
Sources checked October 10, 2026: Google SynthID, Hive model documentation, Reality Defender API. This is a workflow comparison, not an accuracy benchmark.
Tell us what you need to verify
No login is required for an enquiry. Please do not submit confidential clips, passwords or API keys here.