●Media verification for the AI era

If it’s AI, we catch it.

Drop in a photo or clip. Witness shows you what is real, what is AI, and why, in seconds.

Likely real · 96%AI-generated · 0.94Face swap · 0.90
↑

Scan a file

JPG · PNG · WebP · MP4 · up to 50 MB

Drag & drop hereor
No file handy? Try a sample
Free demo, no signup required.Try 5 scans now. Sign up for 20 a month.
Sign up free →
98.9%AUROC on public detection benchmark datasets
<3sFrom upload to verdict
75+Generators identified
0 daysFiles never kept

Trained on the generators people actually use

Diffusion modelsGANsAutoregressiveHybrid generatorsText-to-imageText-to-videoFace swap toolsUpscalersInpaintingLoRA fine-tunesImg-to-imgScreen capturesDiffusion modelsGANsAutoregressiveHybrid generatorsText-to-imageText-to-videoFace swap toolsUpscalersInpaintingLoRA fine-tunesImg-to-imgScreen captures

●The product

Built to catch what the eye can’t.

Every file runs through specialist detectors. One score, the evidence behind it and an audit trail.

01

Catch every AI image

Diffusion models, GANs, upscalers and 75+ generator families. A dedicated image detector, with results in seconds.

  • Heatmap highlights the regions the detector flagged
  • Generator attribution: names the generator family when confidence is high
02

See the exact frame it breaks

Timeline markers pinpoint the exact frames where a face was swapped or a scene was generated.

  • Lip-sync drift, motion coherence and inter-frame geometry checks
  • Separate detectors for AI-generated video and face-swap video
03

Fast. Explained. Defensible.

Every scan returns a likelihood, the signals that drove it and a permanent audit ID your team can cite.

  • Every signal listed with its contribution to the final score
  • Audit ID recorded for every scan
  • Honest about limits: a signal, not a verdict of intent
Blend seam
Noise mismatch
video-call-frame.pngFace swap detected
94% AI
94%LIKELIHOOD AI-GENERATED
Face swap
AI detection
92
Face-swap check
81
Generator confidence
63
Metadata signals
12
audit_id: aud_7f3k9xExport PDF · soon

●Specialist detectors

Detectors trained for the job.

General detection catches most fakes. Specialist detectors catch the ones that matter in your field.

●Where to use it

Check it where you see it.

On the web, in Chrome, on a live call, from your AI agent or inside your own app.

●For teams

Built for the queues that matter.

The same engine, built into the workflows that handle media at volume.

Enterprise & API

Run it your way.
Dashboard or API.

Selectedugc_submission_0412.jpg
AI-generated0.91
witness.vision/queueSample data

Queue

Review needed
AllFlaggedReviewPassed
FileSourceTypeVerdictScore
ugc_submission_0412.jpgNewsroomImageAI-generated0.91
claim_photo_48213.jpgClaimsImageAI-generated0.96
interview_recording.mp4HiringVideoFace swap0.90
refund_evidence_7731.jpgTrust & safetyImageLikely real0.06
exhibit_14.jpgLegalImageLikely real0.12

Case WX-0412

Debris pattern
  • SceneDebris repeats in a generator pattern
  • ProvenanceNo C2PA credentials found
  • ReportReport export coming soon

One call. One verdict.

Put AI detection inside your own product. Score, verdict and an audit ID back in seconds. Self-serve. Usage-based pricing.

  • REST API Early access
  • Batch and webhooks
  • MCP server for agents
POST witness.vision/api/v1/image
# scan a file
curl -X POST witness.vision/api/v1/image \
  -F image=@claim_photo.jpg

→ {
  "verdict": "ai_generated",
  "likelihood": 0.94,
  "signals": { "face_blend": 0.91, "diffusion": 0.88 },
  "audit_id": "aud_7f3k…"
}

●Why Witness

Evidence, not just a score.
Built to be trusted.

98.9%

AUROC on our benchmark on public detection benchmark datasets.

0 days

Files never kept. Media is processed and discarded.

75+
DiffusionGANAutoregressiveHybridText-to-imageText-to-videoFace swapUpscalerInpaintingImg-to-imgLoRA fine-tune

Continuously trained on the latest generators.

Every verdict shows its work

Example scan

A likelihood, the signals that drove it and an audit ID your team can cite.

Blend boundary
92
Skin texture
81
Lighting match
44
Metadata
12

One verdict across media

A specialist detector per media type runs on its own. The results are then combined into one verdict with a confidence score.

ImageVideoDocuments · soon

Honest about limits

Strongest on
  • Fully AI-generated images and video
  • Unedited face swaps
  • Known generator families
Weaker on
  • Heavily edited or re-compressed files
  • Very low-resolution media
  • Very short clips
  • Generators released after our last training update. We retrain continuously.

A likelihood, not a certainty. A signal, not a verdict of intent.

●Pricing

Start free. Upgrade when you need more.

No card needed. Upgrade when you need more.

Free

$0
  • 20 image scans a month
  • 1 min video a month
  • Full heatmaps and frame timeline
  • Generator attribution
  • EXIF, C2PA and SynthID checks
  • Web and Chrome extension
Create free account

Pro

$15 /month or $120/year
  • 1,000 image scans a month
  • 15 min video a month
  • Everything in Free
  • Faces and face-swap specialist
  • Scan by URL: paste any link
  • Email support
Get started

Enterprise

Custom
  • Volume pricing
  • Everything in Pro
  • SSO and audit logs
  • Dedicated support
  • Custom SLA
  • On-premise option
Talk to sales

●FAQ

Common questions.

Cannot find your answer? Our team replies fast.

Still have a question?Talk to us about plans, volume or on-premise.Contact us

Don’t guess. Witness it.

Sign up free and get 20 scans a month. No card needed.

Sign up free →No card needed · Cancel anytime
WITNESS.VISION© 2026 WITNESS