← Back to home
Changelog
Every model version, detection improvement, and milestone Vera has shipped. Newest first.
- v0.10.97ModelApril 7, 2026
v4 burst rescue + sporadic-cap exemption
- • New burst-rescue mechanism: when v4 detects strong AI signals on ≥25% of frames (even non-consecutive), the scan is no longer suppressed by the v12 low-confidence gate.
- • Caught a previously-missed AI music video where v4 fired on 37.5% of frames but the prior heuristic capped the result at 0%.
- • Added a new audio-v2 false-positive cap branch to prevent burst-rescued scans from being downgraded.
- v0.10.95FeatureApril 5, 2026
Detailed PDF reports
- • PDF reports now include attack-category breakdowns for each detection signal so users learn what threats each model catches.
- • Added 'How to read the score' legend and risk-band guidance.
- • Each scan now exports as an educational forensic document, not just a score.
- v0.10.90FeatureApril 3, 2026
Free tier expansion + deep-scan tier gate
- • Free tier increased from 3 quick scans/day to 5 quick scans/day.
- • Detailed scans (full forensic analysis) gated to Starter tier and above.
- • Capability paywall replaces hard quota cliff — free users can keep scanning, but power features are paid.
- v0.10.80FeatureMarch 28, 2026
Pro tier — always-on guardian
- • Pro repositioned around 24/7 protection, not just higher quotas.
- • New Pro features: auto-scan share sheet, public verified report sharing, custom alert rules, browser extension.
- • Watchlist with scheduled scans now triggers user-defined alert thresholds.
- v0.10.50ModelMarch 15, 2026
Multi-model ensemble (v1 + v2 + v4 + v12)
- • Combined four independent detection models for video, image, and audio.
- • v12 acts as a confidence gate to suppress v4 false positives on real footage.
- • Composite score reflects the consensus of multiple models, not a single prediction.
- v0.10.0InfraFebruary 20, 2026
On-device pre-screening
- • RawNetLite (audio) and a lightweight image model now run on-device for the first pass.
- • Scans below the on-device confidence threshold never hit the cloud — protects user privacy and reduces compute cost.
- • Cloud analysis only triggered on borderline or high-risk content.
- v0.9.0FeatureFebruary 1, 2026
Public beta launch
- • Vera launched on TestFlight and Google Play closed beta.
- • Initial signals: voice authenticity, facial landmarks, image forensics, content provenance.
- • Free + Premium tiers available at launch.
Why we publish this
Detection models drift as generators improve. We publish every material change to our scoring so users and partners can audit how Vera evolves. If a result seems wrong, the version stamp on each scan tells you exactly which model produced it.