Skip to content
AI Image & Video DetectorAudio & Video

AI Image & Video Detector

Choose local models or official external verification tools, including SynthID, and keep a separate evidence log. Results are not proof of origin.

Loading Tool…

What Is the AI Image & Video Detector?

AI Image & Video Detector is a local-first media review workspace with eight method choices. PE-Core Tiny and PE-Core Small run directly on your device, with integrity-pinned weights and uncalibrated model scores rather than authenticity probabilities. Video analysis samples 4–24 frames within a chosen range; this is still-image classification, not temporal deepfake, face identity, audio or watermark detection. Provider Checks separately opens Google SynthID Detector, OpenAI Verify, Adobe Content Credentials Inspect, NVIDIA Synthetic Video Detector, Hive and Reality Defender. These are external tools, not bundled or automatically executed engines. Format coverage, account requirements and privacy terms differ. A bounded local byte-marker scan is not authenticated provenance. No missing signal establishes human creation.

How It Works

Select Detection Method, then choose an original PNG, JPEG, WebP, MP4, MOV or WebM file supported by that method. For an on-device model, choose optional video sampling settings and select Analyze Media. Review Overview, Frames and File Evidence. For an external tool, follow Provider Checks and choose Open Official Tool; no file is automatically transferred. If you voluntarily check the original file with the provider, record its finding and notes after confirming the same file was checked. OpenAI API response JSON can be imported as unverified structured evidence, with C2PA and SynthID signals shown separately. Export Report includes local model results and external records without combining them into a confidence score. Frame CSV and External Checks CSV export their respective measurements.

When to Use It

Use it as one source of evidence when checking an unfamiliar image or video, alongside original files, publisher context and independently verified provenance. Never use detector output alone for legal, disciplinary, professional or other consequential decisions about a person.

Frequently Asked Questions

Can This Prove An Image Or Video Was Made With AI?
No. False positives and false negatives occur. Compression, filters, screenshots, edits, new generators and non-photographic media can change results. Low scores do not prove a real camera origin, and high scores do not prove AI generation. The review bands at 0.20 and 0.80 are interface choices, not validated confidence limits.
Which Models Run Directly In BrowserTools?
PE-Core Tiny (12.5 MB) and PE-Core Small (47.5 MB) from Deckard618/imdet are self-hosted and pinned to a revision and separate SHA-256 hashes. Both use a 256-by-256 RGB center crop; parts outside it are not classified. The larger model is not guaranteed to be more accurate. Author benchmarks have not been independently reproduced by BrowserTools. Neither model verifies SynthID or Content Credentials.
How Are Videos Checked?
The tool runs the image model on selected frames, not on motion, voices or the full video. The median and score range summarize those samples only. Short AI inserts, face swaps and changes between samples can be missed. Scores from nearby frames are correlated and must not be treated as independent votes.
Are My Files Uploaded Or Saved?
BrowserTools does not upload media or save files, external findings or API keys to browser storage. Local model assets are downloaded from this site and may be cached. Choosing an external method does not send a file; if you upload on the provider website, that provider handles it under its own terms and privacy policy. Records stay in memory and clear when you replace the source file. Export a report before leaving if you want a copy.
Does It Validate Content Credentials?
No. The file scan only looks for unverified byte markers. It does not parse or validate C2PA manifests, signatures, signer trust or ingredient relationships. A marker may refer to an embedded ingredient and can be spoofed; its absence does not establish authenticity. Use a dedicated Content Credentials verifier for provenance validation.
Can I Choose Google SynthID?
Yes. Select Google SynthID Detector to open the official public verification portal and log its reported finding. It checks supported invisible watermarks, not merely visual artifacts. BrowserTools does not run or reverse engineer Google’s watermark detector locally. Provider coverage can change, and a missing watermark does not mean human-made. The official Gemini alternative is linked separately.
Are External Results Automatically Verified?
No. External log entries are user-recorded, or imported from an unverified OpenAI Content Provenance JSON response. BrowserTools does not rerun the external check, authenticate the response or cryptographically bind it to your file. C2PA and SynthID fields remain separate, and different evidence types are never averaged into an authenticity score. Enterprise services can require accounts, contracts, API access or approved deployments.
What Are The File Limits?
PNG, JPEG and WebP images are limited to 30 MB and 40 megapixels. Videos are limited to 1 GB and must be decodable by your browser; MOV codec support varies. At most 24 frames are sampled. Metadata marker scanning covers the full file up to 24 MB, otherwise the first and last 12 MB. Animated images and audio analysis are not supported.

Found an issue? Send Feedback. See our data sources and update policy for tools that depend on changing information.