Is it AI-generated?
Check images & deepfakes.
Check whether a photo or video frame was AI-generated or edited. Read EXIF and GPS metadata, Content Credentials (C2PA) and generator declarations, and inspect Error Level Analysis and noise maps. Review audio spectrograms and text references, then export your findings as forensic reports. These checks run in your browser, so your files stay on your device.
Open an image or video
Drop a file or choose one. Local checks start right away, and the file stays on your device.
Review metadata and Content Credentials
Check EXIF, GPS, C2PA generator and source-type declarations, plus ELA and noise indicators.
Weigh the evidence
Compare the findings with each other and with independent sources before you conclude.
Browser tools are free, with no account required. Save examined sources, captured frames, text revisions and findings in a local project. Check its save status before closing.
Media forensics & deepfake analysis FAQ
Image checks, video investigation, audio and text analysis, and file privacy.
Getting started
What is PixLab Forensic used for?
PixLab Forensic is a media intelligence and forensic analysis workbench for images, video, audio and text. Investigate deepfakes and AI-generated imagery, inspect photo metadata and GPS locations, review recordings and analyze text references. Keep source files, notes and findings together in local projects and export forensic reports.
How do I inspect a file?
Open a file or drop it into the workbench. Choose Deepfake Detection to investigate an image or video, or use the Images, Video, Audio and Text tabs for specific tools. Paste text directly into the Text tab. Use Resume a project to continue a saved investigation.
Is PixLab Forensic free, and do I need an account?
The browser tools are free and require no account or installation. AI features may require sign-in and are subject to plan limits.
Why does How-to & FAQ open first?
First visits open the guide. After you choose a workspace, PixLab Forensic remembers the last open tab, including Deepfake Detection, when browser storage is available. Use Manage to save examined sources, notes, text revisions, captured frames and available findings in a local project. Check the save status before closing. Resume a project on the guide or reopen retained sources from Manage in the same browser. Unsaved work remains session-only; clearing browser storage removes local projects.
How can I make the text larger?
Open Aa (Text size) in the header, or choose Manage > Settings > Adjust text size. Select Standard (100%), Large (115%), Larger (130%) or Largest (150%). The change applies immediately and is remembered in this browser when storage is available.
Deepfake & AI analysis
How can I check whether an image is AI-generated?
Open an image in Deepfake Detection and choose Check for AI. Review the result alongside its metadata, Content Credentials and forensic indicators. Use these findings together to investigate how the image was created or edited.
How do I check Content Credentials (C2PA) in an image?
Open a JPEG or PNG image. The workbench reads any embedded Content Credentials manifest in your browser and shows the declared generator, the digital source type and the result of the claim-signature check. A source type such as trainedAlgorithmicMedia declares that generative AI created the content. Missing credentials do not show whether an image is authentic or AI-generated, because most images carry none and credentials can be removed.
Which media types can I analyze for deepfakes?
Deepfake Detection accepts images and video. For a video, pause at a moment you want to examine, capture a frame and choose Check for AI. The result describes that frame. Use the Audio and Text tabs to investigate sound patterns and written references.
Can I ask AI questions about an image?
Ask AI for a scene summary, an explanation of a visible detail or a possible location. Your question sends an image preview and relevant metadata for analysis. Compare the answer with the source and other evidence.
Media tools & location
Can I find where and when a photo was taken?
Read recorded GPS coordinates and capture dates from supported photo metadata, then view the location on a map. Online maps need internet access; a local map archive can be used offline. Recorded values may be edited, and missing coordinates or dates cannot be recovered from metadata.
Can I estimate an image's location without GPS metadata?
Ask AI where the scene may have been photographed. Landmarks, signs and other visible details can suggest possible locations. Check these suggestions against independent sources; a location estimate is different from recorded GPS coordinates.
What text analysis tools are available?
Paste text or import a TXT or Markdown file up to 5 MB. Count words, search terms and follow URL, email and date references to their passages. Keep the original alongside a working revision, add notes and save it in a project. Ask AI to review a short excerpt. These tools support investigation, rather than identifying text as AI-written.
How do audio and video forensic analysis work?
Analyze audio with waveform playback and local FFT spectrograms, including channel selection and frequency inspection. For video, play and seek through a recording, capture timestamped frames and inspect them in Images or Deepfake Detection. Each captured JPEG keeps its parent recording, timestamp, dimensions, encoding settings and hash. Save the recording and its frames in a local project to reopen the investigation. Playback depends on browser codec support; transcription and automated speaker identification are not provided.
Can I search for matching images or identify people?
SHA-256 and MD5 hashes identify exact file matches; perceptual dHash comparisons highlight visually similar session images for review. Imported hash lists support known-file checks. The entity graph links examined sources through recorded metadata and verified source identity. Reverse image search and automatic identification of people are not available.
What is Error Level Analysis (ELA)?
ELA re-saves the image as a JPEG at a chosen quality and amplifies the difference from the original. Areas with a different compression history can stand out, which may point to edits or pasted content. Re-saving, resizing, text and sharp edges can produce similar patterns, so treat ELA as one indicator rather than proof.
Privacy & evidence
Does my media stay on my device?
Metadata checks, playback, frame capture, spectrograms and reports run on your device. AI features send the image or text excerpt you ask to analyze; opening a file does not send it. Online maps request map data. You control online features in Settings. Save sources in a local project to keep them after a refresh.
Can I view EXIF metadata without uploading the photo?
Yes. Camera details, capture dates, GPS coordinates, hashes and Content Credentials are read by your browser on your device. Opening a file does not send it anywhere; only AI features you choose to use send the image or text excerpt you ask about.
Can the workbench prove that media is authentic?
No single metadata field, visual observation or AI assessment proves authenticity. Recorded information may be incomplete or altered, and AI interpretations may be wrong. A low AI-generation assessment does not prove that an image is a genuine camera photograph. Compare the source and its context with independent information.
Can I check whether an image has been edited?
Use Error Level Analysis, JPEG quantization-table diagnostics, local noise-variance overlays and embedded-thumbnail comparison to inspect possible inconsistencies. SHA-256 and MD5 identify exact bytes, and dHash supports similarity review. Compression, resizing and ordinary editing can produce similar patterns, so these indicators do not independently prove manipulation or AI generation. For a captured video frame, JPEG diagnostics and missing EXIF or Content Credentials describe the derived image, not the original recording.
Does a valid Content Credentials signature prove the image is authentic?
No. The local check can verify a supported signature against the displayed claim and embedded certificate. Assertion hashes, the binding to the image bytes, certificate trust and revocation are not validated by this workbench. Unsupported or malformed current credentials are reported rather than replaced by an older claim. A valid claim signature is not a verified authenticity verdict.