LmCast :: Stay tuned in

Show HN: Is this photo edited? Client-side image forensics

Recorded: Sept. 14, 2026, 5:08 p.m.

Original Summarized

Is This Photo Edited? - Free Image Forensics Tool | Vajba

Vajba/Image Tools·Search

Check a photo for signs of editing, AI generation, hidden metadata, and compression artifacts

Is This Photo Edited?
Check a photo for signs of editing, AI generation, hidden metadata, and compression artifacts. Runs entirely in your browser. Your file is never uploaded.
Finds clues worth investigating, not proof that a photo is edited or authentic.

Upload a photo for forensic analysis

Drop a photo to analyze
or paste an image (Cmd/Ctrl+V) or image URL
JPEG, PNG, WebP, HEIC, AVIF

Select Image

Load

Pasting a URL fetches the image from that site to your browser. Vajba never receives the file, but the remote site will see your browser request. Downloaded reports include the Source URL, and reverse-search links pass it to those services.

No photo handy? Try an example:
A Photoshop edit with cloned sky
An AI image (ChatGPT)
A Photoshop export
A screenshot

Also useful:
EXIF Viewer ·
Remove EXIF ·
File X-Ray ·
Hide Message in Image

Overall Read

Running tests...

Findings are per-test. The overall read sums only suspicious evidence from Hard and Supportive tests.

What this result means, and what it does not
What it means: every finding here comes from the file's own bytes, parsed in your browser. We look at metadata, encoder signatures, embedded provenance claims, pixel statistics, and a handful of structural clues.
What it does not mean: "no findings" is not a proof of authenticity. AI-image detection can be unreliable, particularly after resizing, recompression or other processing, and metadata is trivial to strip or forge. Reverse image search is usually a better authenticity check than any algorithm here.
How to read suspicious findings: hard evidence is a fact from the file. Supportive evidence is a hint that needs corroboration. Correlated findings from one cause are down-weighted so they do not stack into an exaggerated verdict.

Expand all details

Cancel analysis

Print / save as PDF

Download JSON

Compare with another image

Analyze another photo

Forensic findings grouped by evidence class

About AI Detection

What works, what doesn't, and what this app actually does
In principle: cryptographic content provenance (C2PA / Content Credentials) is the most reliable way to know how an image was produced. When a Photoshop, Firefly, DALL-E, Leica, or Sony tool signs its output, a verifier can confirm who signed it and what edits they recorded.
In this app: we detect and parse Content Credentials from the file bytes, cryptographically verify the COSE_Sign1 signature in your browser via WebCrypto (ECDSA P-256/P-384/P-521 and RSA-PSS), check that every assertion the signer created hashes to the value in the signed claim, verify the asset hash binding (the file's bytes must match the signed SHA-256), and walk the certificate chain to the official C2PA trust list (vendored from c2pa-org/conformance-public), matching anchors by subject and public key. "Verified" means those checks passed. Not checked: certificate revocation, the signing-time timestamp, key-usage policy and ingredient manifests - the official verifier at contentcredentials.org does more.
For images without Content Credentials: AI-image detection can be unreliable, particularly after resizing, recompression or other processing. These checks cannot establish that an image is, or is not, AI-generated. External AI detectors below can give a second opinion at the cost of uploading the image to their servers.

Verify a Content Credential
Hive (uploads your image)
AI or Not (uploads your image)
Sightengine (uploads your image)

Home
·
Tools
·
Image Tools
·
PDF Tools
·
Audio Tools
·
Text Tools
·
Developer Tools
·
Calculators
·
Automotive
·
Generators
·
Everyday Tools
·
Fun
·
About
·
Privacy

Say hi at [click to reveal email]

The Vajba tool functions as an in-browser image forensics utility designed to detect signs of editing, artificial intelligence generation, hidden metadata, and compression artifacts within digital images. The process involves analyzing the file's intrinsic byte structure directly within the user's browser, ensuring that the file itself is never uploaded. The analysis draws upon examining metadata, encoder signatures, embedded provenance claims, pixel statistics, and various structural clues to gather evidence. However, the findings generated by the tool are intended to provide clues for investigation rather than offering definitive proof of an image's authenticity or manipulation.

The reliability and interpretation of the findings are governed by the evidence class. Hard evidence represents verifiable facts derived directly from the file's data, while supportive evidence serves as hints that require external corroboration. The tool applies a system where correlated findings from a single source are down-weighted to prevent exaggerated verdicts.

A significant aspect of the tool’s forensic capability centers on cryptographic content provenance, specifically utilizing Content Credentials (C2PA). This method is presented as the most reliable means of determining how an image was produced. When image generation or editing software, such as Photoshop, Firefly, DALL-E, Leica, or Sony tools, applies a signature to their output, a verifier can use C2PA to confirm the identity of the signer and the specific edits recorded. The system performs rigorous cryptographic checks, including verifying the COSE_Sign1 signature using algorithms like ECDSA and RSA-PSS, ensuring that all assertions made by the signer hash correctly to the claimed values, and verifying the asset hash binding to confirm that the image bytes precisely match the signed SHA-256 hash. Furthermore, the process involves walking the certificate chain against the official C2PA trust list to validate the subject and public key anchors.

For images that lack this Content Credential data, the detection of AI-generated content becomes inherently less reliable, particularly after the image has undergone resizing, recompression, or other processing, as metadata is susceptible to being stripped or forged. In such scenarios, the tool’s algorithmic detection capabilities may be unreliable. Consequently, the text suggests that reverse image searching is often a more robust method for assessing authenticity than relying solely on these algorithmic tools. The overall forensic findings are consolidated by summing evidence from the hard and supportive tests performed on the file.