What this fact-checking toolkit does
Verification professionals don't rely on a single engine or a single technique. This page bundles the standard open-source verification workflow into one place: cross-engine reverse image search (Google Lens, Yandex, Bing, TinEye, Baidu), Error Level Analysis (ELA) for detecting manipulated regions, keyframe extraction for tracing video clips to their origin, EXIF and metadata reading, UNIX timestamp conversion, OCR for pulling embedded text, and a temporal search builder for finding the earliest eyewitness uploads of an event.
Why run one image through five engines?
Every reverse-image engine indexes a different slice of the web. Google is strongest on Western commercial pages, Baidu covers Chinese-language sources, and TinEye's "oldest first" sort is the fastest way to date an image. Yandex deserves special mention: its matching is more forgiving than Google's — it aggressively matches on environment (backgrounds, rooms, skylines) rather than requiring the subject itself to be a near-duplicate. That's why the OSINT community treats Yandex as the go-to when everything else comes up empty, especially for identifying locations and faces.
How Error Level Analysis actually works
JPEG compression is lossy: every re-save discards slightly different information. When you paste an object into a photo, that object has already been JPEG-compressed on a different schedule than the background around it. ELA re-saves the whole image once at a fixed quality and visualizes the per-pixel error: regions with a mismatched compression history show up brighter than untouched areas. It will not hand you a verdict — a bright patch can also come from an innocent local edit, a sticker, or a screenshot tool — but it reliably tells you where to look closer.
Keyframe fragmentation: tracing a video to its source
A video can't be reverse-searched directly, but its frames can. The trick is choosing the right frames: transition frames and motion blur match nothing. This tool samples the clip, scores every candidate frame on sharpness (Laplacian variance) and contrast, drops the blurry ones, and hands you the stable keyframes — the frames most likely to appear in the original upload, unblurred enough for Yandex or TinEye to match. Pull two or three of them, not one: the first frame that returns a hit from an upload older than the viral copy is your origin.
What metadata can and can't tell you
EXIF can give you the capture time, device, editing software and sometimes GPS coordinates — hard evidence when it's present. But its absence proves nothing: Instagram, X/Twitter, WhatsApp and Facebook all strip EXIF from every upload by design, so a clean viral image is expected, not suspicious. That's why the timestamp converter and timeline builder matter more than metadata in most modern verifications: the play is finding the earliest public copy, not the richest one.
Honest limits
This is an aggregator and analyzer, not a magic oracle: it automates the workflow, but the judgment call — does the compression anomaly mean manipulation, does the "first" upload predate the event, is the matching result really the same scene — stays yours. For AI-generated images, ELA and metadata are weak signals; look instead at hands, text, reflections, and run the image through multiple engines to see if it appears nowhere else on the internet at all.