Fact-Check Toolkit

The free Swiss-army knife for verifying images and video. Run one picture through five reverse-search engines at once, stress-test it for manipulation, pull hidden metadata, and trace where a clip actually came from — all in your browser. No upload, no account, nothing leaves your device except the searches you trigger.

One image, five engines

Paste the image URL, then hit each engine. Different engines index different webs — Yandex in particular matches aggressively on background and setting, which is why OSINT researchers run it even when Google finds nothing.

Or pick a local file — it is auto-hosted on a temporary relay that deletes itself after 24h, so every engine gets a one-click link. No manual uploads to each engine.
Direct URL works too if the image is already online. Yandex is the OSINT workhorse — always run it, even when Google finds nothing.

Error Level Analysis

ELA re-saves the image at a known JPEG quality and measures where the error spikes. Every region re-touched, spliced in, or pasted from a different photo has a different compression history — so it lights up brighter than its surroundings. Works fully offline on your file.

Read it like a pro: uniform dim glow = normal. One object glowing much brighter than the rest = that region was saved differently (edited, inserted, or just locally re-compressed — e.g. by a screenshot app). Bright ≠ automatically fake. ELA is a lead, not a verdict. PNG screenshots will light up everywhere — ELA only means something on JPEGs.

Video keyframe extraction

Uploads a clip, samples it frame by frame, scores each frame for sharpness and contrast, and throws away blurry transition frames — leaving the few stable keyframes most likely to identify the source when reverse-searched.

Waiting for a video…
Every keyframe is auto-hosted so each engine link is one click — no manual uploads. Check the oldest results first to find where the clip originated. Frames from news clips, TikToks and YouTube usually match; surveillance or drone footage rarely does.

Metadata & EXIF reader

Reads the raw EXIF block: capture time, camera model, software edits, orientation and GPS. Also converts UNIX timestamps both ways.

No file loaded.

UNIX timestamp converter

OCR — pull the hidden text

Runs text recognition over the image in your browser — useful for signs, licence plates, screenshots, documents, and partially blurred handles.

Timestamp & timeline builder

The classic verification move: if a photo or clip claims to show an event, find the earliest eyewitness uploads inside that time window. Feed in your keyword and window — this builds the exact search URLs for you.

Then compare: does the earliest upload predate the "official" narrative? Do eyewitness versions look sharper than the viral copy? Re-uploads compress — the crappiest copy is usually the oldest.
See who's behind the photo Put a name, number or face to a person in the US — public records lookup. Research without being tracked OSINT-grade privacy for your browsing — investigators don't leak their own IP. Learn this properly Structured OSINT & verification training — from image forensics to sock-puppet hygiene.

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.