Is This Photo Real Skill (useOSINT): Verify Images

Makes an agent check whether a photo or video is authentic, original and correctly captioned, starting with provenance.

Installs: 58.3K; on skills.sh
Stars: 58; on GitHub
Files: 2; besides SKILL.md
Updated: Aug 2026; last repo commit

This is a method skill from useOSINT. It makes the agent verify a photo or video in a fixed order, provenance first and pixels last: find the earliest copy and its caption, read metadata, check for C2PA Content Credentials, check details and lighting, and only then run signal-level tests. It ends with a report of what was verified and a confidence grade for each claim.

The skill bundles no scripts. Its text names tools such as ffprobe, MediaInfo and yt-dlp for video, and warns that error level analysis and AI-detector scores are weak evidence. It tells you to hand the earliest-copy search to the find-the-original-image skill, which is a separate install.

Good for

Fact-checkers and analysts testing whether footage is recycled or miscaptioned
KYC and onboarding fraud teams checking submitted photos
Insurance claim reviewers who need to know if an image was altered

Install is-this-photo-real

Claude Code

Copy this prompt.

Install the is-this-photo-real skill in Claude Code. Run this command:

/plugin marketplace add useosint/skills
/plugin install useosint@useosint

Then check that it loaded and tell me how to use it.

2
Paste it into Claude Code. It shows the change and asks before running it.

Pick one route. The Claude Code plugin installs all 29 skills of the repo (plugin.json and marketplace.json say 29, the README text says 28, and the repo tree has 29 skill folders). The plugin folder holds only plugin.json and marketplace.json, with no hooks, agents or MCP servers. The first skills CLI line and the folder routes install just this skill; the helper line adds the skills its method names. The README still names the repo useosint/osint-skills; the repo is now useosint/skills and GitHub forwards the old name. install.sh replaces any folder in the target that has the same name as one of the skills (it runs rm -rf on each destination first). Each skill can run on its own, and its sibling skills are called out in its method or its pivot table. There is no Claude app tab because the README documents no upload route and the skill calls for local tools.

Built from the package data we hold. Not yet tested by us. Source.

What's in the skill

When it loads

Verify whether an image or video is authentic, original and correctly captioned — provenance checks, error level analysis, noise and JPEG compression analysis, clone and copy-move detection, lighting and shadow consistency, C2PA Content Credentials, deepfake and AI-generation tells, and the honest limits of AI-detector tools. Use when fact-checking a photo or video, checking for a deepfake or AI-generated image, spotting manipulation, or testing whether footage is recycled or miscaptioned. Applies to KYC and onboarding fraud, insurance claim review, disinformation analysis, and evidence admissibility. Reference at useosint.com/skills/is-this-photo-real.

Sections in SKILL.md

Triage: what question are you actually answering
Method
Lighting, shadow and geometry — the checks that hold up
Error level analysis, and why it is mostly used wrongly
C2PA and Content Credentials
AI generation — durable tells and tells that rot
Video
Where this goes wrong
Confidence grading
Worked example
Pivots
Legal and ethical notes

Key facts

Author: useOSINT (plugin.json lists the author as useOSINT; the repo is useosint/skills)
License: MIT (LICENSE file, README badge and plugin.json)
Installs on skills.sh: 58,275 (Nov 2, 2026)
skills.sh audits: Gen Agent Trust Hub pass, Socket pass, Snyk pass (Nov 2, 2026)
GitHub stars: 58 for the whole useosint/skills repo (Nov 2, 2026)
Last repo commit: Aug 3, 2026
What it includes: 3 files including SKILL.md (counted from the repo tree): SKILL.md, reference/tool-catalogue.md and reference/verification-checklist.md. It bundles no scripts.
Needs: The image or video file, ideally the least-processed copy. Optional local tools named in the text: ffprobe, MediaInfo, yt-dlp
Works in: The README names Cursor and Claude, and says "other agents" can use the skills; the plugin.json and marketplace.json make a Claude Code plugin
Who it is for: Fact-checking, deepfake and AI-image checks, KYC and onboarding fraud, insurance claim review, disinformation analysis, evidence admissibility (SKILL.md description)

What it includes

Picks the test that matches the claim
A table maps five claims (event and place, unaltered, said on video, AI-generated, genuine screenshot) to what to do first and what not to do.
Try: “Is this photo really from the event it claims? Here is the file.”
Read · Triage
Finds the earliest copy before any pixel test
Gets the least-processed copy, hashes it, and searches for earlier appearances with a different caption. For video it extracts keyframes. The search itself is handed to find-the-original-image.
Network · Best copy and provenance
Checks the editing chain and Content Credentials
Reads the editing chain, thumbnail and timestamps (via secrets-in-file-metadata) and looks for a C2PA manifest, noting that a missing manifest means little.
Read · Metadata and C2PA
Compares details, lighting and shadows with the claim
Checks signage, plates, season and weather against the claimed place and date, then shadow convergence, reflections, perspective and lens traits.
Read · Consistency checks
Noise, JPEG compression and clone checks as weak input
Notes where error level analysis and double-compression tests help and where platform re-encoding makes them void.
Read · Signal-level forensics
ffprobe, MediaInfo and audio and face-swap tells
Inspects encoder strings, frame types, generation loss, audio and face-swap behaviour, using commands such as ffprobe -show_format -show_streams.
Runs code · Video checks
Writes what was verified with a confidence grade
Grades each claim separately as confirmed original, confirmed recontextualised, probably manipulated, probably synthetic, unconfirmed or cannot be assessed.
Read · Report and grading

From the official docs

Permissions

Low risk

The skill bundles no scripts, but it suggests running tools on your files and using online sites that may keep what you upload. Keep sensitive images off public forensic tools.

What it can do

Read your data
Create and change things
Delete things
Read-only mode

Watch out for

Online tools can keep your upload
SKILL.md warns that uploading material to online forensic services discloses it, and that some publish submissions to public galleries. Check before sending anything sensitive. github.com
Harmful media must not be analysed
For media that depicts a crime, a victim or intimate content, SKILL.md says to minimise copies and not redistribute. For child sexual abuse material it says to stop at once and report to the authority. github.com

Security advisories

No published security advisories as of Nov 2, 2026.

Technical details

Turning on read-only	No read-only mode. The skill bundles no scripts.
Sign-in and keys	SKILL.md names no account or key.
Scopes	Reads the image or video file you give it May run local tools such as ffprobe, MediaInfo and yt-dlp that SKILL.md names Its text points to online reverse-search and forensic sites

Questions

What does is-this-photo-real do?

It gives the agent a method for checking a photo or video: find the earliest copy and caption first, then metadata, C2PA, consistency and only last pixel forensics. It ends with a graded report, never a bare "fake" or "real".

Can it tell me if an image is AI-generated?

Only partly. SKILL.md lists durable tells such as impossible lighting and missing camera noise, and warns that detector scores are a weak signal that should never be the sole basis of a conclusion. Its grades include "Unconfirmed" and "Cannot be assessed".

Does it need the other useOSINT skills?

Its method steps tell the agent to run find-the-original-image, secrets-in-file-metadata and geolocate-from-pixels, so install those with it for the full method. The install box has a skills CLI line that adds the three, and the plugin route installs all 29.

Is it safe to use?

It bundles no scripts, and it suggests running tools like ffprobe and yt-dlp on your files. SKILL.md warns that uploading material to online forensic services discloses it to them and that some publish submissions. Our advice: do not upload sensitive images to online tools.

Does it work in Claude Code, Codex or Cursor?

The install box has Claude Code, Codex, Cursor and standalone tabs. The README names Cursor and Claude and says other agents can use the skills. It documents no Claude app upload.

How is it different from find-the-original-image?

This skill decides whether a photo or video is authentic and correctly described, and hands the earliest-copy search to find-the-original-image. That one only finds where a picture came from. Both are in the same useOSINT repo.

Information

updated Nov 2, 2026

Repository: useosint/skills
Folder: skills/is-this-photo-real

useosint/skills skills.sh

Sources

Docs, issue trackers and discussions this page is built from.

Show all 7

is-this-photo-real SKILL.md · github.com
useosint/skills README · github.com
useosint/skills LICENSE · github.com
useosint plugin.json · github.com
useosint marketplace.json · github.com
useosint ETHICS.md · github.com
tool catalogue · github.com