Content Credentials Checker for AI-Generated Images in Newsrooms
Newsrooms and fact-checkers receive images every day over WhatsApp and social media and must decide quickly whether they are real, edited or AI-generated. Build a desk tool that reads C2PA Content Credentials, checks for known AI watermarks and metadata, explains in plain language what can and cannot be concluded, and keeps a verification log for each image.
Intermediate
A small team, or one strong student willing to learn something new.1 semester · 2 students
7 to ship
3 optional extrasSuggested stack
What you should ship
- Upload and URL intake that preserves the original file bytes and records a hash
- C2PA manifest reading and validation: signer, claim generator, edit history and whether the signature checks out
- Metadata checks (EXIF, XMP, IPTC digital source type) and a note when metadata has been stripped, as most messaging apps do
- A plain-language verdict screen that separates 'proven', 'suggested' and 'unknown' and never claims more than the evidence supports
- A verification log per image with notes, reverse image search links the checker used, and the final decision
- A test set of at least 200 images you create or that carry open licences: camera originals, edited, AI-generated with and without credentials, and re-shared through messaging apps
- An evaluation of what survives re-sharing, with a table per platform
If you have time left
- Integrating a public AI-image detector as one extra, clearly labelled signal
- Urdu interface for regional newsrooms
- A browser extension that checks an image in place
The problem: is this picture real?
During floods, protests and elections, images spread faster than anyone can check them. Image generators now produce photo-realistic scenes in seconds. A fact-checker needs to answer quickly: where did this image come from, was it edited, was it made by AI? Detectors that output a single "92% fake" score are unreliable and hard to defend in print.
What already exists and the gap
- C2PA Content Credentials are an open standard for signed provenance: who made an image, with what tool, and what was changed. Adobe, Microsoft, OpenAI, Google and camera makers have adopted it to varying degrees.
- The Content Authenticity Initiative's Verify site reads credentials from an uploaded file.
- Invisible watermarks such as Google's SynthID mark content from some generators.
- The EU AI Act's transparency rules require providers to mark AI-generated content in a machine-readable way.
The gap is a newsroom workflow. Credentials are stripped by most messaging apps, so most images arrive with nothing. Journalists need a tool that combines every signal, says clearly what is unknown, and records the verification so an editor can review it later.
Inspiration
Fact-checking organisations publish their methods openly, and the C2PA specification is public. Your project turns the specification into a practical desk tool and measures how much provenance survives the way images really travel in South Asia.
What you build
- Intake that keeps the original bytes.
- Checks for credentials, metadata and known markers.
- A verdict screen written for journalists.
- A log for each verification.
The technically interesting part
- Validating credentials correctly. Check the signature chain, not just the presence of a manifest. A manifest can be copied onto a different image.
- What survives sharing. Send your test images through WhatsApp, Facebook, X and Telegram (using your own accounts) and record what metadata survives. This table alone is valuable.
- Wording. "No credentials found" does not mean "fake". Designing language that is accurate under pressure is part of the project.
How you will evaluate it
- Correct reading of credentials and metadata on your test set.
- The survival table per platform.
- A usability session with two or three journalism students or fact-checkers: time to a decision and whether they understood what the tool could not prove.
Path to a product
Newsrooms, NGOs and brand-safety teams all need verification workflows. A hosted desk tool with team logs, or an open source tool with paid support for media organisations, are realistic paths.
Scope warning
Do not build your own deepfake detector; that is a research project of its own. Use only images you create or that carry open licences, and do not publish verdicts on real people's images.
Learn what you need on Sefism
- AI creative roadmap: how image and video generators work, and the disclosure and provenance practices around them.
- AI creative cheat sheet: content credentials, watermarks and labelling in brief.
- AI governance roadmap: where transparency obligations for AI-generated content come from.
- Stable Diffusion: generate your own test images locally.
- FastAPI tutorial: the backend for intake and checks.
- Can I make a career with AI image and video generation?: how AI images are made and sold, from the creator side.
Get help with this idea
The risk is promising certainty the evidence cannot give. Tauseef can help you scope the checks, build a fair test set and design a verdict screen journalists trust. If you are new here, create a free account first.
Ideas and guidance, not finished projects
These are project ideas and scoping guidance, published free for students to use as a starting point. I do not build, write, or sell final-year projects, and I do not complete coursework for anyone. Take an idea, make it yours, and build it.
Not sure this one fits you?
I help students pick an idea that matches their skills and their deadline, then scope it down until it is finishable. Guidance only, never done for you.
Stuck on something specific?
Writing only gets you so far. If you want an answer to your situation rather than the general case, book a session and we will work through it together. Sessions are free for approved Sefism members, and a few slots open each week.
Follow along
New writing, resources and project ideas land here first.