Find where your images are being used online. PixelProof runs fully on your Windows PC. It never uploads your images anywhere.
This is step 1, the quick prototype:
- Fingerprint your images. Point it at a folder (it works with Google Drive for desktop).
It reads each image once, makes a small "fingerprint", and remembers the photographer,
year and quarter from your folders (
Photographers \ Name \ Year \ Q1) and the date from the file name. It never copies your images. - Check one page. Paste a link. PixelProof downloads the pictures on that page, compares them with your library, and opens a report in your browser showing ours vs theirs side by side, with a match ID for each hit.
The crawler, the full desktop app and Proof of Discovery come in later steps.
You need: Windows 10 or 11, Python 3.10 to 3.13, and an internet connection for the setup downloads.
- Check Python. Press the Windows key, type
cmd, press Enter, then typepy --versionand press Enter.- If you see
Python 3.12.x(or 3.10, 3.11, 3.13), you're good. - If not, install Python 3.12 from https://www.python.org/downloads/. On the first install screen, tick "Add python.exe to PATH".
- If you see
- Put the PixelProof folder somewhere simple, for example
C:\PixelProof. (Not inside Google Drive. Thedatafolder it makes holds your fingerprint index.) - Double-click
install.bat. A black window opens and installs everything. The first time it downloads about 3 GB (the AI engine with graphics card support), so be patient. At the end it runs a check and should say All good! and name your graphics card.
Double-click PixelProof.bat. You get a menu:
| Type | What it does |
|---|---|
1 |
Fingerprint a folder of your images |
2 |
Check one page (paste a link) |
3 |
Check a picture file on this PC (handy for testing) |
4 |
Check a list of links (a .txt file, one link per line) |
5 |
Show what's in your library (images per photographer) |
6 |
Open the reports folder |
7 |
Edit your ban list (sites to skip, one per line) |
8 |
Check that everything works |
Pasting a folder path: in File Explorer, right-click the folder and choose Copy as path (on Windows 10, hold Shift while you right-click). In the black window, right-click to paste.
Stopping: press Ctrl+C or close the window. Your progress is saved. Run the same folder again and it carries on where it stopped. Later runs only look at new or changed files.
If you prefer typing, open cmd in the PixelProof folder and run:
.venv\Scripts\activate
pixelproof index "G:\My Drive\Photographers\Maya Chen"
pixelproof check https://example.com/some-article/
pixelproof status
pixelproof doctor
| File | What it is |
|---|---|
library.db |
One row per image: path, photographer, year, quarter, date, fingerprints, and every match found |
clip.faiss, hash.faiss |
The fast search indexes (rebuilt automatically) |
reports\ |
Every check report, as a web page you can open again |
banned.txt |
Sites to skip |
models\ |
The AI model, downloaded once |
Roughly 2 KB per image, so about 2 GB per million images.
Each picture gets two fingerprints:
- Perceptual hash (64 on/off values): catches straight copies, even resized, recompressed or flipped.
- CLIP (512 numbers from a free AI model that runs offline): catches cropped, filtered or watermarked copies.
| Category | Rule (starting values) |
|---|---|
| Exact copy | hash differs by 8 bits or fewer |
| Edited copy | flipped, or CLIP similarity 0.92 or higher |
| Possible | CLIP similarity 0.89 to 0.92, needs a human look |
The starting values live in pixelproof\config.py and will be tuned on your real images.
pip install pytest
python -m pytest