AI Watermark Popper (Apify Actor)

Strip AI provenance metadata (C2PA / EXIF / PNG text) and apply quality-gated imperceptible pixel disruption.

What you get

• Social auto-labels (Instagram / YouTube / X / LinkedIn): high success (~90–95%) against automatic file-based AI labels.
• SynthID-class pixel verifiers: best-effort — not guaranteed.
• Style classifiers / disclosure checkboxes: out of scope.

This Actor is the headless twin of the Chrome extension pipeline. It does not inject into ChatGPT/Gemini tabs — you pass image URLs.

Input

Field	Type	Default	Notes
imageUrls	string[]	—	Required. Public HTTP(S) image URLs
strength	light | standard | strong	standard	Quality-gated disrupt dose
outputFormat	same | png | jpeg | webp	same	Output container
maxImages	int	50	Cap (1–200)

Output

• Key-Value Store: cleaned-000.png (etc.) and OUTPUT = last successful file
• Dataset: one row per URL with ok, filename, report summary, or error

Local run

# from monorepo root — needs Apify CLI + Python 3.11+
cd apps/apify-actor
python -m venv .venv
# Windows:
.\.venv\Scripts\activate
pip install -r requirements.txt
pip install -e ../../packages/py-cleaner

# Put sample input (already under storage/.../INPUT.json) then:
apify run

Without Apify CLI you can still smoke-test the pipeline:

$python scripts/smoke_local.py

Deploy (apify push)

Apify uploads only apps/apify-actor. The shared package is vendored at vendor/py-cleaner.

After editing packages/py-cleaner, refresh the vendor copy before push:

# from repo root
Copy-Item -Recurse -Force packages\py-cleaner apps\apify-actor\vendor\py-cleaner
Remove-Item -Recurse -Force apps\apify-actor\vendor\py-cleaner\py_cleaner.egg-info -ErrorAction SilentlyContinue
cd apps\apify-actor
apify push

Monetization

Price this Actor on the Apify Store (rental and/or compute). No Stripe inside the Actor.

Chrome extension

Unchanged. This package does not modify apps/extension.