protecting your timeline from the egregore since 2026
Some people find that a significant percentage of their timeline consists of accounts using aesthetically identical chibi avatars posting aesthetically identical content. This extension addresses that.
A bundled ONNX classifier scans avatars as you scroll. When it spots a match, you pick what happens:
- Hide — collapsed behind a click-to-reveal row.
- Fade — visible but at half opacity.
- Debug — borders and confidence scores on every post.
- Off — does nothing.
The popup tracks session stats (posts scanned, match rate, last sighting), keeps a list of detected accounts you can exempt individually, and collects avatar data you can export for offline labeling.
Everything runs locally. No server calls, no telemetry, nothing leaves your browser unless you explicitly export it.
There is no Chrome Web Store release. Install from GitHub Releases instead:
- Download the latest
milady-shrinkifier-vX.Y.Z-unpacked.zipfrom Releases. - Unzip it somewhere permanent on disk.
- Open
chrome://extensions. - Enable
Developer mode. - Click
Load unpacked. - Select the unzipped folder.
- Development, debugging, and training workflow commands live in
DEVELOPMENT.md. - Runtime model artifacts live in
public/models/andpublic/generated/. - Training runs, labels, downloaded avatars, and dataset manifests live under ignored
cache/. - The review app supports both individual labeling and 9-up batch labeling.
- The extension runtime is ONNX-only.