Repository navigation
Blockerflow for Mac 1.1.2
Important
macOS may say "Blockerflow is damaged and can't be opened"
The app is not damaged. Blockerflow for Mac is distributed without an Apple Developer licence, so it is not signed or notarized by Apple, and macOS blocks apps it cannot verify.
After you drag Blockerflow into Applications, open Terminal and run this one command:
xattr -cr /Applications/Blockerflow.app
Then open Blockerflow again. You only need to do this once per install. You can also try right click, then Open, then Open again.
Support: help me get an Apple Developer ID (€99 a year) to sign Blockerflow for macOS and iOS, so it opens without warnings. Donate via PayPal
What's new in 1.1.2
- Block adult images no longer blocks ordinary photos as often. A photo the model was unsure about (swimwear, gym, leggings) could be blocked after a few seconds just by staying on screen, because each look added the same weak score again. Now a look must be reasonably sure to count at all. Pictures the model is sure of are still blocked at once, still pictures on the second look, and videos as before.
- Tested on 50 ordinary photos of people in swimwear, sportswear and evening wear, each opened on its own and inside a web page: 6 of the 100 views were blocked before, 3 now.
- Still known: the model can be confidently wrong, and those pictures are still blocked. For example, it can read a shirtless man's chest as a woman's.
- Block suggestive images (lingerie and swimwear) works as before.
Everything from 1.1.1 is included: Instagram, Facebook, YouTube and Snapchat are never decrypted, and Reels, Shorts and Spotlight are blocked from the browser's address bar.
After updating: macOS ties Screen Recording to this exact build, so Block adult images will wait for permission again. In System Settings › Privacy & Security › Screen & System Audio Recording, select Blockerflow, remove it with −, add it again with +, then quit and reopen Blockerflow.
Not tested yet: real explicit content with the new rule, Safari and Arc with the address-bar blocking, Intel Macs.
Licence
The image check uses NudeNet 320n (https://github.com/notAI-tech/NudeNet), trained with Ultralytics YOLOv8, both AGPL-3.0, and ONNX Runtime (MIT). The AGPL-3.0 text: https://www.gnu.org/licenses/agpl-3.0.html.