4chan captcha solver userscript (original by AUTOMATIC1111)
Automatically fills in the captcha when the captcha loads. If you get a slider captcha, the slider will be moved into position automatically. If the automatic slider detection is incorrect, you can move it yourself and press the "Solve" button to redo the solution with the new position.
No error handling in place - if solving fails, you can find the reason in js console.
I can't upload it to any userscript js site because of size. Most of the size is model weights.
IF YOU'RE ON FIREFOX/LIBREWOLF AND YOU HAVE privacy.resistFingerprinting ENABLED IN about:config YOU NEED TO GRANT CANVAS PERMISSION TO 4CHAN.ORG OR THE SCRIPT WILL NOT WORK
- New model - sligthly larger but more precise
- Now handles the new slider puzzle used in red boards.
- New model trained by Yukariin.
- Yet another more accurate model trained by Yukariin.
- Added an alternative script using Tensorflow.js' WebAssembly backend which may replace the main script in the future. On all of my devices it solves the captcha about 150% faster compared to the CPU backend.
- And if you're reading this on the 25th of December: I wish you a Merry Christmas!
- New model trained by moffatman.
- New model trained by Yukariin.
- New slider solver code
stolentaken from the 4chan Mass Reply browser extension, which offers a much higher success rate compared to the old code. - The script will now ask for Canvas permission on Firefox-based browsers with privacy.resistFingerprinting enabled.
- Built a new model trained on 4chan's new captcha with 5-6 characters, specks, lines, and more extra garbage scattered everywhere.
- Disabled some of tooltip debug functionality I imagine no one used.
- Changed the algorithm to not produce empty result wjhen the length of captcha output is not what we expected.
- Tested on Firefox with Violentmonkey.
- Built a new model from scratch, without using a pretrained one, on 50k synthetic samples (as opposed to previous one trained on 400 images + their augmentations).
