BotDetect is a CAPTCHA library that sites install on their own servers, with versions for .NET, .NET Core, Java and PHP. Nothing goes through a cloud service: the server draws a distorted-text image (plus an audio version), keeps the answer in the session and checks the visitor's input itself. There is no external verification API to talk to.
How to identifyLink to How to identify
- Images served by
CaptchaImageHandler.ashxon ASP.NET, or a comparable library endpoint - Fields such as
CaptchaCode,CaptchaInputTextorCaptchaId - A "BotDetect" watermark on the free edition
- Warped text over lines, dots and a coloured background
- An audio button that reads the code aloud
- A hidden
CaptchaIdfield naming the widget instance
How it worksLink to How it works
- The server control picks a random code and draws it as an image.
- It saves the code in session state under the
CaptchaId. - The visitor types what they see and the server compares it to the stored code.
- A wrong answer burns the code, so a new image is needed.
- Codes time out after a configurable delay (20 minutes by default) and allow a single validation attempt.
- The audio version speaks the same code, which makes speech-to-text a valid route.
Free & open-source bypass methodsLink to Free & open-source bypass methods
| Tool | What it does | Success Rate | Link |
|---|---|---|---|
| ddddocr | Good out-of-the-box reader for warped text, no training needed | 85–97% | sml2h3/ddddocr |
| Tesseract + preprocessing | Classic OCR after thresholding, deskewing and a character whitelist | 60–85% | tesseract-ocr/tesseract |
| EasyOCR / PaddleOCR | Handles noisy or mixed-font images better | 75–92% | JaidedAI/EasyOCR · PaddlePaddle/PaddleOCR |
| Whisper | Transcribes the audio version | 70–90% | openai/whisper |
| captcha_trainer | Trains a model on one site's BotDetect style | 95–99% (trained) | kerlomz/captcha_trainer |
Practical approachLink to Practical approach
- Fetch the image with the same session cookies the form will use.
- Clean it up: convert to grayscale, raise contrast, remove noise and crop to the text.
- Run ddddocr or Tesseract, limited to the characters the site actually uses.
- Send the answer with the matching
CaptchaId, in the same session. - When audio is offered, download it and let Whisper transcribe it.
- After a miss, fetch a new image, because the previous code is no longer valid.
Practical tipsLink to Practical tips
- Each site configures its own image style, so tune the preprocessing per target.
- The free edition's "BotDetect" watermark can confuse OCR. Crop it out when you can.
- Codes time out, so don't let them sit before submitting.
- The answer lives in the server session: lose the session cookie and you lose the answer.
- Sites often restrict the character set. Matching it with a whitelist gives a big accuracy boost.
- Some sites turn the audio version off or rate-limit it.