Shumei (数美) is a Chinese risk-control company, and its CAPTCHA shows up on many sign-up, login and promotion pages. The widget can serve slide, click, rotate and word challenges, and it is wired directly into Shumei's device fingerprinting and risk scoring.
How to identifyLink to How to identify
- Traffic to
captcha.fengkongcloud.cn - The endpoints
/ca/v1/conf,/ca/v1/registerand/ca/v2/fverify - Parameters such as
organization,appId,model,rversionandsdkver - CSS classes like
smCaptcha,sm-buttonandsm-slider - The prompt "请完成安全验证" ("Please complete the security verification")
- A slider, a rotating picture or a click-in-order task
Common challenge typesLink to Common challenge types
- Slide: drag the piece into the gap.
- Rotate: turn the picture until it sits at the right angle.
- Click: tap characters or icons in the requested order.
- Word: type the characters shown.
How it worksLink to How it works
- The page calls
/ca/v1/confto fetch the settings for itsorganizationandappId. /ca/v1/registerreturns the challenge assets: backgroundbg, foregroundfgand a request IDrid.- Once the visitor answers, the SDK posts the answer, the
ridand a set of risk signals to/ca/v2/fverify. - A pass returns a token or result that the site's backend then checks.
- Each answer belongs to one
ridand one session, works once and expires fast.
Free & open-source bypass methodsLink to Free & open-source bypass methods
| Tool | What it does | Success Rate | Link |
|---|---|---|---|
| OpenCV template matching | Finds the slide gap from bg and fg |
70–90% | opencv/opencv-python |
| PuzzleSolver | Slider solver with eased drag paths | 70–90% | vsmutok/Puzzle-Captcha-Solver |
| ddddocr | Gap detection plus simple OCR and click helpers | 65–85% | sml2h3/ddddocr |
| YOLOv8 | Spots the targets in click and icon tasks | 80–95% (trained) | ultralytics/ultralytics |
| DrissionPage / Playwright | Grabs the assets and answers inside the same session | 70–90% | g1879/DrissionPage · microsoft/playwright-python |
Practical approachLink to Practical approach
- Load the page and intercept the
/ca/v1/registerresponse so you work with the originalbgandfgimages. - Run Canny edge detection and
matchTemplateto locate the gap. - Drag with acceleration, a small overshoot and a little vertical wobble.
- Wait for the success callback, then send the form with the token it returns.
- Rotate tasks: estimate the angle from the picture, then drag along a curve.
- Click tasks: use an object detector or a maintained solver.
Practical tipsLink to Practical tips
- A
ridis good for one attempt. After a failure, start a new challenge. - Shumei fingerprints the device, so a real browser with a stable profile is safer than raw HTTP calls.
- When the site targets China, use a Chinese locale and an IP from the region.
- The gap can be faint. Edge detection finds it more reliably than comparing raw pixels.
- Keep failing and Shumei will switch to a harder challenge type, or block the session outright.