Skip to main content

CAPTCHA guide

TikTok CAPTCHA

ByteDance's slide, rotate, 3D and icon puzzles, bound to the device and session.

Difficulty: High
Risk engineImage puzzle
Exploded concept drawing of how the TikTok challenge is built

ByteDance has its own image-puzzle CAPTCHA, shared by TikTok, CapCut, Lark and Douyin. It comes in four looks: slide, rotate ("whirl"), 3D object matching and icon selection. Each challenge is tied to the device ID, the cookies and the session IP, so an answer produced somewhere else is worthless. It has to be applied inside the same session.

How to identifyLink to How to identify

  • A wide background with a cut-out and a separate piece (slide)
  • Two concentric round images (whirl / rotate)
  • Several 3D-rendered objects and a "find the matching pair" prompt
  • A picture plus a text instruction to click the matching icons
  • Requests carrying device_id and iid to TikTok/ByteDance anti-bot endpoints
  • A drag or click task that comes back after a failure

Common variantsLink to Common variants

Variant Task Answer format
Slide Move the piece into the gap X distance or ratio
Whirl Turn the inner disc until it lines up with the ring Angle or drag distance
3D Click the two identical 3D objects Two points
Icon Click the icons named in the prompt, in order A list of points

How it worksLink to How it works

  1. A challenge is requested during login, sign-up or any action the risk engine flags.
  2. The server sends back one or more images plus the challenge type.
  3. While the visitor drags, rotates or clicks, the client logs the path and its timing.
  4. The answer is submitted along with device and session identifiers.
  5. TikTok checks both the answer and the behaviour. A wrong position or an unnatural path fails.
  6. A failure brings a new challenge. There is no long-lived token to fall back on.

Free & open-source bypass methodsLink to Free & open-source bypass methods

Tool What it does Success Rate Link
OpenCV matchTemplate Finds the slide gap after Sobel or Canny preprocessing 75–92% opencv/opencv-python
PuzzleSolver Slide solver with natural drag paths 70–90% vsmutok/Puzzle-Captcha-Solver
xtekky/TikTok-Captcha-Solver A research solver for the slide puzzle experimental xtekky/TikTok-Captcha-Solver
YOLOv8 / CLIP Detects 3D objects and matches icons 70–90% (trained) ultralytics/ultralytics · openai/CLIP
DrissionPage / Playwright Performs the drag or click inside the real session 60–85% g1879/DrissionPage · microsoft/playwright-python

Practical approachLink to Practical approach

  1. Take the original challenge images from the network response, keeping their size and transparency.
  2. Recognise the variant: wide background plus piece means slide, concentric circles mean whirl, several 3D objects mean match, a prompt with icons means icon.
  3. Slide: run Sobel or Canny with matchTemplate to find the gap, then drag along an eased, jittery path.
  4. Whirl: estimate the angle and drag along the circular track.
  5. 3D and icon: use a detector or classifier and click the targets in the order requested.
  6. Do all of it in the same browser session and submit at once.

Practical tipsLink to Practical tips

  • An answer generated on another device or IP is refused, because the challenge belongs to the session.
  • The drag path is analysed: a straight move at constant speed fails even when it ends in the right place.
  • Work on the original image bytes rather than a resized screenshot, or the geometry will be off.
  • After a failure a new challenge appears. Try again with a different path.
  • TikTok switches variants often, so keep solvers ready for slide, whirl, 3D and icon.
  • Mobile and desktop apps may serve different sets of challenges.

Skip the solver: let BrowserSolver handle TikTok

Website AntiBot runs a real browser session with matching TLS, headers, cookies and IP, clears the challenge, and returns the page. You send a URL, you get HTML back.

For education and authorized testing only. Respect the law, privacy, and the terms of the sites you work with. Difficulty labels are editorial estimates. Adapted from the CC0 community field guide on GitHub.