Temu runs its own in-page CAPTCHA, not GeeTest or reCAPTCHA, and rotates through many challenge types: slide puzzle, tile swap, semantic click, arced slide, 3×3 grids, shapes and items, and line alignment. Its pages are heavily guarded, so a clean browser fingerprint and a residential IP matter as much as solving the puzzle.
Challenge typesLink to Challenge types
- Slide: move a piece into the gap
- Arced slide: the piece follows an unpredictable curve towards one of two possible targets
- Puzzle slide: the pieces move once the drag has started
- Swap: two tiles have been switched and need putting back
- Semantic click: instructions such as "click the smaller green cylinders"
- Three-by-three: choose the grid cells that match the instruction
- Two image / line: line things up, or pick one of two pictures
How to identifyLink to How to identify
- The challenge is drawn inside the Temu page itself, often in a modal
- A background image with one or more draggable pieces (transparent PNGs)
- A written instruction for click and selection tasks
- Heavy anti-bot measures: fingerprinting, TLS checks and frequent re-challenges
- No sitekey or iframe of the kind reCAPTCHA and hCaptcha use
How it worksLink to How it works
- Temu shows a challenge to sessions it considers risky.
- The client sends the geometry of the answer, a drag distance or click coordinates, to Temu's backend.
- Temu checks it against the session, the IP and the fingerprint.
- A wrong or late answer brings another challenge, often of a harder type.
- Some types move the pieces after the drag starts, which breaks pipelines that compute the answer once and replay it.
Free bypass methodsLink to Free bypass methods
| Method | What it does | Success Rate | Link |
|---|---|---|---|
| OpenCV | Matches gaps and pieces for slide and swap variants | 70-85% | opencv/opencv-python |
| PuzzleSolver | An OpenCV slider pipeline | 70-85% | vsmutok/Puzzle-Captcha-Solver |
| DrissionPage | Real Chromium that keeps the session and performs the drag | 65-85% | g1879/DrissionPage |
| Patchright / Camoufox | Stealth browsers to cope with Temu's fingerprinting | 60-80% | Kaliiiiiiiiii-Vinyzu/patchright-python |
| ddddocr / YOLO | Handles semantic click and grid challenges | 55-80% | sml2h3/ddddocr |
Practical approachLink to Practical approach
- Use a residential IP and a stealth browser profile. Temu challenges datacenter traffic almost at once.
- Work out which variant you are facing from the DOM and the instruction text before picking a solver.
- Slide: capture the background and the piece, locate the gap, then drag along a curved path.
- Arced slide: move the mouse along a curve and allow for the two possible targets.
- Semantic click and 3×3: read the instruction with OCR, then find the targets with object detection.
- Apply the answer in the same session and submit immediately.
Practical tipsLink to Practical tips
- The real obstacle is Temu's anti-bot, not the puzzle: TLS fingerprint, headers and the JS environment are all inspected.
- Some variants shift the pieces mid-drag. If the first try fails, read the DOM again once the drag has begun.
- Avoid brute force, because repeated failures escalate to harder types.
- Keep the same cookies and IP from the challenge through to the submission.
- Image solvers often return top-left corners. Click the centre by adding half the piece's size.