Challenge fundamentals
reCAPTCHA image grids: 3×3, 4×4, and dynamic tiles
reCAPTCHA image grids: 3×3, 4×4, and dynamic tiles. Learn the inputs, limits, and result handling for easycaptcha reCAPTCHA recognition.
By the easycaptcha team · Published

The easycaptcha recognition API uses three grid names. Send imageselect for a full 3×3 image, tileselect for a full 4×4 image, and dynamic for separate replacement tiles. The native API expects these names rather than numeric aliases.
Map the returned selections
A full 3×3 result contains nine booleans; a 4×4 result contains sixteen. Read both left to right, top to bottom. A dynamic result has one boolean per submitted image, in submission order. Keep a position map when replacement images come from nonadjacent cells.
Keep recognition separate from verification
A true value means select that image. It does not mean the website accepted the challenge or completed the form. Apply selections to the current images and observe the next state before recording an outcome.
Choose a mode from the actual image layout
A grid mode describes the image you send, not just the number of objects you expect to find. A 3×3 image can contain no matching cells, several matching cells, or an object spanning cell boundaries. Those possibilities do not change the request from imageselect into another mode. Your integration should decide the layout before constructing the request and preserve that decision until the result is applied.
Do not infer the mode from the length of a previous response. A reset can replace a nine-cell challenge with a different layout. Capture the current instruction and image together, then validate the expected answer length against that request. This catches a class of bugs where the API returns a valid answer but the application interprets it using old geometry.
Compare all three request shapes
| Mode | Image input | Result length | Position reference |
|---|---|---|---|
imageselect |
One complete 3×3 image | 9 | Full-grid cell order |
tileselect |
One complete 4×4 image | 16 | Full-grid cell order |
dynamic |
1–9 replacement images | Number of submitted images | Input array order |
The distinction between the last two columns matters. Nine images in a dynamic batch do not make it a full-grid request: they are nine separate image inputs. Conversely, a full image containing nine cells is still one image_data entry. Validate both the number of entries and what each entry represents before sending it.
Walk through a replacement example
Suppose the initial nine-cell result selects the top-left and center cells. Apply those decisions to the original grid and wait for those two cells to finish replacing. If the replacement images are submitted in that order, a dynamic result of [false, true] means leave the new top-left image alone and select the new center image. It says nothing about the other seven cells.
If selecting the center creates another replacement, collect that image for the next dynamic request. Keep the instruction associated with the sequence. A changed instruction starts a different recognition context, so decisions from the previous question should not be carried over merely because the cell positions are unchanged.
Estimate work by requests, not cells
For easycaptcha reCAPTCHA recognition, a full-grid request is one recognition request, and each dynamic request is another. A 4×4 grid does not become sixteen billed recognitions because the answer contains sixteen booleans. Similarly, a coherent dynamic batch containing several replacements remains one request.
When estimating usage, record how often workflows require replacements and how many batches they produce. Two users can see the same initial layout and consume different request counts. The replacement-round metrics guide explains how to track that difference without confusing image count, request count, and completed challenges.
Validate a grid request before submitting paid work
Build local validation around the mode you selected. A full-grid request should have exactly one image entry, an instruction that belongs to that image, and an expected result length recorded with the task context. A dynamic request should have between one and nine image entries and the same number of stored position records. These checks catch construction mistakes before they become paid API requests or wrong-cell clicks.
Keep validation of media separate from validation of geometry. A correctly encoded JPEG can still be the wrong crop, and a correctly cropped grid can still be encoded as a filename rather than image bytes. During development, decode a prepared data URI and inspect its contents locally. Then exercise result mapping with a known boolean fixture instead of using live recognition to test indexing arithmetic.
When a new layout appears, fail explicitly if the integration does not support it. Guessing based on the most recently seen layout can produce valid-looking requests with unusable results. Retain the selected mode in your task record so later investigation can compare what the browser displayed, what the client submitted, and how the answer was interpreted.
Frequently asked questions
Does easycaptcha return a reCAPTCHA token?
No. reCAPTCHA recognition returns boolean image selections, not a token.
Can I send nine separate images as imageselect?
No. imageselect takes one complete 3×3 image. Separate replacement images belong in a dynamic request.
How should I estimate recognition usage?
Count the initial full-grid request and every dynamic request. Cell count and visible challenge count are different units.