We use cookies.This website uses essential cookies to operate core features. With your consent, we also use analytics cookies to understand traffic and improve the service. For more details, see our .
We use cookies.This website uses essential cookies to operate core features. With your consent, we also use analytics cookies to understand traffic and improve the service. For more details, see our .
If this tool helped you, you can buy us a coffee ☕
Upload an image to check whether it may have been photographed again from a phone, monitor, tablet, or another display, and compare recapture and original-image confidence.
Drop an image here or select a file
, ,
Max 4 MB • Max 1 files
A recaptured image is created when a picture is shown on a phone, monitor, tablet, television, or another medium and then photographed again with a camera. That process can introduce screen-pixel patterns, moiré, reflections, perspective distortion, visible bezels, uneven brightness, or local blur. This tool analyzes those visual signals and returns separate confidence values for recapture and original.
The result addresses a narrow question: whether the submitted file shows characteristics associated with photographing a display. It does not prove that the depicted event is true, that a file has never been edited, or that the submitter owns the image. Recapture screening, manipulation detection, metadata review, and human verification answer different questions.
Select one JPG, JPEG, PNG, or BMP file. The file must be no larger than 4 MB; its shortest side must be at least 15 pixels and its longest side no more than 4096 pixels. Review the preview and start the check. ToolKK encodes the image for this request, sends it to the third-party recognition service, and then displays the two scores and the decision threshold used by the interface.
A model confidence score is not a statistically guaranteed probability of authenticity. The interface uses 0.95 as its default recapture alert threshold. A recapture score at or above that value is shown as “likely recaptured.” A lower score only means that this model did not produce a strong recapture signal under the current threshold; it does not certify the file as original.
Risk tolerance varies by workflow. A large content platform may route only high-scoring cases to manual review, while insurance documents, identity evidence, contracts, or regulated records require the original file, capture history, metadata, business context, and additional checks. Never turn one model response directly into a penalty, denial, or identity decision.
Content moderation: flag material that may have been displayed and photographed to evade duplicate-image controls. Document collection: ask a user for a clearer source when a screen border, glare, or moiré is detected. Commerce and support: screen product, delivery, or after-sales evidence. Claims and risk operations: add a quality signal to a review queue.
The tool is not a standalone forensic examination and should not determine copyright ownership, news truthfulness, cheating, fraud, or personal credit. Those conclusions may require preservation of originals, provenance records, qualified investigators, and a documented appeal or review process.
Heavy compression, sharpening, low-light noise, reflections on glass, printed halftone patterns, or photos of electronic dashboards can resemble screen-recapture characteristics. Preserve the highest-quality source and avoid screenshots, collages, filters, beauty processing, or repeated transcoding before the check. Do not edit an input merely to obtain a preferred score.
Some legitimate business images intentionally photograph a monitor, display board, television, or instrument panel. The model may correctly detect screen recapture in that situation without implying deception. Interpretation must follow the purpose for which the image was requested.
If no result is returned, verify the format, file size, pixel dimensions, and network connection. For a score close to the threshold, obtain the source file and run a second comparison rather than cropping, brightening, or retouching the same copy. If repeated results conflict, rely on human review and the original evidence chain.
Providers may update their models, so an old result and a later result can differ. An audited workflow should record the check time, request reference, threshold, and review decision while retaining no more personal imagery than necessary.
ToolKK does not write uploaded images from this tool to its site disk. The image is transmitted for the current request to a third-party provider, whose handling is governed by its own terms. Do not upload identity numbers, payment-card details, medical records, children’s information, or other unnecessary sensitive data. Where processing is necessary, confirm authorization and minimize or redact the image first.
Automated screening can produce false positives, false negatives, and temporary service errors. Treat the output as a prompt for further review, not as a professional forensic opinion, legal conclusion, or guarantee of authenticity.
Is a screenshot the same as a screen photo? No. A screenshot does not pass through a second camera, although editing and compression can still affect model features. Does a score below the threshold prove originality? No; it only means the alert threshold was not reached. Can I check a batch? This page processes one image per request so that cost and evidence remain easy to trace.
Endpoint fields, supported formats, dimensions, and response names follow the Aliyun Marketplace recapture-recognition documentation. Apply the result together with your organization’s privacy, moderation, and human-review rules.
