Start with provenance and context
A forensic assessment is most useful when the input’s history is known. Keep the supplied file, record where it came from and obtain the original camera file when available. A social-media download can differ from its source in dimensions, compression and metadata before any analysis takes place.
Read and accept the ImgProcess analysis disclaimer before running the tool. Record the case information requested by the interface. A report should be connected to the actual input and context, rather than treated as an interchangeable certificate for every copy of the same scene.
Separate a label from supporting evidence
An assessment label summarizes the tool’s available signals. Confidence describes the reported assessment or detector response; evidence strength and human review address different limitations. Inspect the individual sections instead of interpreting one large percentage as the entire explanation.
A confidence score is not the percentage of pixels altered. A camera photo with a small generative edit can produce a strong synthetic signal, while a heavily compressed copy can weaken other checks. The score alone cannot tell you exactly how much of a face or background was changed.
Treat unavailable checks as missing evidence
Some report sections depend on configured external detectors or specialized engines. If a detector is unavailable, that status must remain visible. A missing camera-verification engine is not evidence that the image failed physical camera verification; it means the check was not performed.
Likewise, metadata absence is not a conclusive synthetic label. File handling can strip metadata, and editable fields do not establish reliable provenance on their own. Corroborate the report with originals, contextual records and a suitable human review.
Know when the selected image leaves the device
Local checks can run on the device. Connected external detection can transmit the selected image to a configured service. Read the processing acknowledgment and the owner’s published privacy policy before using that feature for sensitive material.
Do not upload a private evidentiary image merely because an unrelated compression tool is local. Per-tool processing information matters. A local metadata inspection and an activated external detector have different data flows even though they appear within the same website.
Download a report with its qualifications
The tool requires a second acknowledgment before report download. Review the thumbnail, case details, unavailable checks and disclaimer in the PDF. Keep the exact input alongside the report so that a later reviewer can identify which file was assessed.
Do not use the report as a legal or professional conclusion beyond its stated scope. If the decision affects a person or an important event, seek the relevant specialist review and additional evidence. The useful role of the tool is to organize available signals and limits, not to replace missing provenance with certainty.
Keep unavailable checks visible in your conclusion
A report can have useful local observations while a connected engine is unavailable. State that limitation explicitly; missing checks do not constitute a negative test result. Preserve the input and the report together so a later reviewer knows which bytes were assessed.
A hash and browser-generated audit entries are useful records but not an independently verified identity or complete chain-of-custody system. Important cases require the appropriate handling procedure and qualified review beyond this browser workflow.
