For scans, photo archives and publishing

Check TIFF image metadata and AI signals

Upload an original, single-page TIFF and read its AI declarations, creation records and image measurements straight from the original bytes, with no conversion. The report keeps a scan’s history separate from the file format: a .tif or .tiff extension says nothing about who made the picture, and the records inside do.

1. Check the original TIFF before making an export

  1. Keep the source file. Work from the scanner output, camera-processing export or original supplied by the publisher. Record where it came from and whether it has already been edited.
  2. Open the AI image checker. In the Image tab, choose Check my image now and select one supported .tif or .tiff. Selecting the file starts the check.
  3. Read the result with its basis. Open How we checked, then review What we found and Checks run. Distinguish a completed check from an unsupported or unavailable one.
  4. Keep a report beside the file. Use Save report or Copy report and record the source, check date and any subsequent export. A later PNG or JPEG is a different file to review.

The TIFF report can show preview unavailable. Open the original in a compatible image application for the visual review; the file check runs on the original bytes regardless of the browser thumbnail.

Uploads are private by default and processed on isGenAI’s server. The main checker is separate from the browser-local editing tools. Read the upload privacy details before including material that needs special handling.

2. Separate a scan’s history from the picture’s origin

A scanner records the material placed in front of it. A photograph of a real scene, a printed AI-generated picture and an edited composite can all become TIFF scans. A scanner name describes a stage in that process; it does not settle what happened before the scan.

For an archive or publication, preserve three useful copies when they exist:

  • The supplied original or scan: the starting point for the records you can inspect.
  • The edited master: the copy after retouching, restoration, color correction or other changes. Ask which edits used generative tools when that distinction matters.
  • The publication export: the version that actually appears in the article, catalogue or website.

Example: a scanned print is restored with a generative editing tool and exported for a catalogue. “Scanned from a print” and “AI was used during restoration” can both be true. Record the stage supported by each finding instead of applying one description to the whole history.

3. Read the saved records without overinterpreting them

TIFF-related metadata can describe equipment, software, dimensions and resolution. Adobe’s TIFF metadata reference defines fields such as Make, Model, Software and XResolution. These fields describe different things: a software name is not a signature, and print resolution is not pixel width.

isGenAI reads EXIF, IPTC and XMP content for recognized AI declarations and validates C2PA records, focused on the signals that establish origin rather than an inventory of every TIFF tag. Each finding appears exactly when the file carries a readable record for it.

What a TIFF finding can support
FindingWhat to reviewWhat it does not establish
AI declaration in metadataThe actual declaration and the field or mechanism reported.An uneditable history or a calibrated probability of AI authorship.
Content CredentialsRecorded actions, signature validation and signer trust separately.That every depicted event is true, or that every edit is non-generative.
A matching copyWhether the match is exact or approximate and what the earlier check found.A search of every copy online or a complete publication history.
No AI detectedWhich checks completed and which records were available.Human authorship or proof that no AI was used before this copy existed.

Read more about signed creation records and why identical pixels can have different metadata. Editable declarations and validated signatures provide different kinds of evidence.

Try it: identical TIFF pixels, different file records

These three downloadable files contain the same 32 × 24 opaque pixels: the left half is black and the right half is white. We drew the image in code. It was not generated by an AI model or captured by a camera.

The third file deliberately contains an editable XMP declaration, DigitalSourceType = trainedAlgorithmicMedia. We added it to demonstrate how a file can declare AI use even though these particular pixels were drawn in code. It is a test assertion, not the image’s verified history. None of the files contains a signed C2PA manifest or a saved generation prompt.

  1. Download a file and upload it to the image checker.
  2. Compare the image measurements: brightness mean 127.5, near-black pixels 50%, near-white pixels 50% and transparency 0% in each file.
  3. Compare the metadata findings. The first two have no added AI declaration. The third produces a declarative_metadata finding with the mechanism trainedAlgorithmicMedia.
  4. Keep the report and repeat with the other files. Their SHA-256 hashes differ even though their decoded pixels match.

In the controlled offline check on 24 September 2026, both untagged files returned ORIGIN_UNRESOLVED, and the local C2PA check reported no manifest. The added declaration was detected in the third. Live combined results can also reflect matching history; compare the specific metadata finding and measurements when repeating this example.

Read the method, SHA-256 hashes and CC0 reuse terms. Download the standalone reproduction script and run node reproduce.mjs ./tiff-examples with Node.js. It uses no dependencies or network calls and refuses to overwrite files. This is a reproducible file-behavior example, not a classifier-accuracy benchmark.

4. Use image measurements to guide a visual review

A supported TIFF can include dimensions, megapixels, brightness, tonal spread, near-black and near-white pixel percentages, transparency and neighboring-pixel differences. Measurements use an oriented sample that fits within 512 × 512 pixels. They describe that sample and do not alter the origin verdict.

In a scanned letter, a high proportion of near-white pixels describes the paper background. A low edge difference reflects a smooth subject or a small sampled preview. Inspect the original at an appropriate zoom before deciding that detail is missing. These measurements describe the sample; OCR accuracy, print-color proofing and archive acceptance are separate tests with their own tools.

Pixel dimensions and print density answer separate questions. A 3000 × 2400 image placed at 300 pixels per inch corresponds to 10 × 8 inches: divide each pixel dimension by the intended density. That calculation does not establish sharpness or authorship. Use the image diagnostics reference if you need the response fields and unavailable states.

Supported TIFF files and limits

  • Format: classic TIFF in either byte order, using .tif or .tiff. BigTIFF is not supported.
  • Pages: one page per file, without subimages or image pyramids. Multipage TIFF and files with SubIFDs are rejected rather than treated as a complete review of every stored image.
  • Upload size: up to 20 MiB, or 20,971,520 bytes, for the file.
  • Decoded size: no more than 40 million pixels. A file with efficient compression can still exceed this limit when decoded.
  • Image-model availability: the statistical image-model adapter reports TIFF as unsupported media, and the checker never silently converts your file into a model input.
  • Other tools: TIFF support lives in the main image checker. The saved-prompt extractor and the 25 browser image tools keep their own format lists.

For a concrete size check, a US Letter scan at 600 pixels per inch is 5100 × 6600 pixels, or 33.66 million pixels. At 1200 pixels per inch, it is 10,200 × 13,200, or 134.64 million pixels, exceeding the decoded limit even if its compressed file is small. File size and pixel count must both fit.

If a file is rejected, keep the original and create a separate single-page copy within the limits using an application that supports your TIFF. Record any reduction or conversion. Renaming an extension never converts the contents, and a damaged or unsupported TIFF encoding is rejected rather than guessed at.

TIFF checking questions

Can a TIFF image be AI-generated?

Yes. TIFF is a file format, not a record of how the picture originated. A scan, a camera photograph, an AI-generated image or a mixture of edited material can all be saved as TIFF. isGenAI reads the records inside the particular file and its source, so the extension never stands in for an authorship verdict.

Does scanning an AI image make it a human-made image?

No. Scanning creates a new digital copy of the printed material and says nothing about how that material was originally made. The scan carries scanner or editing records while the original generation records are absent. Keep the printed source, supplied original and scan history together and the report tells you which stage each finding belongs to.

Can isGenAI check every page in a multipage TIFF?

The TIFF workflow checks one page per file: single-page files with no subimages or image pyramids. Multipage TIFF and BigTIFF are rejected up front rather than half-checked. Export the specific page you need from a compatible desktop application and check that file, keeping the original and the page number. An exported page is a new copy with its own metadata.

Will I receive an AI percentage for my TIFF?

No, and that is deliberate: the statistical image classifier reports TIFF as unsupported media, so you never get a number that was not measured. What you do get is the full file-evidence path: metadata, creation records, matching copies and eight image-quality measurements. Brightness, transparency and edge values describe the pixels; origin comes from the records.

Should I convert my TIFF to PNG before checking it?

Check the original TIFF first. Converting changes or removes saved records, and the main checker reads TIFF directly. If you need a PNG or JPEG for a publishing system, create it as a separate copy and record how it was exported. The browser image tools and the saved-prompt extractor have their own supported formats; TIFF support lives in the main checker.

Sources and scope

The Adobe TIFF metadata reference describes field meanings. Sharp’s input and page controls and metadata documentation describe the image reader used by isGenAI. The upload restrictions above are isGenAI’s own limits; broader format support in a library does not mean every variant is accepted here.

This guide is a file-review workflow with a reproducible example, distinct from a TIFF detection-accuracy benchmark. Report an incorrect instruction or unsupported example and we check it against the implementation.