Skip to content
checkaifree

Guides

How to check an image without overclaiming

Follow a concrete file task with downloadable examples, expected output and a failure branch. Start with a local metadata operation, or use the image-source guides when you need to investigate a broader claim.

The checker itself stays on the home page. It reads source clues and, when needed and available, can request a model score. It does not verify C2PA signatures, and it does not publish an overall accuracy. The method page is the short account of that pipeline. The earlier metadata experiment covers eight licensed photographs and two defined local transforms. New file-operation guides provide their own source-defined inputs and observed outputs; none of these examples measures social-platform retention or detection accuracy.

Photo privacy

  1. Read image metadata locally and identify each field's source

    Open a JPEG locally, identify EXIF field IDs and source offsets, export the reading, and distinguish empty metadata from incomplete or damaged input.

  2. Find GPS coordinates stored in a photo

    Read JPEG EXIF GPS degrees, minutes, seconds and hemisphere references locally; check altitude, UTC time, and invalid or missing coordinates.

  3. Remove photo GPS while keeping supported camera fields

    Create a JPEG copy without supported GPS, inspect the exact camera allowlist, and compare the downloaded copy with the original before sharing.

  4. Check photo privacy fields before sharing a copy

    Inspect supported location, author, device and workflow fields locally, preview a report without stored values, clean selected categories and recheck the copy.

  5. Compare image metadata before and after a local export

    Compare an authorized original with a documented Pillow JPEG export, inspect retained, changed, added and missing fields, and save a local difference report.

  6. Export image metadata as JSON or CSV without losing repeated fields

    Select metadata groups, preview sensitive values, preserve duplicate and Unicode fields in JSON, and inspect CSV quoting and spreadsheet formula protection.

Camera and photo metadata

  1. Batch-check photo metadata and keep failed files visible

    Inspect mixed JPEG, PNG and WebP files locally, distinguish same-name inputs with stable IDs, and export a per-file summary without raw metadata values.

  2. Remove metadata from multiple images and verify the ZIP

    Apply a local whole-category metadata policy to a mixed batch, keep failed files visible, and inspect verified JPEG, PNG and WebP copies plus their ZIP manifest.

  3. Find a photo's stored capture time without confusing file dates

    Read capture, digitization, image modification, GPS UTC and browser file modification dates separately, preserve subsecond digits and export a local report.

  4. Understand photo timezones before converting a stored date

    Match EXIF date fields with their offset and subsecond tags, keep absent timezones unknown, and verify an example that crosses midnight when converted to UTC.

  5. Fix sideways photo orientation and verify the display copy

    Apply one of eight EXIF orientations once, check a labeled corner chart, and download a re-encoded PNG or JPEG with explicit size and metadata limits.

  6. Extract and check the real JPEG thumbnail stored in EXIF

    Extract the original IFD1 JPEG thumbnail, check its dimensions and TIFF range, download its bytes, and distinguish a missing thumbnail from a broken offset.

  7. Inspect EXIF, XMP and IPTC in the same image

    Read a known JPEG in three local viewers, compare its deliberately different author and caption declarations, and export each result with its original source context.

  8. Check stored photo author, copyright and keyword fields

    Inventory actual XMP and IPTC creator, rights and keyword values, preserve Unicode and repeats, and export a reviewable handoff without treating declarations as proof of ownership.

  9. Inspect an embedded ICC profile and check whether cleaning preserved it

    Extract ICC bytes from JPEG APP2, PNG iCCP and WebP ICCP, inspect actual header fields, and compare a profile-preserving copy with an explicitly profile-removed copy.

PNG text and parameters

  1. Inspect PNG text chunks, duplicate keywords and decoded values

    Locate tEXt, zTXt and iTXt in a real PNG inventory, read duplicate Author values and Unicode text, and distinguish compression limits from missing parameters.

  2. Find PNG metadata stored after IDAT image data

    Compare the same compressed Comment before and after IDAT, save real chunk offsets, and verify why a scan that stops at the first image-data chunk misses stored text.

  3. Edit one PNG text field without re-encoding its image data

    Select an exact duplicate-keyword chunk, stage a Unicode replacement, create a verified PNG copy, and re-read its changed text, valid CRCs and preserved IDAT bytes.

  4. Read the generation parameters actually stored in a PNG

    Open a known A1111-related original PNG, copy and download its exact parameters field, and distinguish saved settings from a model verdict or missing metadata.

  5. Parse A1111 generation settings without losing the original text

    Separate multiline prompts, seed, steps, CFG, sampler and model fields from saved A1111 text while retaining quoted commas, unknown settings and exact large integers.

  6. Find out what a missing PNG parameters result does and does not mean

    Compare a known parameters PNG with a defined locally saved text-removed copy, separate missing fields from unsupported reads, and choose a practical next file to obtain.

  7. Remove selected prompt, workflow and parameters chunks from a PNG copy

    Delete three specific stored-text occurrences, retain the original PNG, re-read the downloaded copy and verify unchanged IDAT bytes without claiming complete anonymity.

ComfyUI workflows

  1. Extract the workflow and API prompt stored in a ComfyUI PNG

    Read the official SDXL original PNG, download its exact workflow/prompt JSON, compare their actual node counts, and handle a prompt-only or metadata-free file without inventing a graph.

  2. Tell a ComfyUI UI workflow from an API prompt JSON

    Open official and self-authored UI/API JSON, locate nodes and stored connections in each shape, and separate valid unrelated JSON, empty workflows and malformed syntax.

  3. List the model references actually stored in a ComfyUI workflow

    Export known checkpoint, LoRA and VAE literals with their node IDs, keep connected/custom-node values unresolved, and retain the fixed adapter/catalog reference without checking installed files.

  4. List ComfyUI node types with a fixed reference and useful compatibility questions

    Export actual type names and counts, retain the pinned base-node catalog scope, and turn an unknown type into a precise question without guessing an installation URL.

  5. Compare two saved ComfyUI graphs without rounding seeds or matching by guesswork

    Find exact node, input and connection changes; verify a key-order-only control; and keep renumbered nodes unmatched instead of claiming semantic or visual equivalence.

  6. Find and correct a known ComfyUI saved-structure error

    Locate duplicate IDs and a dangling connection in a defined workflow, edit the local text, recheck it, and distinguish errors, warnings, unsupported checks and runtime questions.

  7. Create and preview a ComfyUI support report with private fields omitted by default

    Generate a separate aliased diagnostic document, inspect every exported field, test prompt/path/note opt-ins and download JSON or Markdown without changing the input workflow.

  8. Build a concrete ComfyUI pre-import checklist from a saved file

    Record source and format, static diagnostic scope, model literals, node-version questions and report privacy before deciding what to import elsewhere.

Image files and comparisons

  1. Compare SHA-256 before and after a metadata edit

    Hash a known PNG, replace one exact text field, and check the saved output against the expected digest while its compressed image data remains unchanged.

  2. Exact duplicate files versus decoded pixel matches

    Find differently named byte-identical files, keep same-name variants separate, and compare native-size sRGB RGBA with explicit orientation, alpha and color conditions.

  3. Check an image extension against its real container

    Identify a PNG named .jpg, compare its browser MIME and actual structure, and distinguish renaming, damaged data and genuine conversion.

  4. Prepare an image file for the check you need

    Check actual format, byte size, encoded/display dimensions and known file history, then choose a supported local field, hash, pixel or optional model workflow.

Content Credentials

  1. Missing Content Credentials versus an unreadable manifest

    Use absent, readable unsigned and present-but-unreadable controls to distinguish storage and reading states, then export the actual status without claiming signature verification.

  2. Compare Content Credentials before and after removing their container

    Create a defined metadata-removal copy, re-read its supported store and hash, and retain the original report without claiming a signature-validity transition.

Image-source checking

  1. How to tell if an image is AI-generated

    A file-first checklist: earliest copy, embedded text, Content Credentials, reverse search, then a model score.

  2. Can AI images be detected?

    Why AI detectors disagree, what two public benchmarks actually reported, and how to read a conflict.

  3. Does Instagram, WhatsApp or Facebook strip image metadata?

    Our local screenshot experiment, the IPTC 2019 tests, and what those sources do not measure.

  4. How to check Content Credentials (C2PA)

    Use a validator. This site observes source text and does not verify signatures.

  5. Reverse image search vs AI detector

    Prior publication and generated pixels are different questions. Use the method that matches the claim.

A tool, not another essay

If you have an original PNG and want the prompt or the graph that was saved inside it, open the Stable Diffusion PNG info reader. It runs in the browser, shows tEXt, iTXt and zTXt fields, and does not execute a ComfyUI workflow. The same reader also sits on the Stable Diffusion page beside the general checker.

German readers looking for a native explanation can start with KI-Bilder erkennen, Bild auf Echtheit prüfen or Fake-Bilder erkennen. Spanish readers can start with detectar imágenes generadas por IA, then detector de imágenes IA gratis or verificar si una imagen es real. Italian readers can start with riconoscere immagini IA, then verifica immagini AI or verificare se una foto è falsa. Those pages are written for those queries. Where the job matches, the language tags point at each other. They are not machine translations of this index.