Metadata calculator
What the four numbers mean
| Metric | Meaning | How to use it |
|---|---|---|
| Total metadata | Sum of every removable block, in bytes | Over 5% means the file is carrying a lot |
| Share by category | How much C2PA / XMP / EXIF / PNG text / IPTC each take | Find the dominant cause and decide which switches to turn off |
| Upload risk score | Weighted by platform reading priority, not by size | A high score means several triggers stacked up |
| Post-clean estimate | File size minus removable bytes | An estimate only; the real value depends on the rebuilt structure |
How the risk score is calculated
Platforms do not rank an AI label by bytes, they rank it by block type. So the weights here are per type as well: C2PA is highest (most platforms read it first), XMP and PNG text chunks next, IPTC and EXIF Software after that, and pure camera EXIF or GPS only affects privacy, barely touching the AI decision.
- HIGH — C2PA content credentials, AI tags in XMP, IPTC AI attribution. These three drive Instagram, Facebook, Pinterest and TikTok.
- MEDIUM — an EXIF Software field naming an AI tool, or PNG text chunks carrying workflow data. Indirect, but readable.
- LOW — the remaining camera EXIF (time, GPS, lens). A privacy matter, generally not an AI trigger.
Typical metadata sizes
Knowing the usual values helps spot a file that is unusually heavy:
| Source | Typical metadata | Mostly made of |
|---|---|---|
| Straight from a phone camera | 5–15 KB | EXIF camera parameters |
| Midjourney PNG | 10–50 KB | XMP parameters plus PNG text chunks |
| DALL·E / ChatGPT | 5–20 KB | C2PA signature |
| Photoshop (with AI edits) | 15–60 KB | C2PA plus XMP |
| Stable Diffusion / ComfyUI | 20–100 KB+ | The full workflow in PNG text chunks |
Why the size itself matters
Metadata can add more than 10% to a file. For a creator uploading dozens of images it directly affects upload time and storage; for portfolio sites and email attachments it eats bandwidth and size limits. Version management is worse still — two JPEGs that look identical may differ by tens of KB because one kept IPTC keywords and C2PA assertions.
So the recommendation is simple: clean before distributing, keep a master copy locally with the full metadata. Use the master when you need provenance, use the cleaned copy when you upload.