What is screenshot sanitizing?
Screenshot sanitizing means finding and hiding sensitive text, such as emails, usernames and API keys, before you post or send a screenshot. This tool runs on-device text recognition (OCR), suggests matches for you to check, and lets you draw extra boxes for anything the scan missed. Nothing is redacted automatically.
Why review is mandatory, not optional
Automatic text detection is a starting point, not a guarantee. OCR can misread low-contrast or stylized text, and a pattern match can flag something harmless or miss a real secret written in an unusual format. That is why every auto-detected match starts as pending: you accept or reject it yourself, and the download button stays locked until you confirm you reviewed the regions. This mirrors general advice on handling sensitive data, such as the OWASP guidance on protecting sensitive data.
What the scan looks for
| Pattern | Example |
|---|---|
| Email address | jane.doe@company.com |
| @handle | @teamlead_92 |
| Prefixed API key | sk-live_…, AKIA…, ghp_… |
| Generic long token | A 20+ character mixed-case/number string with no spaces |
How to sanitize a screenshot in four steps
- Add your screenshot. PNG keeps UI text sharpest.
- Click Auto-Detect. The first run downloads a small OCR model, and later scans are faster.
- Accept or reject each suggestion, and add manual boxes for faces, logos, account numbers or anything the scan missed.
- Pick a redaction style, confirm you reviewed every region, and download the PNG.
Choosing a redaction style
A solid bar fully removes the pixel data underneath and is the safest choice for real secrets like passwords or live API keys. Blur and pixelate look lighter but can sometimes be partly reversed if the setting is too weak, especially on short strings against a plain background, a risk documented in general research on de-blurring text recovery. When in doubt, use the solid bar.
What this tool cannot catch
- Sensitive information with no matching text pattern: real names, phone numbers, physical addresses, account balances.
- Text inside a logo, icon or embedded image rather than rendered UI text.
- Faces or other visual identifiers. Use a manual box for these.
- Secrets split across two OCR words, or written in an unusual format the patterns do not match.
Always look over the whole screenshot yourself before sharing it, even after a clean scan.
Related tools
Shrink your sanitized screenshot with the Image Compressor, or check a link for phishing red flags with the Link Safety Checker before you click it. Browse more free utilities in the ImageFix tools hub.
Frequently asked questions
What does the auto-detect feature find?
It runs on-device text recognition (OCR) on your screenshot, then checks each piece of text against patterns for email addresses, @handles and API-key-like strings such as long mixed-case tokens or keys prefixed with sk-, AKIA, ghp_ and similar.
Why do I have to review each detected region?
OCR and pattern matching both make mistakes. They can miss stylized or low-contrast text, and they can flag harmless text that merely looks like a key. This tool never redacts anything automatically — you accept or reject every suggestion before anything is applied.
Does auto-detect catch everything sensitive?
No. It cannot read names, phone numbers, account balances, physical addresses or anything in a screenshot without matching text, and it cannot find sensitive information inside logos, QR codes or embedded images. Always scan the full screenshot yourself and add manual boxes for anything the scan missed.
Is my screenshot uploaded anywhere?
No. The image and the text recognition both run locally in your browser. The OCR engine downloads its model files from a public host the first time you use it, and your browser caches them.
What is the difference between the black bar, blur and pixelate styles?
A solid bar fully hides the content and cannot be reversed. Blur and pixelate are faster to apply visually but can sometimes be partially reversed with specialized tools if the blur or pixel size is too light, so use a solid bar for anything truly sensitive.
Can I redact more than text, like a face or a logo?
Yes. Switch to manual mode and draw a box over any region, text or not. Auto-detect only helps you find text; manual boxes work on anything.