How the tools work

Where your files go

Each tool does its work in your browser unless it can’t do the job well there. Today that means:

• AI Image Checker: runs entirely in your browser. Your image is never uploaded.
• AI Doc Editor: runs in your browser and keeps your documents in this browser’s storage. Only writing help, which you start yourself, sends the selected text to an AI model.
• PDF Editor: runs entirely in your browser. Your PDF is opened, edited and saved on your device and never uploaded.
• Image Compressor: uses our server, because server-side encoders give smaller files and don’t drain a phone’s battery. The image is sent over an encrypted connection, held in memory rather than written to disk, and deleted as soon as your browser has the result, or after 30 minutes at the latest.
• Image Resizer: uses our server in the same way, with the same deletion. The preview of the new size is drawn in your browser before anything is uploaded.
• Image Converter and PDF Compressor: use our server in the same way, with the same deletion.

The only things we count are anonymous page views and that a tool was used, never the content of your files or their names. See the privacy policy for the details.

Image Compressor

You choose either a quality level or a target size. With a quality level, the image is re-encoded once at that setting. With a target size, the compressor tries quality levels until it finds the highest one that fits. If even a low quality is too big, it scales the image down and searches again. PNGs get a smaller colour palette first, which suits screenshots and graphics.

Phone photos are turned the right way up, and metadata such as camera details and GPS location is removed unless you choose to keep it. The colour profile always stays, so colours don’t shift. If compressing would make a file bigger, which happens with images that are already well compressed, you get the original back unchanged.

Sizes count 1 KB as 1,024 bytes, the way Windows and most forms do, so a file “under 50 KB” here is under 50 KB everywhere.

Image Resizer

Images are resized with Lanczos resampling, after phone photos are turned the right way up. With the shape kept, the side you type sets the size and the other follows. For an exact size with a different shape, the image is cropped evenly from both sides, padded with a colour, or stretched, as you choose. Sizes in cm, mm or inches become pixels at the DPI you set (an inch is 2.54 cm), and JPEG and PNG files record that DPI.

An optional file-size limit lowers the encoder quality until the file fits but never changes the pixel size you chose; if it can’t fit, you’re told rather than given a smaller image. Metadata and the colour profile are handled as in the compressor.

Image Converter

The converter saves the same picture in the format you choose and never changes its pixel size. PNG loses nothing; JPG, WebP and AVIF use the quality level you pick. Transparent areas become white in JPG, which can’t store them, and you’re told when that happens. A file that’s already in the chosen format comes back unchanged. Metadata and the colour profile are handled as in the compressor.

PDF Compressor

Most of a large PDF’s size is its pictures, so each one is scaled down and saved again as a JPEG. The limit is worked out as if the picture filled the largest page, so a picture placed smaller on the page ends up sharper than the DPI shown, never softer. Text, fonts, links and form fields aren’t touched. With a target size, the compressor tries settings until it finds the best one that fits, and tells you if none does.

Pictures in CMYK or spot colours are left alone so their colours don’t shift, and so are black-and-white scans, which are already compact. If compressing wouldn’t help, you get the original back. Protected PDFs are refused, and a digital signature won’t survive compression.

PDF Editor

The editor shows your PDF with pdf.js and writes the result with pdf-lib, both in your browser. What you add (text, ticks, pictures, signatures, highlights and white-out boxes) is drawn on top of the original pages; the text already in the PDF isn’t changed. White-out only covers what’s under it: the original text stays in the file, so it isn’t a way to remove private information. The first PDF you open is edited in place, so its form fields, links and bookmarks survive; pages added from other PDFs are copied in without their form fields.

Text uses Helvetica when every character fits it and DejaVu Sans otherwise, so Greek, Cyrillic and symbols such as ✓ come out right. Scripts that need special layout, such as Arabic, Hebrew and Hindi, aren’t supported yet; the editor says so rather than saving broken text. Signatures are pictures, not certified digital signatures, and editing a digitally signed PDF makes its signature invalid.

AI Image Checker

The checker reads what the file says about itself. It doesn’t look at the pixels or try to guess. It reads four things:

1. Content Credentials (C2PA). A signed record of how the image was made, read with the official open-source library. The checker verifies the signature, follows the record back through earlier edits and looks the signer up in the public C2PA trust list.
2. The IPTC source type. A standard label for how an image was made; “trained algorithmic media” means generative AI.
3. Generator settings. Some image generators save the prompt and settings in the file.
4. Camera and software fields. EXIF and XMP fields that name the camera or the app that saved the file. An AI tool named there is flagged.

Each finding shows where it came from and whether a signature covers it. A signed record is strong evidence; unsigned labels are usually right, but anyone can add or remove them.

AI Doc Editor

The AI Doc Editor turns an answer from ChatGPT, Claude or Gemini into a document you can edit and export. It runs in your browser and keeps your documents in this browser’s storage. The one exception is writing help, which you start yourself: it sends the selected text to an AI model hosted by Groq to rewrite it.

Reading an answer

Chatbots write their answers in Markdown. The editor reads it with a standard parser that understands tables, task lists and strikethrough. If you copy rich text from the chat window instead, it’s turned back into Markdown first, so both ways of copying give the same result.

Formulas

Maths written in LaTeX is converted to Word’s own equation format, so it opens as an equation you can edit. Because answers about money are full of prices like $5, single dollar signs only count as maths when the answer has no dollar amounts. A formula that can’t be read is kept as LaTeX text, never dropped, and the editor tells you.

Tables and numbers

For Excel, a table cell becomes a number only when it plainly is one: digits with optional separators and decimals, a leading currency sign or a trailing %. Anything else, such as dates, ranges and codes with leading zeros, stays text exactly as written.

Known limits

• Most images from social networks, messaging apps and screenshots have had their metadata removed, so the AI Image Checker will find nothing either way.
• A missing AI label never proves an image wasn’t made with AI.
• A very small target size can force the compressor to shrink an image’s pixel dimensions.
• In the AI Doc Editor, matrices in formulas become plain rows in Word.

Found a problem?

If a tool gets something wrong, email admin@samprix.com with a description, or a sample file if you can share one, and we’ll look into it.