What every tool on PDF & Image Tools does (and doesn't)

Last updated: June 2026. If anything here goes stale, that's a bug. Tell me and I'll fix the page, not the wording.

No marketing on this page. I run pdfandimagetools.com, and this is the whole list: every tool, what it actually does, and where it breaks. If a tool can't honestly do something you'd reasonably expect it to, it says so right here. I'd rather you read this, decide it's not for you, and leave, than open the tool, hit the wall, and feel tricked.

There's exactly one rule the whole site is built around. It's the first thing I want in writing.

The one rule

Your files never leave your browser.

Every tool runs locally, in the tab you have open. There is no upload step. When you drop a PDF or a photo onto a tool, it gets read into your browser's memory and processed by JavaScript and WebAssembly running on your machine. Nothing goes to a server. There is no server waiting to receive files. I couldn't read your documents if I tried, because they're never on a computer I control.

That's the gap between this site and Smallpdf, iLovePDF, Adobe's online tools, pi7.org, and basically everyone else. They upload your file to their servers, do the work there, and send it back. Nothing sinister about it; it's a completely normal architecture. It also means a copy of your private file sits on someone else's disk for some window of time, governed by a privacy policy you didn't read. (Be honest. You didn't.)

Here the trade is different. You get privacy, you skip the upload-and-download round trip (which is a big part of why it often feels faster), and in return your own device does the heavy lifting. On a ten-year-old phone, the chunkier tools will crawl. That's the real cost, and I'm not going to bury it.

Don't trust me. Check.

You should not take a privacy claim on faith from some guy with a website. Two ways to prove it yourself, both under a minute:

1. The Network tab. Open DevTools (F12, or right-click then Inspect), go to the Network tab, clear it, then run any tool on a file. Watch the requests. You'll see page assets and model files load, but nothing that uploads your document. The big bytes go out, to your downloads folder, not up to a server.

2. Cut your Wi-Fi. Load the page once with the internet on, so the tool's code (and, for the background remover, its AI model) can download. Then go offline. Now use the tool. It still works. A tool that was secretly uploading your file simply could not function with the network gone. This is the cleanest proof there is, and it's the one I'd reach for.

The background remover is the most interesting case, because it's a genuine AI model doing inference. The first time you use it, it pulls down a model file once, then caches it. After that you can be fully offline and remove backgrounds all day. The model runs on your GPU through WebGPU where your device supports it, or on your CPU through WebAssembly otherwise. Your images never touch the network on either path.

Device support matrix

The site is built to work everywhere, but "works" isn't a yes-or-no thing when one of your tools is an AI model running in a browser tab. So here's the honest breakdown.

Tool group	Desktop (Mac/Win/Linux)	iPad	iPhone / Android phone
Image compress / resize / crop / rotate / convert	Full	Full	Full
Watermark / meme / HTML to image	Full	Full	Full
Image to text (OCR)	Full	Full	Full (slower; the sharpening pre-pass is tuned off on phones to save memory)
Background remover	WebGPU, fastest	WASM, single-thread (lean engine)	WASM, single-thread: works, slower, can hit memory limits on big images
All PDF tools	Full	Full	Full, but very large files (100 MB+) or huge batches can run out of memory

A few specifics worth stating flat out:

• Batch limits depend on your device. Desktop and iPad let you run up to 50 files (800 MB total) at once. Phones cap at 15 files (300 MB total). That's not me being stingy. Phones have a hard RAM ceiling, and pushing past it doesn't throw a tidy, catchable error. It just kills the tab. So the cap is me picking a number you'll survive over a number that looks bigger in a feature list.
• iPhone and iPad Safari don't support WebGPU. So the background remover falls back to WebAssembly on those. Same result, it's just doing the matrix math on the CPU instead of the GPU.
• Old or low-RAM phones are the genuine weak spot, both for the AI tools and the rasterizing PDF tools (heavy compress, PDF-to-image). If a big job kills your tab, that's the memory ceiling, not a secret bug I'm hiding. Try a smaller file, or fewer of them.

The image tools

Compress Image. Shrinks JPG, PNG, and WebP. The thing most competitors don't do: you can compress to an exact target size. Tell it "under 200 KB" and it binary-searches the quality until it lands below your number, then shows you the real result. Honest limit: if you ask for something physically impossible, like a detailed photo under 5 KB, it hands you the smallest it can make and says so, instead of producing mush and pretending it hit the mark.

Resize Image. Change dimensions by pixels, by percentage, or by a max bound, with or without the aspect ratio locked. No AI upscaling magic here. It's a straight, fast resize. Scale a small image up and it'll look soft, exactly like any normal resize would.

Crop Image. Trim to a region, with aspect-ratio presets. Nothing fancy, and that's the point.

Rotate Image. Rotate in 90s, set a free angle, or flip and mirror. It's also the engine the crop tool sits on top of.

Convert (20+ formats). JPG, PNG, WebP, AVIF, HEIC, TIFF, GIF, BMP, SVG, ICO, JFIF, in a lot of combinations. The ones people actually need: HEIC to JPG/PNG/WebP, for those iPhone photos nothing else will open, and PNG/JPG to ICO for favicons. SVG to raster renders the vector at whatever size you choose. Limit worth knowing: GIF converts the first frame to a still. It does not keep the animation. Animated output isn't a thing here.

Watermark Image. Text or logo, with position, opacity, rotation, tiling, and a real font library. Does exactly what it says on the label.

Meme Generator. Top text, bottom text, the classic look. It's honestly the watermark engine wearing a hat.

HTML to Image. Paste HTML, get a PNG. It renders in a sandboxed iframe with a strict content policy, so it can't reach into your private network or leak anything out. Limit: it's a screenshot of rendered HTML, not a full browser, so some exotic CSS or pages that lean hard on external resources won't come out pixel-perfect.

Image to Text (OCR). Pulls text out of a photo or screenshot, fully in the browser, using Tesseract with a sharpening pre-pass for faint or dense scans. This is the one with the most caveats, so I'll be blunt. It's genuinely good on clean printed text and screenshots. It struggles with handwriting, heavily stylised graphics, charts, and text baked into busy infographics. In-browser OCR has a real ceiling and I won't pretend it doesn't. If a chart's labels come out as gibberish, that's the model hitting its limit, not a setting you forgot to flip.

Background Remover. The flagship. An AI model lifts the background off a photo and hands you a transparent PNG, all on your device, offline after the first load. Now the honest limit, and it's a real one: on cluttered scenes (a gym-mirror selfie with ten things behind you, say) the lightweight model can fail to find a clear subject and give you back an empty or ragged mask. It's tuned for a clear subject on a background: people, products, pets, cars. It is not a tiny Photoshop artist living in your tab. On a clean portrait or a product shot it's excellent. On chaos, it tries its best, and sometimes it just shrugs.

The PDF tools

Compress PDF. Two modes. "Keep text" does a lossless structural re-save plus safe image recompression, so text stays selectable and the file still gets smaller. "Maximum" rasterizes each page and rebuilds a much smaller file, with the same optional target-size search the image compressor has. There's an opt-in OCR toggle (off by default) that layers an invisible searchable text onto scanned PDFs. Honest limits: Maximum mode turns text into an image, so it's no longer selectable unless you switch OCR on. OCR is desktop and iPad only, because on a phone it can take ten-plus minutes on a large file. And if a PDF is already optimised, the tool just gives you your original back and tells you so, rather than making it bigger to look busy.

Merge PDF. Combine multiple PDFs, reorder them, or drop into a page-level manager to merge specific pages. Phones can run out of memory on very large merges, lots of big PDFs at once, which is the device limit again, not a logic bug.

Split PDF. Split into individual pages or into ranges.

Organize PDF. The full page manager: reorder, delete (with undo), rotate, and insert blank pages, across several documents at once.

Rotate PDF. Rotate the whole document, or pick specific pages. Lossless.

Crop PDF. Trim page margins.

PDF to Text. Extract the text content to a .txt file, with sensible handling of columns. Limit: it pulls the text layer. A scanned PDF that's really just images of pages has no text layer, so for that you want OCR, not this.

PDF to JPG / PNG. Each page becomes an image.

Image to PDF (also JPG, PNG, HEIC, TIFF, and WebP to PDF). Combine images into one PDF.

Word to PDF / Excel to PDF. Convert .docx and .xlsx. Honest limit: these render in the browser, so they're faithful but not byte-identical to what Microsoft Word's own print engine spits out. Complex page-break logic especially leans on the hints already inside your document. For a normal letter or a normal spreadsheet, it's solid.

HTML to PDF. Render HTML to a PDF, the same sandboxed approach as HTML to Image.

Add Text to PDF. Place text anywhere on any page.

Fill PDF Form. Fill interactive form fields and flatten them.

Annotate PDF. Draw, highlight, add notes and shapes.

Sign PDF. Add a signature, drawn, typed, or uploaded. One thing I want to be clear about with the word "sign": this puts a visual signature on the page. It is not a cryptographic digital signature backed by a certificate authority. If you need a legally certified e-signature with an audit trail, this isn't that, and I'd rather tell you up front than let you find out later.

Watermark PDF. Text or image watermark, with page ranges and positioning.

Page Numbers. Add page numbers, with options for position and format.

Delete Pages. Remove pages, or flip it around to keep only the pages you want.

Protect PDF. Add a password. Real encryption, through a WebAssembly build of qpdf running in your browser.

Unlock PDF. Remove a password, but you have to know the password. This decrypts a PDF you already hold the key to. It is not a cracker, and it will not break into a document you don't have the password for. I get search traffic for "unlock PDF without password," and I deliberately don't chase it, because I can't do it and I won't pretend to.

Redact PDF. Black-bar sensitive content. It draws the bars and, the part that actually matters, deletes the content underneath so it can't be copied back out. (A surprising number of "redaction" tools just lay a black rectangle over text that's still sitting there, fully selectable. This is not one of those.)

What the site does not do

The gaps, stated plainly, because a spec sheet that only lists wins isn't a spec sheet. It's a brochure.

• No accounts, no cloud storage, no file history. Close the tab and the file is gone from memory. There's nowhere to "find your old documents." That's the privacy model, not a missing feature.
• No animated GIF output. No video tools either (a video background remover is on the maybe-someday list).
• No image upscaling, face blur, or face redaction yet. Those need heavier in-browser models with the same device constraints as the background remover, and I'd rather not ship them half-working.
• No OCR for handwriting, and only limited OCR on charts and stylised graphics. Same ceiling as before.
• No password cracking. Unlock needs the real password.
• No certified digital signatures. Sign is visual.
• It can be slow on old phones, for the AI and rasterizing tools especially. The work runs on your hardware, so your hardware sets the pace.

Why this page exists

I built the whole thing solo, and I'm self-taught, so I have no brand to hide behind and every reason to just be straight with you. The honest version of a tool is the one that tells you where it stops. A privacy promise you can't check is only a slogan, so I handed you the Network tab and the airplane-mode test. A capabilities list that quietly skips the gaps is a trap, so I wrote the gaps down too.

Every technical write-up I publish links back here as the canonical "what it actually does" reference. If a claim on this page turns out to be wrong, that's the worst kind of bug, and I'd genuinely want to hear about it. The tools are free, there's no signup, no limits beyond your own device's, and nothing stamped on your output. The only thing I'm really selling is that I'm not lying to you, and this page is me putting the receipts on the table.

If you want to poke at any of it, the whole list is here. Load a tool, cut your Wi-Fi, and watch it keep working.

— Swathik