What RoundCut is
RoundCut is a suite of free image tools. It covers the small, specific jobs people keep needing: cropping (round avatars, rectangles, social presets), converting between formats, compressing, removing backgrounds, resizing, AI upscaling, and turning images into PDF. Some of those run entirely in your browser, others route a single request through named infrastructure to do work the browser doesn’t yet do as well, always with an invisible client-side fallback. No signup, no watermark, no resolution cap.
The site exists because the alternatives, in our experience, get one of three things wrong: they upload everything to a server (whether or not the work needs one), they paywall basic operations, or they bury simple tools under intrusive ads. RoundCut tries to do none of those, and we describe the actual data flow honestly rather than claim “100% client-side” when the cloud path is what gives you remove.bg-grade results.
Why we built it
The first version of RoundCut launched in 2024 as a single tool, a circle crop for profile pictures, built to scratch a personal itch: the existing options either asked for an account or rendered the image at 256 pixels and offered a “Pro” upgrade for the original resolution. Neither felt acceptable for a one-line operation that every browser has been able to do natively for years.
The site grew organically from there. Each new tool answered a recurring question we kept hitting personally, “how do I shrink this PNG without losing transparency?”, “how do I strip the GPS coordinates from this photo before sharing?”, “how do I make a Discord avatar without installing software?”. The 2026 rebuild is the same project, rewritten on top of modern web standards (Astro, WebAssembly, Canvas API for the browser side. Cloudflare Workers + a small VPS + a cloud image API for the handful of tools where server-side encoders or AI models still beat the browser).
Who it’s for
RoundCut is built for anyone who needs to do a small, specific thing to an image and doesn’t want to install software, sign up for an account, or hand their photo to a generic uploader that retains it. The tools fall into a few categories of work:
- Cropping. Circle-crop a photo for Discord, LinkedIn, Slack, Instagram, or any platform that masks avatars to a circle, plus rectangular and social-preset crops. Runs in your browser. Export as transparent PNG, WebP, AVIF, or JPEG.
- Format conversion. Move between PNG, JPEG, WebP, and AVIF in either direction. Runs in your browser via Canvas and WebAssembly, except AVIF encoding, which routes one request to a Worker (with a main-thread browser fallback) because the in-browser AVIF encoder is slow on some devices.
- Compression. Preview and the quality slider run in your browser so iteration stays instant. The final download sends your image once to our compression service for sharp-grade encoder quality, and falls back to the in-browser blob if the service is unreachable.
- Background removal. Pull a subject out of its background with AI. The default path sends your image once to a Cloudflare Worker that runs the BiRefNet model on Cloudflare’s edge GPUs (same architecture as remove.bg), with the staging copy auto-deleted within one hour. When the cloud is unreachable, the tool transparently falls back to an in-browser model (ISNet via ONNX Runtime + WebAssembly), no upload at all on the fallback path.
- Resizing. Change pixel dimensions with high-quality resampling, entirely in your browser. No upload.
- AI upscaling. Enlarge an image with super-resolution. The default path uses a cloud model, with an in-browser resampling fallback when the cloud is unreachable.
- Image to PDF. Wrap one or more images into a PDF. Runs in your browser for a single image, and routes to our service for multi-image documents, with a client fallback.
Common to all of them: no upload that we keep, no account, no watermark. The server-touching tools document exactly what happens to your image, where, and for how long.
Our principles
Honest about the architecture
“Your photos stay on your device” is a marketing line we deliberately don’t use, because it’s not true on every path. The truth is: the cropping, resizing, format conversion (non-AVIF), and compress-preview work never leaves your browser, and you can verify that in DevTools’ Network tab. The server-touching paths (background removal, the compress final download, AI upscaling, AVIF encoding, and multi-image PDF) send your image exactly once, to named infrastructure we operate (a Hostinger VPS in Germany, Cloudflare Workers, and a cloud image API), do their work, return the result, and either auto-delete or rely on a content-addressed cache that’s not tied to you. We tell you which path you’re on, and we don’t pretend the cloud path doesn’t exist.
If you need a hard guarantee that bytes never leave your device for a particular job, the browser-only tools cover that. The cloud-backed tools each ship an in-browser fallback that runs without any upload when you turn off the network.
Free should mean free
Every tool that exists today will stay free. We don’t watermark exports, we don’t cap resolution, we don’t require an account, and we don’t show interstitial ads. We may add a paid tier in the future for extra capability (batch processing, larger upload sizes, higher cloud quotas), but it would not gate anything that’s already shipped.
Truthful copy
What we say a tool does and what the code actually does are the same thing. If the page says “Export as PNG, WebP, or JPEG”, you’ll find all three in the format picker. If the page says “up to 4096 px”, that’s the real cap in code. If a tool’s default path is cloud-routed, the page says so, not “all-in-browser” with an asterisk. We have an internal rule that copy and code ship in the same commit, exactly so this never drifts.
Accessibility from the start
Every page passes WCAG 2.2 AA contrast in both light and dark themes. Every interactive element is reachable by keyboard, has a visible focus ring, and has a minimum 24 × 24 CSS-pixel touch target. Screen-reader landmarks (header, main, footer, nav) are in place. The site is built so that the experience for someone using a screen reader, a keyboard, or a small phone is not noticeably worse than for someone with a high-end desktop.
No tracking by default
The site uses Cloudflare Web Analytics for aggregate page-view counts. There are no tracking cookies, no fingerprinting, no third-party analytics scripts. The full data flow, including the cloud paths for the server-touching tools, is described in the privacy policy.
How we stay free
RoundCut’s static site runs on Cloudflare Pages’ free tier, and the browser-only tools cost us essentially nothing per use because the work happens on your machine.
The server-touching tools have a small but real cost:
- Cloud processing (background removal, AVIF encoding, AI upscaling) runs on Cloudflare Workers, an R2 bucket holding a few MB at any time, a KV namespace for rate-limiting, and a cloud image API for super-resolution. Mostly on free tiers, with per-IP and per-day caps that keep usage sustainable.
- The compression and PDF service runs on a small Hostinger VPS (KVM 2, ~$10/month) shared across two tenants, returning sharp-grade encodes to thousands of requests per day. The cache hit ratio amortizes most of the work.
As traffic grows, we plan to fund continued development with one primary layer:
- One discreet ad below the tool, served by Google AdSense on tool pages only. The home page, satellite pages, and legal pages remain ad-free. This is the standard pattern competitors like iLoveIMG, TinyPNG, and Pixlr use, and it’s what funds free tools at scale on the open web.
The AdSense layer is live on roundcut.com.br and pending review on roundcut.app.
A paid tier, if we add one, would come only after sustained traffic, and free use
stays free.
How to reach us
For questions, feedback, or bug reports, email support@roundcut.app. We read every message and try to respond within five business days.
For privacy-related requests, see the privacy policy. For a technical look under the hood, see how it works.