From Idea to 2M Users: The Pixly Story

Two years ago, Pixly was a weekend project. Today, it processes millions of images per month for over 2 million users. The journey wasn't planned, funded, or smooth — but it was deliberate. Here's the story.

The Problem

It started with frustration. I needed to remove backgrounds from 20 product photos for a client project. I tried three online tools — all required uploads, two required accounts, and one watermarked the output unless I paid. I thought: why can't this just work, in my browser, for free?

The First Prototype

That weekend, I built a prototype using TensorFlow.js and a pre-trained segmentation model. It was slow — 30 seconds per image — and the results were rough. But it worked entirely in the browser. No uploads, no account, no watermark. That was the seed of Pixly.

The first version was ugly and slow. But the core idea — private, free, in-browser image processing — resonated immediately.

Finding Users

I posted the prototype on Hacker News and Reddit. The response was overwhelming. 50,000 visitors in the first week. People loved that their images never left their devices. The feedback was clear: make it faster, add more tools, and keep it private.

• Month 1: 50K visitors from Hacker News and Reddit
• Month 3: Added upscaling and compression, reached 200K users
• Month 6: Launched Pro plan, hit 500K users
• Month 12: 1M users, added batch processing and format conversion
• Month 24: 2M users, WebGPU acceleration, full tool suite

The Business Model

From the start, I wanted the core tools to be free forever. But servers cost money — except they don't, because Pixly runs client-side. The only costs are CDN hosting for the models and the website itself. This is why we can offer powerful tools for free and charge only for Pro features like batch processing and 4K upscaling.

Lessons Learned

The biggest lesson: privacy isn't a feature, it's a positioning. When users learned their images never leave their device, it changed the entire conversation. We didn't have to promise not to look at their data — we structurally couldn't. That trust is what built Pixly.

You don't need to promise privacy when you architect for it. The best privacy policy is one you never have to write.

Technical Challenges Along the Way

The journey from prototype to platform was not without technical hurdles. The first major challenge was model size. The initial U²-Net model was 176MB — too large for a web download. We spent weeks on quantization and pruning, eventually getting it down to 44MB with negligible quality loss. This was the difference between a tool that took 30 seconds to load and one that was ready in 3 seconds.

The second challenge was browser compatibility. WebGPU was not widely supported when we started, so we had to build a three-tier fallback system: WebGPU for best performance, WebGL for decent performance, and WASM for basic functionality. Writing the same inference pipeline three times was painful, but it ensured that every user could access the tool regardless of their browser.

The third challenge was memory management. Processing large images in the browser can consume significant RAM, leading to tab crashes. We implemented streaming processing for large images, breaking them into tiles and processing each tile independently. This allowed us to handle 50+ megapixel images without exceeding browser memory limits.

Community and Feedback

One of the most surprising aspects of Pixly's growth was the community that formed around it. Users started sharing workflows, creating tutorials, and requesting features. We received feature requests from freelance designers, e-commerce managers, journalists, researchers, and teachers — each with unique needs that shaped our roadmap.

The most requested feature, by far, was batch processing. Users who discovered Pixly for a single image immediately wanted to process hundreds at once. This feedback directly led to the Batch Mode feature, which is now one of our most-used capabilities. Listening to users — not analytics, not assumptions, but actual user messages — was the best product strategy we ever employed.

• Background removal was the first tool and remains the most popular
• Upscaling was the second most requested feature — added in month 3
• Batch processing was the top community request — added in month 12
• AI compression was driven by e-commerce users needing smaller files
• Format conversion was requested by web developers standardizing on WebP

What's Next

We're just getting started. New AI tools, better performance, and deeper integrations are all on the roadmap. But the core promise remains: your images are yours. We'll never upload them, never look at them, and never compromise on that.