Rooze.ai is live: photorealistic SD1.5 text-to-image, fully in your browser

Rooze.ai is a text-to-image web app that runs Stable Diffusion 1.5 locally in your browser — no signup, no install, and no server generation.

The big deal isn’t just “Stable Diffusion in a web page.” It’s that you can run SD1.5 community photorealistic checkpoints (the kind people normally run in ComfyUI / Automatic1111) with a real end-user UX — without asking anyone to download apps or configure workflows.

TL;DR: Photorealistic text-to-image in-browser. Models download once to your browser. Generation runs locally on your machine.

Privacy by default: because inference runs on-device, your prompts and generated images are kept entirely private.

Two more examples from the Rooze showcase — different prompts, different vibes, all generated locally in-browser.

Why I built this

Text-to-image has reached a point where the models are incredible — especially the SD1.5 ecosystem. But actually using them still often means:

• Installing tooling (ComfyUI, Automatic1111, Python environments…)
• Downloading huge model files (anything from 2GB to 10GB)
• Learning a bunch of settings and workflows
• Dealing with GPU requirements and setup friction

I wanted the opposite: a link you can send a friend where they can get to high-quality results with minimal friction — and where their data doesn’t need to leave their device.

The only real “setup” is the first model download: Rooze uses SD1.5 checkpoints and can ship them in FP16, so photorealistic community quality is often a ~2GB download (not 10GB+). After that, the model stays stored locally in your browser — and generation runs on-device.

What’s special about Rooze

1) SD1.5 + photorealistic community quality in-browser

To my knowledge, Rooze is one of the first web apps (perhaps the very first) where end users can run SD1.5 community photorealistic checkpoints fully in-browser — not just a research demo.

That matters because SD1.5 community variants can look seriously good — the kind of photorealism many people associate with “desktop workflows only.”

A huge thank you to the amazing community model creators. Without the hundreds of hours they’ve spent refining SD1.5 into such strong photorealistic checkpoints, Rooze wouldn’t be possible.

2) No signup + no install

You can generate without creating an account. No local install. No setup — just open the site, pick a model, type a prompt, and generate. The first time you choose a model, it may take a one-time download (often around ~15 minutes on a good connection), then it’s cached locally. Because generation runs locally, your prompts and outputs stay on your device.

3) Fast on Apple Silicon

In my testing:

• Under ~30 seconds per image on a MacBook Pro M4
• Under ~15 seconds per image on an M3 Pro

Your mileage will vary by device, browser, and settings — but the fact this is even possible in a browser is the fun part.

4) Batch mode (because local changes the economics)

Since generation happens locally, there’s no “per-image cost.” That enables a workflow where you can leave Rooze running and generate lots of variations — great for prompt iteration, mood boards, character exploration, and more.

To make it concrete: on an off-the-shelf MacBook Pro with an M4, a quality photorealistic image can take around ~30 seconds. Leave your laptop generating overnight and in ~8 hours you can end up with ~1,000 images — all completely free (and generated on-device).

Rooze’s batch processor lets you paste in 100+ prompts and choose how many variations per prompt you want, so you can “set it and forget it” while you iterate.

5) Model Manager (because 2GB models are real life)

Some models are huge. Rooze includes a Model Manager page so you can see what you’ve downloaded, delete models you don’t want, and keep your storage tidy.

What you can do today

• Generate photorealistic images from prompts
• Try different SD1.5 community variants
• Batch-generate many variations
• Manage/delete local models when you’re done

Try it here: https://www.rooze.ai

What I’d love feedback on

If you try Rooze, the most helpful feedback is:

• Your device + browser (and roughly how fast generation is)
• Which models feel best to you
• Any UX friction (especially model download and storage management)
• What tool or feature would make this a daily driver for you

If you post results, tag/share Rooze — I’ll be collecting examples and performance notes across devices.