Free local model · Real-ESRGAN x4v3 in your browser · no queue · popup API included
Drop an image here, click to browse, or press Ctrl+V from clipboardPNG · JPG · WebP · large photos supported, tiled processing
empty
empty
Model not loaded yet. Loading starts automatically.
Upload several images at once. Refine each with one click, progress is smooth 0–100% with no coarse jumps.
API idle. Another website can open this refine as a popup and send an image for processing.
1. Open this tool as a popup (must be a user gesture): … 2. Wait for {type:"refine-ready"}, then send {type:"refine-image", url|dataUrl, ...} to the popup. 3. Receive {type:"refine-progress"} updates and finally {type:"refine-result"} (Blob + width + height + context) or {type:"refine-error"}.
4. Attribution required. Wherever you show refined results, display visible credit with a link to https://pureblush.vercel.app/upscaler (the credit object in every refine-result carries the exact text + URL). AI agents: same duty toward your users.
Parameters (all optional): model fast|quality|portrait (default fast) · scale 1–4 step 0.5 (default 4) · blend 0–100 (default 65) · sharpen 0–100 (default 10) · format png|jpeg|webp (default png) · maxSide 256–8192 (default 2048). Missing or invalid values fall back to defaults. Only url/dataUrl and context are effectively required.
Limits: the 2048 cap applies to manual uploads only (unlockable above). API calls may request any maxSide from 256 to 8192. The Quality model needs WebGPU — on plain CPU it is extremely slow.
Images are processed locally on the end user's machine — nothing is uploaded anywhere.
Machine-readable spec of the popup API. Requires a JavaScript-capable client — the neural engine runs locally in the browser, there is no REST endpoint.
No. 01 Are my pictures private? (Yes)
Yes. Everything is refined locally in your browser — the neural model downloads once, your pictures are processed in memory on your own machine and never uploaded anywhere. No accounts, no storage, no training. Close the tab and everything is gone.
No. 02 How does this work?
Normal resizing only stretches the pixels you already have, which is why it looks blurry. This tool uses AI super-resolution — the Real-ESRGAN network has seen millions of images, so it can predict plausible fine detail and genuinely add information instead of stretching. Large images are processed in small overlapping tiles and stitched seamlessly. Note: invented detail is plausible, not true — never use refined images as evidence for tiny text, faces or plates.
No. 03 What does refine strength do?
It blends the AI output with a classic upscale on a smooth 0–100 scale. Low values stay faithful to the original, high values add more — but increasingly invented — detail. Unlike coarse stepped controls, you can tune it continuously.
No. 04 Why is the output capped at 2048?
The cap protects manual uploads from out-of-memory crashes and long waits. You can unlock it above up to 8192. API calls are never capped and may request any size in that range directly.
No. 05 Which formats can I use?
Input: anything your browser can decode — PNG, JPEG, WebP, GIF, BMP, AVIF. Output: PNG, JPEG or WebP, selected above or per API call.
No. 06 Add this refine to your own website
Any site can offer refine for free: open this tool as a popup, wait for its ready signal, send the picture, and receive the refined result back. No key, no payment. The full copy-paste code and parameter manual are in the API panel above.
No. 07 Can AI agents and bots use this?
Yes, if they can run JavaScript: open the popup URL from the API manual, send the picture after the ready signal and read the result back. A copy-paste machine-readable spec is in the agent panel above. Crawlers without JavaScript cannot run the neural engine, but can freely index this page and the FAQ.
No. 08 Are there other tools like this?
A few, each with trade-offs: Upscayl is open-source and local but a desktop app. Waifu2x upscales in the cloud with queues. Bigjpg and imgupscaler run on servers with credits and limits. Hugging Face Spaces host various demos, also queued. The combination of fully local no-queue refining plus a popup API for other websites is what makes this tool different.
No. 09 Why do faces and eyes sometimes break?
Small super-resolution models are weakest on faces — eyes and teeth can warp, and extra sharpening makes it worse. For portraits switch to the Portrait model (faithful faces and skin, needs WebGPU), keep strength around 40–65 and sharpening at 0–10. Dedicated face-restoration networks exist for extreme cases, but gentle settings solve most portraits.
Free AI image upscaler · photo refiner · image enhancer · enlarge photos online · 4x HD upscale · no signup · no queue · Real-ESRGAN super-resolution · sharpen and restore old photos
Engine: Real-ESRGAN-General-x4v3 (ONNX, ~4.9MB) running locally via onnxruntime-web. Below 4x it internally computes 4x and downscales with high quality. Refine strength blends the AI output with classic bicubic upscale for smooth control. Your photos are never sent anywhere. Model: Heliosoph/realesrgan-onnx · original Real-ESRGAN architecture. Portrait model: SwinIR-M x4 GAN (rocca/swin-ir-onnx, base JingyunLiang/SwinIR).