Quick Answer
An image upscaler enlarges a picture beyond its original resolution while recovering sharpness that plain stretching destroys. Our free tool does it two ways: an AI engine (Swin2SR super-resolution network) that reconstructs fine detail as it doubles the resolution, and a classic engine (Lanczos3 resampling plus unsharp masking) that is instant and works offline. Both run entirely in your browser — the image is never uploaded — and you can export 2x or 4x PNGs with no sign-up or watermarks.
How It Works
Upload a PNG, JPEG, or WebP image — it stays on your device the whole time
Choose 2× or 4× and pick an engine: AI upscale (Swin2SR super resolution, ~20 MB model downloads once with a progress bar) or the instant Classic engine (Lanczos3 + sharpen)
Click Upscale — 4× runs the AI engine twice for maximum reach; if the CDN is unreachable the tool auto-switches to Classic
Compare the side-by-side before/after view, then download the result as a PNG named like upscaled-2x-photo.png
Key Facts
- Upscale by 2x or 4x (4x runs the AI engine twice for maximum reach)
- AI engine: Swin2SR super-resolution via transformers.js, ~20 MB quantized model, downloaded once and cached
- Classic engine: real Lanczos3 resampling plus unsharp mask — instant, no download, works offline
- AI input is capped at 1024 px longest side to bound memory and speed
- Side-by-side before/after comparison at full zoom-in detail
- Download as PNG with an honest filename like upscaled-2x-photo.png
- Falls back to the classic engine automatically if the CDN is unreachable
Frequently Asked Questions
How do I upscale an image for free?
Upload your PNG, JPEG, or WebP, choose 2x or 4x, pick an engine, and click Upscale. The AI engine (Swin2SR) reconstructs edges and textures as it enlarges; the classic engine uses Lanczos3 resampling plus sharpening and finishes instantly. Then click Download PNG. It is completely free — no account, no credits, no watermarks.
Is my image uploaded to a server?
No. The upscaler runs entirely in your browser. The only network request is a one-time download of the open-source AI model files from a public CDN (Hugging Face / jsDelivr); the image itself is processed on your device and never transmitted or stored. The classic engine needs no network at all after the page loads.
What is the difference between the AI and classic engines?
The AI engine (Swin2SR) is a neural network trained for super resolution: it hallucinates plausible fine detail — sharper text, cleaner edges — and generally looks best, but needs a ~20 MB model download and several seconds of compute. The classic engine applies true Lanczos3 resampling followed by unsharp masking: it is instant, deterministic, and adds no new detail but keeps the image crisp. If the AI CDN is unreachable, the tool automatically falls back to classic.
Are there size or resolution limits?
For the AI engine the input is capped at 1024 px on the longest side (larger images are downscaled first), so a 4x result is up to 4096 px. The cap bounds memory and keeps processing under a few seconds. The classic engine accepts inputs up to 2048 px and produces up to 8192 px. Very large outputs may be slow on phones or devices with little RAM.
Why does the first AI upscale take longer?
The first run downloads the quantized Swin2SR model — about 20 MB — plus the transformers.js runtime from the CDN, with a progress bar showing the download. Your browser caches both, so every later run starts immediately. A 4x upscale runs the 2x model twice, so expect roughly double the processing time of a 2x upscale.