Zero upload · Browser-local · AI repair + clean resize

Upscale images right — AI for photos, clean resize for graphics.

Two modes, both browser-local: AI super-resolution rebuilds detail in low-res or blurry photos; a clean Lanczos resize enlarges posters, screenshots and logos with crisp edges. Leave it on Auto and the tool picks for you.

Auto analyses the image — AI repair for photos, clean resize for graphics. You can always override. AI photo repair — best for low-res, compressed or blurry photos. May soften fine text and sharp graphics. Clean resize — best for posters, screenshots, logos, documents. Enlarges with crisp edges; does not invent new detail. · auto-picked for this image

Why upscale here

The right enlargement for what your image actually is.

01

No limits

No watermark, no resolution cap, no sign-up. Most free upscalers cap previews at 1 MP and stamp a watermark — we don't.

02

Nothing uploads

Both modes run in your browser — AI via WebGPU/WebAssembly, clean resize via a Lanczos resampler. Your image never leaves this tab.

03

Honest about detail

AI mode reconstructs plausible detail in low-res photos; clean resize keeps graphics sharp. Neither can invent detail an already-sharp image never had — and the tool tells you which mode fits.

Frequently Asked Questions

An AI upscaler (Swin2SR in this tool) is a neural network trained on millions of high- and low-resolution photo pairs. Given a low-resolution image, it predicts what the missing pixels should look like — recovering detail rather than simply enlarging existing pixels.
Swin2SR runs at fixed ×4 by default — a 512 × 512 input becomes 2048 × 2048. For larger factors, chain a second pass, or use the Lanczos pipeline for plain ×2 / ×3 / ×8 resizing on graphics where AI hallucination is not wanted.
Use AI upscale for photos where you want to recover missing detail (old scans, low-resolution social posts, small phone screenshots). Use plain Lanczos resize for logos, screenshots, and UI assets where the AI's "creativity" can actually add artefacts.
Traditional interpolation (Lanczos, bicubic) computes a weighted average of surrounding pixels — it never invents detail. AI upscaling synthesises plausible detail based on training data, which looks much sharper on natural photos but can introduce hallucinated patterns on text or logos.
No. Swin2SR runs in your browser via ONNX Runtime Web (WebGPU when available, WebAssembly as a fallback). The model file is about 30 MB and is cached after the first run.

Related guides

Related tools

Paste an image and enlarge it now.

Free, no registration, no watermark. AI repair for photos, clean resize for graphics — all in your browser.