AI Image Upscaler
Make small images bigger and sharper with on-device AI — or classic Lanczos for any size.
Image
AI draws sharper edges and textures.
Keeps every pixel and any transparency.
Drag to move around a zoomed picture; pinch or Ctrl + scroll (⌘ + scroll) to zoom. With the picture focused, + and − zoom and 0 fits it.
Enlarged image
AI upscaling adds plausible detail — it cannot recover what the photo never captured. Do not rely on enlarged images as evidence, to read text or number plates, or to identify people.
About the AI Image Upscaler
Enlarge a small photo, a screenshot or an old picture so it can be printed or used at a bigger size. AI mode uses a neural network trained on high-resolution photos to enlarge 2×, 3× or 4× and draw edges, lines and textures more crisply than plain stretching. Classic mode uses Lanczos resampling, the sharp traditional method, for any size up to the device’s limit — it is fast and invents nothing.
Everything runs in your browser: the AI model runs on your graphics processor (WebGL) and the image is never uploaded. A before/after slider lets you compare the result at full size before you save it as PNG, JPG or WebP.
How to use it
- Choose, drop or paste an image (JPG, PNG, WebP, HEIC, GIF or BMP).
- Pick AI (2×, 3× or 4×) or Classic (1.5× to 4×, or type a width). The output size is shown before you start.
- Select Upscale. AI mode works through the image in tiles and shows the time left; you can cancel at any time.
- Drag the slider (or use the arrow keys on it) to compare before and after; zoom to 100% to check details. Then Download as PNG, JPG or WebP, or copy the image.
Examples
AI · 3× → 1,920 × 1,440 px
Enough pixels for a 6 × 4 in print at about 300 DPI, with crisper edges than stretching
PNG with transparency · AI 4×
A 4× PNG that keeps the transparent background (the outline is enlarged with Lanczos)
Classic · width 2,400 px
Exact size, sharp text, no invented detail
Common uses
- Enlarging old or small photos for printing.
- Making product photos and logos big enough for banners and slides.
- Upscaling artwork and pixel-light illustrations for wallpapers.
- Enlarging screenshots exactly, with Classic mode.
What the AI does — and what it cannot do
The AI model is the “ESRGAN medium” network from the open-source UpscalerJS project (MIT licence): a residual dense network trained on the DIV2K collection of high-resolution photos to predict how a small image would look at 2×, 3× or 4× the size. It sharpens edges and restores some texture, but it cannot recover detail the original never captured: everything it adds is a plausible guess. Do not use upscaled images as evidence, to read number plates or to identify people. Faces, text and fine patterns can come out smoother or slightly different from reality.
AI or Classic?
- AI suits photos, illustrations and logos that are small (up to about 2,000 × 2,000 px). It takes from a few seconds to a few minutes depending on the size and your device’s graphics processor.
- Classic (Lanczos) takes a second or two for most images, works on any image up to the size limit, and is the honest choice for documents, screenshots and anything where accuracy matters.
- Compression blocks in heavily compressed JPGs get enlarged too; start from the best copy of the image you have.
Size limits
The enlarged image has to fit in your device’s memory, so the tool makes images of up to about 40 megapixels on a computer and 16.7 megapixels (4096 × 4096) on phones and tablets — the largest canvas older iPhones and iPads allow, and a safe amount of phone memory. The AI model takes inputs up to 4 megapixels on a computer and 1 megapixel on a phone. If a request is too big, the tool tells you the largest scale that fits.
Limitations
- AI mode adds plausible detail; it cannot restore what the original never showed.
- AI inputs are limited to about 4 MP on computers and 1 MP on phones; outputs to about 40 MP and 16.7 MP.
- On a device without hardware graphics acceleration the AI runs on the processor and is slow; use Classic mode there.
- One image at a time.
- Animated GIFs are enlarged as a single still frame.
Privacy
The image is processed on your device and never uploaded. The first AI upscale downloads the AI engine and the model for that scale (about 2.8 MB) from MySmartCoPilot, not from a third party; Classic mode downloads nothing. Saved files are encoded fresh from the pixels, so they carry no EXIF data from the original (no camera details or GPS location).
Frequently asked questions
Is my image uploaded?
No. Both modes run in your browser. Only the AI model (about 2.8 MB per scale) is downloaded from MySmartCoPilot the first time you use AI mode.
Can it make a blurry photo sharp?
Only up to a point. The AI makes edges crisper and textures more convincing, but it guesses detail rather than recovering it. A badly blurred face or unreadable text stays unreliable.
Why does AI mode take longer than Classic?
The neural network performs millions of calculations for every pixel, tile by tile, on your graphics processor. Lanczos is a simple filter. On a phone without fast graphics, AI mode can take a few minutes for a larger image.
Which format should I save in?
PNG keeps every pixel and transparency (largest file). JPG is much smaller and fine for photos. WebP is small and keeps transparency; all modern browsers open it.
Does it keep the transparent background of a PNG?
Yes. The colours are enlarged by the chosen method and the transparency (alpha channel) with Lanczos, so edges stay smooth. Save as PNG or WebP to keep it.
How many pixels do I need to print?
For a sharp print, about 300 pixels per inch: a 6 × 4 in print needs about 1,800 × 1,200 px, an A4 page about 2,480 × 3,508 px. The image resizer can set the DPI.