The two methods
| AI model (ESRGAN) | Lanczos and sharpen | |
|---|---|---|
| Scales | 2x or 4x | 1.5x, 2x, 3x or 4x |
| First-time download | About 4.1 MB in total (TensorFlow.js plus the model) | None |
| Largest result | 16 megapixels | 36 megapixels |
| Good at | Photos and textures that need crisper detail | Clean, quick enlargement of graphics and photos |
| Runs on | Your graphics card when available, otherwise the processor | Your processor |
Upscaling a picture
Choose the image
Its current size is shown once loaded.
Pick a method
AI model (ESRGAN) with a 2x or 4x scale, or Lanczos and sharpen with a scale and a Sharpening slider from 0 to 1.5.
Check the output size
A line shows the final dimensions. If the result would be too large for a browser tab, choose a smaller scale.
Upscale and save
Choose PNG, JPG or WebP and press Upscale. The result shows its size and which method and hardware were used.
Example: a supplier's tiny product photo
A gift shop wants to put a supplier's new candle on a printed shelf card, but the only photo is a 400 x 300 pixel image from the supplier's old website.
The AI model at 4x produces a 1600 x 1200 picture. Edges of the jar and the label lettering are crisper than a plain enlargement, which makes it usable at a small print size. It still cannot recover words that were never legible in the original.
candle.jpg, 400 x 300 px, AI model (ESRGAN), 4x
candle-4x.png, 1600 x 1200 px
Questions people ask
How can I make a small image bigger without it going blurry?
Use the AI model at 2x or 4x for photos, which sharpens edges and textures as it enlarges. For flat graphics, Lanczos with a little sharpening is often enough and needs nothing downloaded. Starting from the largest original you can find always helps most.
Is the AI upscaling done online?
No, it runs in your browser. The first time, TensorFlow.js and the ESRGAN model, about 4 MB together, download from the jsDelivr public code library and are cached. Your image is processed on your own device, using the graphics card if the browser allows it.
Why does upscaling take so long?
The AI model works through the picture in pieces, and a 4x result has sixteen times as many pixels as the original. On a phone or an older laptop without graphics acceleration this can take a while. Try 2x first, or use Lanczos for a near-instant result.
Where your data goes
Upscaling runs in a background worker in your browser. For the AI method, open-source TensorFlow.js and the ESRGAN model are downloaded once from the cdn.jsdelivr.net public code library; your image is never uploaded.
Checked against the tool on