Small images fall apart when you make them bigger. Stretch a 600 pixel product photo to fill a 2,000 pixel slot and every edge turns soft and every JPEG block shows. An AI upscaler takes a different approach: instead of just spreading the existing pixels further apart, it predicts what sharp edges and fine texture should look like at the bigger size and draws them in.
This tool runs an upscaling model of the Real-ESRGAN kind on our server. You choose 2x or 4x, see the result next to the original with a zoomable before and after slider, and download a PNG. Below: how to use it, what it really improves, where it fails, and which sizes marketplaces and print need.
How to upscale an image
- Drop a JPG, PNG or WEBP image onto the box, or click to choose one.
- Check the size note: it shows the image's pixel size and megapixels. The upscaler accepts up to 1.5 megapixels, for example 1,414 x 1,060 or 1,224 x 1,224 pixels.
- If the image is larger, either click Shrink to 1.5 MP (your browser makes a smaller copy, nothing is uploaded yet) or crop the part you need with the crop tool.
- Choose 2x or 4x. The line below shows the exact result size and how large it prints at 300 DPI.
- Click Upscale image. The model runs on a CPU server, one image at a time, so you may wait in line briefly.
- When it is done, the PNG downloads, and a before and after view appears. Drag the slider, and use Zoom or tap the picture to look closely at edges, text and faces.
What an AI upscaler improves
A normal resize (the bicubic or bilinear scaling in most editors and browsers) can only blend neighboring pixels. The result is bigger but blurry. A trained upscaling model has learned, from large sets of sharp images paired with blurred and compressed copies of them, how low-resolution detail usually looks at high resolution, so it can:
- Draw crisp edges on product outlines, packaging lines and architecture
- Remove the blocky squares and color smudges that JPEG compression leaves behind
- Add plausible fine texture to fabric, wood, foliage and skin
- Smooth noise in flat areas such as walls, sky and studio backgrounds
The before and after view on this page shows the original enlarged by your browser on the left of the slider, which is what a normal resize looks like, and the AI result on the right. At 2x zoom the difference on edges is usually obvious.
What it cannot do
Upscaling is educated guessing. The model draws detail that is likely, not detail that was there. That has real consequences:
- Lost detail stays lost. A face that is ten pixels wide in the original will become a sharp face, but not necessarily that person's face. The same goes for a car's number plate or a serial number.
- Small text gets garbled. Tiny letters on labels and screenshots often turn into neat-looking shapes that are not real letters. Check every word you need to read.
- Motion blur and missed focus are not fixed. The model sharpens edges that exist; it cannot reconstruct an image that was blurred when it was taken.
- Heavy compression can be amplified. On a very poor JPEG, 4x can turn artifacts into odd patterns. Try 2x first on images like that.
Warning: Do not use upscaled images as evidence of what something looked like, for example in a damage claim or an ID check. The added detail is invented by a model and may not match reality.
Upscaling, resizing and sharpening compared
Three different tools all promise "bigger and sharper", and they are easy to mix up:
| Method | What it does | Result |
|---|---|---|
| Resize (bicubic) | Spreads existing pixels over a larger grid and blends between them | Bigger and soft; no new detail |
| Sharpen filter | Increases contrast along edges that already exist | Crisper edges, but also more visible noise and halos |
| AI upscaling | Predicts and draws the detail a sharper, larger version would likely have | Bigger and sharp, with invented fine detail |
For a photo that is already big enough but a little soft, a mild sharpen in any photo editor is the honest fix. For a photo that is too small, AI upscaling is the better tool, and you rarely need to sharpen afterwards.
2x or 4x?
| Start size | 2x result | 4x result | Good choice |
|---|---|---|---|
| 500 x 500 | 1,000 x 1,000 | 2,000 x 2,000 | 4x for a marketplace main image |
| 800 x 600 | 1,600 x 1,200 | 3,200 x 2,400 | 2x for web, 4x for print |
| 1,024 x 1,024 | 2,048 x 2,048 | 4,096 x 4,096 | 2x for a store, 4x for a poster |
| 1,414 x 1,060 | 2,828 x 2,120 | 5,656 x 4,240 | 2x unless you need a large print |
A rule of thumb: pick the smallest scale that reaches the size you need. Every doubling asks the model to invent four times as many pixels, so 2x stays closer to the original, while 4x is for images that are truly small. If you need 3x, upscale 4x and resize down with the image resizer; shrinking a sharp image loses nothing important.
Sizes that marketplaces and print need
Most people upscale to hit a minimum size somewhere. These are the common targets, checked October 2026:
| Where | Size to aim for | Note |
|---|---|---|
| Amazon product images | At least 1,000 px on the longest side | Below that, shoppers cannot zoom; 2,000 px gives a better zoom |
| Shopify product images | 2,048 x 2,048 px for square images | Shopify's recommended size; keep all product images the same shape |
| Etsy listing photos | At least 2,000 px on the shortest side | Etsy shows thumbnails cropped, so leave space around the product |
| Print at 300 DPI | Width in inches x 300 | An 8 x 10 inch print needs 2,400 x 3,000 px |
| Large posters viewed from a distance | 150 DPI is often enough | An 18 x 24 inch poster at 150 DPI needs 2,700 x 3,600 px |
After upscaling, you will usually want a smaller file for the web. A PNG from the upscaler is large because it is lossless. Run it through the image compressor or save it as a high-quality JPG before uploading to your store.
Common jobs for an upscaler
Supplier and marketplace product images
Product photos copied from suppliers or wholesale catalogs are often 600 to 800 pixels, too small for a modern store theme. Upscale them 2x or 4x, check edges and any printed text on the product, then compress. If you are importing many products from other stores, AM Jarvis Product Importer brings the images across with the product, and you can upscale the ones that are too small for your theme.
Old photos and scans
Family photos scanned at low resolution, or images saved from an old website, upscale well when they are sharp but small. Faded colors and scratches are not repaired; that is a different kind of restoration. If you still have the print, rescanning it at 600 DPI captures real detail, which always beats upscaling a low-resolution scan of the same photo.
AI-generated images
Many image generators output 1,024 pixel squares. A 2x upscale gives 2,048 pixels for a store banner or social post; 4x gives 4,096 pixels for print. The style usually survives upscaling very well, because it is already clean and noise-free.
Logos and icons
Simple logos with bold shapes and flat colors upscale cleanly, and thin lines usually stay straight. For a logo you will use often, though, the real fix is a vector version (SVG or PDF) from whoever designed it, which scales to any size without any guessing.
Example: a small supplier photo for a Shopify store
A supplier sends a 640 x 480 pixel JPG of a ceramic mug. Your theme shows product images at up to 2,048 pixels wide, and the photo looks blurry on large screens.
- Drop the JPG into the box. The size note shows 640 x 480 px, 0.31 megapixels, well within the limit.
- Choose 4x. The result line shows 2,560 x 1,920 px.
- Click Upscale image and wait for the PNG.
- Drag the slider over the mug's rim and handle at 2x zoom. The edges should be clean, without the stepped outline of the original. Check the printed logo on the mug: if the letters look invented, keep the 2x version or ask the supplier for a larger original.
- Resize to 2,048 x 1,536 with the image resizer and save as a JPG at high quality, which brings the file down to a size your store loads quickly.
Check before you publish
- Edges: zoom in on the outline of the main subject. It should be clean, without halos or a cut-out look.
- Text: read every word on labels, packaging and signs. Anything you cannot read clearly in the original may be invented.
- Faces and hands: look for a waxy or painted look. If you see it, use the 2x result instead.
- Flat areas: backgrounds and skies should be smooth, not patterned.
- File size: compress the PNG or save it as a JPG before uploading it to a store or website.
Why the 1.5 megapixel limit?
The model runs on our server's processors rather than an expensive graphics card, and every job runs one at a time. Upscaling is heavy work: a 1.5 megapixel image at 4x becomes a 24 megapixel image. Capping the input keeps each job to a reasonable wait for everyone in line. Images bigger than 1.5 megapixels usually do not need upscaling anyway; a 1,500 pixel wide photo already meets most store requirements.
If your image is just over the limit, Shrink to 1.5 MP makes a slightly smaller copy in your browser and swaps it in. You still end up with a much larger result than you started with.
Troubleshooting
- "Over 1.5 MP": the image has more pixels than the upscaler accepts. Click Shrink to 1.5 MP, or crop to the area you need. A 3,000 pixel photo rarely needs upscaling for a website.
- The result looks plastic or painted: the model smoothed real texture it could not tell apart from noise. Use 2x instead of 4x, or start from a less compressed copy of the image.
- Odd patterns in hair, grass or fabric: fine repeating detail is the hardest thing to predict. Look at it at normal viewing size; patterns visible only at 4x zoom rarely matter on a product page.
- Letters changed shape: small text is guessed, not read. Replace text in a design tool, or ask for the original artwork.
- The image will not open in the preview: some files use color profiles or formats your browser cannot show. The server still checks the image when you upload it; if it fails, save it again as a standard JPG or PNG.
- "Not available yet": the upscaling engine is being set up or updated on our server. Try again later; the message has a button for it.
What happens to your image
The image is uploaded over HTTPS, upscaled, and deleted: the upload as soon as the job ends and the PNG 30 minutes after it is ready. Only the image, the tool name and the scale you chose are sent. The before and after comparison is built in your browser from your original and the downloaded result.