AI Image Upscaler
Enlarge a small or soft photo 2x or 4x with an AI model that redraws fine detail, running entirely in your browser — photo never uploaded
On-device AI Runs in your browser. It does not use the AI Studio allowance.
Drop an image to upscale with AITap to choose an image to upscale with AI
or click to browseor tap to choose
Runs in your browser: the picture never leaves your device
Large files (over 50 MB) may take longer or run out of browser memory
AI model: Real-ESRGAN (realesr-general-x4v3) by Xintao Wang, Liangbin Xie, Chao Dong and Ying Shan, released under the BSD 3-Clause licence.
Next step: Premium application
Document Scanner
PremiumScan a paper page with your camera: crop, straighten, clean up shadows and save pages as one PDF.
2 trial uses a day · $3 day pass or $9.99 a month
Open Document ScannerMakes a small or soft photo bigger with a neural network that redraws edges, texture and fine detail, instead of only stretching the pixels. 2x or 4x, entirely in your browser: the picture is never uploaded.
Drop an image, pick 2x or 4x, and click AI upscale. The picture is processed in small overlapping parts, with a progress bar and a Cancel button. When it finishes, drag the slider over the picture to compare the original with the result, then download the PNG. The first run also loads the model (about 5 MB), which your browser keeps for next time.
The AI adds believable detail; it cannot recover what was never captured, so faces and small text can come out invented rather than exact. Pictures up to 2 megapixels are accepted; for bigger ones you can shrink first or use the plain resize (which only adds pixels). It works best on photos, product shots and artwork; clean screenshots and logos are usually better enlarged as vector or with the plain resize. Model: Real-ESRGAN (realesr-general-x4v3) by Xintao Wang, Liangbin Xie, Chao Dong and Ying Shan, BSD 3-Clause licence.
- How is this different from Image Quality Enhancer's Resize?
- Resize only makes more pixels by smoothing between the existing ones, so a blurry picture stays blurry. The AI upscaler uses a neural network (Real-ESRGAN) that redraws edges, texture and fine lines, so the enlarged picture usually looks crisper. The trade-off is that it invents plausible detail, which is not always the true detail.
- Is my photo uploaded?
- No. The AI runs inside your browser on your own device. Only the model file (about 5 MB) is downloaded from this site, once, and your browser keeps it for next time.
- Why is there a size limit, and what if my picture is bigger?
- The model runs on your device, so memory and time grow quickly with size. The limit is 2 megapixels in. For a bigger picture the tool says so and offers to shrink it to fit before upscaling, or to use the plain resize, which adds pixels without AI.
- How long does it take?
- It depends on your device and the picture size. The picture is processed in small parts, so you see real progress and a time-left estimate after the first parts finish. A small photo can take under a minute; a 2-megapixel picture can take several minutes, or more on an older phone or computer. You can cancel at any time.
- Can it fix faces, remove noise or restore old photos?
- Not as separate features. The model is a general-purpose upscaler that also softens some compression noise and blur while enlarging. It does not restore damaged faces, colourise black-and-white photos or repair scratches.
- Does it keep transparency, and what format is the result?
- Yes, transparent areas stay transparent. The result is always a PNG, named with -ai-2x or -ai-4x added to your file name.
How to use AI Image Upscaler
Drop in a picture and you see it with its size in pixels. Choose 2x or 4x (each shows the output size), then press AI upscale. The first time, your browser loads the AI model, about 5 MB, from this site and keeps it for next time. The picture is then enlarged in small overlapping parts, with a progress bar, a time-left estimate and a Cancel button.
When it finishes, drag the slider over the picture to compare the original with the result, and download the PNG. The file is named with -ai-2x or -ai-4x added, so your original is never overwritten. Start over clears everything.
When to use it
It suits a small or soft photo that needs to be larger for a print, a banner, a product listing or a slide: an old family snapshot, a thumbnail-sized product shot, a low-resolution download or a web image you took yourself. The model smooths jagged edges and redraws texture and fine lines instead of only stretching pixels.
When not to use it
The AI invents plausible detail; it cannot recover what the camera never captured. Faces, small text, numbers and logos can come out altered, so do not use it for evidence, documents, ID photos or anything where every detail must be exact. Clean screenshots, flat logos and pixel art are usually better served by the plain resize in Image Quality Enhancer. Face restoration, heavy denoising and old-photo repair are not what this model is for.
Tips for best results
Start with 2x: it is much faster and often looks more natural than 4x. Crop away empty borders before upscaling, because the time grows with the number of pixels. If a result looks too smooth or too painted, compare it against the original with the slider and try 2x, or use Image Quality Enhancer for a gentler sharpen. The model is Real-ESRGAN (realesr-general-x4v3) by Xintao Wang and colleagues, released under the BSD 3-Clause licence.
Limitations
Pictures up to 2 megapixels (about 1600 by 1250) go through the AI. A larger picture is not processed silently: you can shrink it to fit and then upscale, or use the plain resize, which only adds pixels. The result is limited to 36 megapixels, which 4x of a 2-megapixel picture (32 megapixels) stays under. It runs on your own device, so a big picture on a slow phone can take several minutes; Cancel stops it at once. 2x is made by enlarging 4x and averaging down. Transparent pictures keep their transparency, which is enlarged smoothly rather than by the AI. The result is saved as a PNG.
Privacy
The picture is processed on your device by a small neural network running in your browser, with no upload and no outside service. Only the model file (the software) is downloaded from this site, once, and your browser keeps it. The original and the result stay in the open page until you download or start over.
More from Image Tools: