How image upscaling works
Upscaling increases an image's width and height in pixels. Ordinary interpolation estimates colours between known pixels. It can make edges smoother, but it does not discover the exact detail that was never captured.
Super-resolution uses a trained model to infer plausible textures and edges. That can improve perceived sharpness, but it may also change tiny details. This build does not label its standard interpolation as AI; AI Quality remains disabled until a tested local model is included.
2× or 4×?
Start with 2× when the source is already usable. A 4× result has 16 times the source pixels and needs substantially more memory.
Pixel art and transparency
Nearest-neighbour scaling keeps block edges crisp. PNG and WebP can preserve alpha; JPG output uses your chosen background.
DPI and browser limits
DPI changes calculated print size, not detail. Large decoded canvases need much more memory than compressed source files.
Privacy and practical limitations
Processing stays on your device. Very compressed photos, unreadable text, and heavily blurred edges cannot be reconstructed exactly. Mobile browsers may enforce lower canvas and memory limits than desktop browsers.
Frequently asked questions
Can I enlarge a small image without losing quality?
Enlargement adds pixels but cannot guarantee missing detail. Standard Smooth reduces common jagged edges, while the result still depends on the source.
What happens to pixels when an image is enlarged?
The output contains more pixels. At 2× each edge, total pixels grow by 4×; at 4× each edge, they grow by 16×.
Does AI upscaling restore original detail?
Not exactly. A genuine model may reconstruct plausible detail, but it cannot know every missing original pixel.
Is this a real AI upscaler?
AI Quality is currently disabled because this build does not ship a tested local super-resolution model. Standard Smooth is clearly identified as interpolation.
What is the difference between 2× and 4×?
4× makes each edge twice as large as 2× and creates four times as many output pixels as a 2× result, so it uses much more memory.
Why can a 4× upscale use 16 times more pixels?
Pixel count is width multiplied by height. Multiplying both dimensions by four gives 4 × 4, or 16 times the pixels.
Can I upscale text and logos?
Yes, but enlargement cannot reconstruct unreadable lettering. Compare at 100% and try Pixel Art for hard-edged graphics if smoothing looks soft.
Can I upscale pixel art?
Yes. Pixel Art mode uses nearest-neighbour scaling and disables smoothing to preserve square pixel edges.
Can I preserve PNG transparency?
Yes. PNG and WebP output retain canvas alpha. JPG has no alpha, so the tool flattens transparent areas onto your selected background colour.
Does changing DPI improve quality?
No. DPI relates pixels to physical print size. Pixel dimensions determine the image information available.
Can I upscale multiple images?
Yes. Add several images, process them sequentially for safer memory use, download individually, or package completed files in a ZIP.
Are images uploaded?
No. Standard Smooth and Pixel Art processing use browser image and canvas APIs locally. Filenames, previews, and outputs are not sent to a server.
What is the largest output size?
The tool enforces 16,384 pixels per edge and 64 megapixels overall, then applies an additional conservative memory warning. Browser and device limits may be lower.
Does the tool work on mobile?
Modern mobile browsers can use the standard modes, but available canvas size and memory are usually lower, so smaller outputs are safer.