Boneyard Tools

AI Image Upscaler

Make a small or low-resolution image bigger and sharper using a Swin2SR AI super-resolution model that runs entirely in your browser. Unlike plain resizing, which just blurs, the AI reconstructs detail. Pick a quality tier: Fast is Swin2SR 2x, crisp and great for clean art and screenshots, while Best is the Swin2SR 4x real-world model (BSRGAN), which is GAN-trained to handle noise and JPEG artifacts in photos and enlarges natively 4x in a single pass. The model downloads once on first use; your image is processed locally and never uploaded.

How to upscale an image with AI

  1. Drop an image in, or click to browse.
  2. Pick a quality tier (Fast 2x or Best 4x) and let the AI enhance it.
  3. Compare the result and download the larger image.

Examples

Enlarge a thumbnail

A 400x300 photo at 2x
An 800x600 image with reconstructed detail, not just stretched

Frequently asked questions

Is my image uploaded anywhere?

No. The AI model runs in your browser with WebAssembly. The image is processed on your device and never uploaded. Only the model downloads, once.

How is this different from resizing?

Resizing stretches the existing pixels, which looks blurry when you enlarge. This uses an AI super-resolution model that reconstructs fine detail and edges, so the larger image stays sharp.

What is the difference between the Fast and Best tiers?

Fast is the Swin2SR 2x classical model: crisp and quick, ideal for clean art, logos, and screenshots. Best is the Swin2SR 4x real-world model (BSRGAN), which is GAN-trained to handle noise and JPEG artifacts in real photos and enlarges natively by 4x in one pass. Each tier upscales by its own factor, so the output is 2x or 4x the original.

Why is the first run slow?

The AI model downloads the first time you use the tool, then is cached. Large images and the Best (4x) tier also take longer because there is more to process.

What works best?

For clean art or screenshots, Fast (2x) gives the crispest result. For real photos with noise or JPEG artifacts, choose Best (4x). Very large inputs may be slow or run out of memory, so start with smaller images for the Best tier.

Related tools