Local AI image upscaler for Windows

What 2× and 4× mean for your files

The scale applies to both dimensions. For an 800 × 600 source, 2× produces 1600 × 1200 pixels and 4× produces 3200 × 2400 pixels. A 4× result has 16 times as many total pixels as the source, so it can also need much more memory and disk space.

Upscaling creates additional pixels, but it cannot guarantee recovery of detail that is absent from the original. Inspect the result at full size, especially around text, faces, and fine patterns.

How to upscale images in Pixel Up

1. Open Upscale and drop in one or more PNG, JPG, WebP, BMP, or TIFF images.
2. Choose a photo or illustration model. The photo models support 2× or 4×; the illustration model supports 4×.
3. Select PNG, JPEG, or WebP output and choose an output folder.
4. Start the queue. When a file finishes, compare the original and result with the split view, zoom, and pan controls.

Use the batch workflow guide if you are processing many images or want to rerun individual items with different settings.

GPU, CPU, and large images

Pixel Up uses DirectML acceleration when a compatible GPU is available. If that path fails, it can retry on the CPU. Upscale can reduce tile size on weaker hardware. CPU processing and very large images may take longer; there is no fixed speed that applies to every PC.

For a large source image, try 2× first, close memory-heavy programs, and check the output dimensions before starting a 4× batch. See the support guide for hardware and memory troubleshooting.