AI Image Upscaler
LOCAL IMAGE UPSCALER
Use tiled lightweight ONNX inference for 2×/4× output. Ordinary resizing is a separate, explicitly non-AI mode.
For best results use an image under 10 MB with moderate dimensions.
AI mode processes at most 2.5 million input pixels in tiles with a native 3× model; output is capped proportionally at 4096 pixels per side. It is not a guarantee of restored detail. Ordinary resizing is explicitly non-AI.
Using AI Image Upscaler
Use tiled lightweight ONNX inference for 2×/4× output. Ordinary resizing is a separate, explicitly non-AI mode.
The image and model inference stay in this browser; the source is not uploaded.