Image Upscaling and Restoration

A free-forever, unlimited Real-ESRGAN 4x image upscaling and restoration tool with general upscaling, photo enhancement, compression repair, line-art enhancement, and pre-upscale sharpening. Images are processed locally and never uploaded.

Category

Image Tools

How to Use

  1. Drag or select a PNG, JPG, or WebP image up to 1080p
  2. Choose General Upscale, Photo Enhance, Compression Repair, or Line / Illustration Enhance
  3. Adjust pre-upscale sharpening for the source image; the default is 35%
  4. Auto Best selects a suitable model from the mode and browser memory budget
  5. Use it free forever with unlimited runs; images are processed locally in your browser

Examples

  • Upscale a web image 4x

    Input: 1200x800 image | General Upscale | 35% pre-sharpen

    Output: 4800x3200 PNG output

  • Repair a lightly compressed screenshot

    Input: 900x600 screenshot | Compression Repair | 45% pre-sharpen

    Output: 3600x2400 PNG output with enhanced detail

  • Run without WebNN

    Input: Browser does not expose navigator.ml

    Output: Automatically uses WASM compatible mode

FAQ

Is my image uploaded to a server?
No. Decoding, tensor conversion, model inference, and export all happen locally in your browser. Nothing is uploaded, so your image stays absolutely safe.
Are there usage limits?
No. The image upscaling and restoration tool is free forever with unlimited runs.
How are the processing modes different?
General Upscale keeps the original flow; Photo Enhance lightly boosts contrast and saturation; Compression Repair applies a subtle smoothing pass before upscaling; Line / Illustration Enhance boosts line contrast and prefers the lightweight model.
How much pre-sharpening should I use?
The 35% default works for most low-resolution images. Text, line art, and screenshots can often use 45%-60%; noisy photos or heavy JPEG artifacts usually look better at 15%-30% to avoid sharpening the noise.
Which model is used by default?
Auto Best is the default. The tool combines the selected processing mode, quality order, and current browser memory budget to pick the best model that fits.
Can it run when WebNN is unavailable?
Yes. Auto Fastest tries WebNN NPU first, then WebNN GPU, then WASM. You can also manually choose a backend.

Related tools