AI Denoiser & Upscaler
Use Real-ESRGAN to reduce photo noise and compression artifacts, then export PNG at ×1, ×2, or ×4. The model loads only when you run it; image inference stays local.
How to Use
- 1
Drag and drop or click to upload your image
- 2
Select the output mode (×1 Enhance / ×2 Upscale / ×4 Upscale)
- 3
Click the "Upscale with AI" button
- 4
Compare Before and After to see the result
- 5
Click "Download Image" to save
Key Features
- ✓Real-ESRGAN AI model for super-resolution upscaling
- ✓Up to 4× upscale with natural detail enhancement
- ✓WebGPU-accelerated for fast processing (WASM fallback included)
- ✓No account is required
- ✓The model and runtime download first; image inference then runs in the browser without uploading the selected image
- ✓Download in high-quality PNG format
FAQ
Q. What kind of images work best?
A. Low-resolution photos, smartphone screenshots, and older images benefit the most. The AI naturally fills in details while upscaling.
Q. How long does processing take?
A. Depending on the device, WebGPU/WASM backend, quality, and image, processing can take seconds to minutes. Progress is shown and the job can be canceled.
Q. Are my images uploaded to a server?
A. No. All processing happens entirely in your browser. Your images are never transmitted over the internet.
Q. What image formats are supported?
A. Input supports JPEG, PNG, and WebP up to 25 MiB. Output is PNG and limited to 4,096 px per side and 16 megapixels total.
Q. Which mode should I use?
A. "×1 Enhance" improves quality without changing size. "×2 Upscale" doubles the resolution. "×4 Upscale" maximizes enlargement for the best detail.
How It Works: Real-ESRGAN Super Resolution
This tool uses Real-ESRGAN (Real-World Enhanced Super-Resolution GAN), an AI model specialized in 4× image upscaling that naturally reconstructs fine details.
All processing runs inside a Web Worker using ONNX Runtime Web. WebGPU-enabled browsers use GPU inference for maximum speed, with automatic WASM fallback for other environments.
The approximately 5 MB model is downloaded when you first click the enhancement button, then cached by the browser. Standard mode tiles a working image up to 256 px on its longest side; High Quality uses up to 768 px before compositing to the selected output size.
Privacy & Security
All image processing happens entirely on your device. No image data is ever sent to any server.
AI inference runs in the browser and image files are not uploaded. Model/runtime downloads occur, and device-side risks such as untrusted extensions are outside this tool's control.