RealScaler
An offline AI image and video upscaler for Windows that uses Real-ESRGAN and supports DirectX 12 compatible AMD, Intel, and NVIDIA GPUs.
RealScaler is a local AI upscaler for people who want to enlarge photos and videos and reduce noise without uploading their media. Its Windows desktop interface uses Real-ESRGAN models, and processing works offline once the required files are available on your PC.
- Images and video: Upscale still images and videos, with support for JPG, PNG, TIFF, WebP, and HEIC images, plus MP4, MKV, AVI, and MOV video files.
- Control over the result: Resize the source before upscaling and blend the original with the AI output to adjust how much the processed image or video differs from the source.
- Resumable video jobs: Stop a video upscale and resume it later instead of completing a long job in one session.
- GPU support: Use compatible AMD, Intel, or NVIDIA hardware, including multiple GPUs, with automatic image tiling to process images that would otherwise exceed available GPU memory.
RealScaler is open source under the MIT license and runs on Windows 10 and 11. It requires at least 8 GB of RAM and a DirectX 12 compatible GPU with at least 4 GB of VRAM. Its processing backend uses ONNX Runtime with DirectML, and image and video processing stays on your PC.
