Real-ESRGAN vs Upscal

Real-ESRGAN is a free, open-source project developing algorithms for image and video restoration. It offers pretrained models for general images, anime images and animation video, including AnimeVideo-v3 and smaller anime models. Image inputs include JPG, PNG and WEBP; inference can handle alpha-channel, grayscale and 16-bit images, and outputs can be JPG, PNG or WebP. Portable inference supports scale ratios of 2, 3 or 4, with 4 as the default; the Python implementation also offers arbitrary output scaling. Portable NCNN executables are available for Windows, Linux and macOS, with required binaries and models included and no CUDA or PyTorch environment needed. These executables do not include every Python feature, including outscale. Python use requires Python 3.7 or newer and PyTorch 1.7 or newer. Online inference is available through an ARC demo and Colab demos. GFPGAN is integrated for face enhancement, and the training code supports fine-tuning on user data or paired data. The project uses synthetic training data and is released under the BSD-3-Clause license.

Upscal uses built-in AI models to enhance low-resolution photos and videos on a device. Its image tools target blurry, compressed and old photos, aiming to improve edge sharpness, texture and detail. Video enhancement works frame by frame, and built-in super-resolution models can enlarge images and videos up to 8×. Batch processing is supported, with output formats including JPG, PNG, WebP, MP4, MOV and M4V. Upscal is available on iOS, macOS and Windows. Once the app and required resources are installed, normal image and video enhancement can run offline. The privacy policy says processed media is handled locally rather than sent to an external processing server, though anonymous technical diagnostics may be transmitted, including app and operating-system versions, device category, crash traces and error logs. The free plan costs nothing, has no subscription and includes core image and video enhancement with offline processing; no account is required. The maker permits commercial use when users follow applicable laws and have the necessary rights to the input content. Results vary with source quality and device capability, and enhancement cannot perfectly recover detail that was never captured.