Changelog
- Added two next-generation image upscaling models supporting 1x, 2x, 4x, 6x, and 8x scaling factors:
- VisionNext: Latest-generation architecture with fine-structure restoration, recommended for maximum visual quality.
- SharpNext: Multi-step restoration model with reverse noise prediction, focused on noise-free clarity and sharp edges.
- Added a dedicated Models Manager tab, providing a centralized interface to download, inspect, and remove local model packages across supported features, including Image Upscaling and ReShade depth processing.
- Added GPU device selection for Image Upscaling, allowing you to explicitly assign AI upscaling workloads to a preferred graphics adapter on multi-GPU systems.
- Added direct conversion mode, allowing images and videos to be processed and exported with all enhancement stages disabled so the pipeline can be used for standalone format, container, and codec conversion.
- Updated ReShade depth estimation models to deliver higher spatial accuracy, cleaner object boundaries, and more precise depth-map generation for depth-aware shaders.
- Fixed Ray Reconstruction pipeline compatibility, resolving an issue where enabling Ray Reconstruction prevented other processing features and effects from running in the same workflow.
- Fixed AMD CAS and NVIDIA NIS execution issues on certain GPUs where sharpening and scaling passes failed to initialize or apply properly.
- Fixed FFmpeg command-line failures (Windows Error 206) triggered when oversized embedded video comment metadata exceeded the operating system's command-line length limit.