Changelog
- Added interactive Compare workspace, introducing a dedicated tab to compare images and videos using Side-by-Side, Split View slider, fast Cycle switching, and Difference Heatmap modes, complete with session saving and media export.
- Added Windows ML AI acceleration, bringing native multi-vendor GPU support across NVIDIA, AMD, and Intel graphics cards with automatic DirectML fallback and smart disk caching for faster startup times.
- Added AI Inferencing Framework selection, giving users direct control to choose between Auto, Windows ML, or DirectML backends in the Upscale settings and CLI.
- Improved Image Upscaling responsiveness, keeping VisionNext and SharpNext models warm in memory for up to 120 seconds to eliminate reload delays between consecutive previews and renders.
- Upgraded multimedia processing engine to FFmpeg 9, providing unified, rock-solid audio-video decoding, synchronized playback scrubbing, and reliable high-quality video encoding.
- Improved DLSS Neural Rendering engine architecture, directly interfacing with official and community models to resolve false-positive antivirus detections while guaranteeing identical, bit-perfect visual quality.
- Improved ReShade effects and AI Depth processing, delivering better temporal stability, updated MoGe-2 depth estimation, and smoother rendering across complex shader chains.
- Improved application navigation and interface design, transforming the Models manager into a convenient modal dialog over the active workspace, refining dropdown controls, and expanding localized interface support across 11 languages.
