Merserk/dlss5-visual-enhancer v14.0

Visual Enhancer v14.0

on GitHub

Changelog

• Significantly improved DLL and native processing performance across Neural Rendering, DLSS, RTX Video, Frame Interpolation, previews, and video pipelines, reducing unnecessary memory transfers, format conversions, buffer allocations, and processing overhead for faster overall rendering.
• Added a new Grain processing stage for images and videos, with adjustable Grain Amount, Grain Size, Color Grain, Tonal Response, Grain Seed, and optional Animated Grain for video. Grain can be reordered within Neural Rendering and is fully supported by presets, previews, Stage by Stage processing, and Rolling Cache.
• Significantly improved single-stage video processing, allowing supported Neural Rendering, DLSS Super Resolution, RTX Super Resolution, RTX Video HDR, and Frame Generation workflows to render directly to the final output without creating an unnecessary intermediate video and performing an additional export pass.
• Improved GPU-resident video processing with more direct CUDA processing paths and automatic NVIDIA NVENC selection for compatible H.264, H.265, and AV1 workflows, reducing unnecessary transfers between GPU and system memory.
• Significantly improved Neural Rendering video preview performance with a new direct CUDA preview path that can seek into compatible source videos and process only the requested preview range without first preparing a complete intermediate clip.
• Improved DLSS Super Resolution and RTX Super Resolution previews with faster timeline seeking and more direct processing from the selected source position, reducing preparation time when generating short previews.
• Improved single-frame Neural Rendering previews with in-memory processed-frame caching and new direct frame-processing paths, allowing repeated previews at the same position and settings to reuse completed work instead of unnecessarily processing the frame again.
• Improved Rolling Cache performance with more processing remaining on the GPU, reduced intermediate conversions, and more efficient handling of Neural Rendering, Coloring, Scaling, Sharpening, Grain, scene analysis, stabilization, and other compatible processing stages.
• Changed Rolling Cache (5 GB) to the default video processing workflow, providing a more efficient default for multi-stage Neural Rendering while keeping Stage by Stage processing available when needed.
• Significantly improved Frame Interpolation performance with optimized decoding, CUDA frame handling, color processing, output surfaces, encoding, and fallback behavior, reducing unnecessary CPU processing and improving throughput in supported GPU-accelerated workflows.
• Improved RTX Video and Upscale processing performance with more efficient host and CUDA frame handling, additional native pixel-format processing paths, reduced synchronization overhead, improved decoder fallback behavior, and more direct transfer of processed frames to the encoder.
• Improved application responsiveness when loading media and large batches, reducing unnecessary work on the interface thread and improving file scanning, metadata loading, media probing, decoder cleanup, cancellation handling, and behavior with slow or problematic files.
• Improved processing speed reporting and output estimates, providing more accurate FPS measurements and better processing-time estimation across RTX Super Resolution, Frame Generation, and other GPU-accelerated workflows.
• Improved presets and settings compatibility with support for the new Grain stage, updated processing defaults, stage ordering, and automatic migration of existing Neural Rendering configurations.
• Improved overall performance, stability, preview reliability, export handling, GPU processing, decoder fallback behavior, media switching, cancellation, memory management, and interface responsiveness across the application.