Merserk/dlss5-visual-enhancer v8.0

DLSS 5 Visual Enhancer v8.0

on GitHub

Changelog

• Added the new Neuroframe Engine, a completely new processing engine designed specifically for images and videos with full GPU acceleration:
  • Images: v8.0 is 92.7% faster than v7.0 (13.7× speedup).
  • Videos: v8.0 is 71.4% faster than v7.0 (3.5× speedup).
• Added Multi Pass for DLSS 5 Neural Rendering, allowing 1 to 4 Neural Rendering passes to be applied for stronger enhancement and additional detail refinement.
• Added significantly more Neural Rendering controls, including NR Color Strength, Tone Preservation, Face/Skin Protection, Grain Preservation, Shimmer Suppression, Mask Feather, and Custom NR Mask.
• Added Custom NR Mask, allowing Neural Rendering to focus on selected areas of an image or video instead of affecting the entire frame equally.
• Added Mask Feather with up to 128 pixels of feathering, allowing custom masks to blend smoothly into surrounding areas.
• Added Detail-Only mode, allowing Neural Rendering detail enhancement to be applied while keeping the original colors and tones much closer to the source.
• Added Shimmer Suppression for video and Live processing, helping reduce flickering, unstable fine details, and shimmering between frames.
• Massively upgraded Live mode with the new Neuroframe Engine and expanded DLSS 5 Neural Rendering controls, allowing supported parameters to be changed and applied while the video is playing.
• Expanded real-time controls in Live mode with Multi Pass, color strength, tone preservation, face and skin protection, grain preservation, shimmer suppression, masks, and other new v8.0 Neural Rendering settings.
• Reworked the Live GPU pipeline so supported video can remain in GPU memory through decoding, DLSS 5 Neural Rendering, and NVIDIA encoding, reducing unnecessary transfers and improving processing efficiency.
• Added a new VRAM and RAM Processing Engine Path option, allowing Neural Rendering to use the new GPU-focused VRAM path or switch to the RAM path when needed.
• Added automatic real-time Neural Rendering previews, allowing supported settings to generate an updated preview without starting a full render.
• Added a dedicated Neural Rendering image Preview mode, allowing DLSS 5 settings to be tested on an image before creating a production output.
• Added full-size image preview quality in Settings for inspecting Neural Rendering and Upscale results at their complete output resolution.
• Completely reworked image batch processing to prepare upcoming images and save completed outputs alongside GPU processing, reducing idle time between images and improving overall batch throughput.
• Completely reworked batch processing outputs across Neural Rendering, Upscale, and Frame Interpolation. Single-file jobs now provide a direct Save button, while multi-file jobs use Save as ZIP.
• Changed ZIP creation so batch archives are generated only when Save as ZIP is requested instead of automatically after every render, reducing unnecessary waiting and disk activity.
• Improved batch processing safety so completed outputs remain available when processing is stopped, while outdated previews and download actions are automatically invalidated when the input changes.
• Reworked Neural Rendering video processing with a dedicated GPU-resident pipeline for NVIDIA NVENC output, reducing CPU and memory transfers between decoding, rendering, and encoding.
• Improved MP4 and MOV processing by combining video encoding, source audio, and metadata into the same finalization path, removing an unnecessary additional remux step in supported jobs.
• Improved single-frame video previews by avoiding unnecessary additional frame decoding and final remux processing.
• Reworked the Neural Rendering Scale control with Source, 75%, 50%, and 25% processing sizes, providing a simpler way to balance processing speed and detail.
• Improved Skin Structure behavior so making an explicit skin adjustment automatically activates the required masking behavior while still respecting manually selected Mask Off settings.
• Expanded output metadata to include important v8.0 settings such as Multi Pass, composition controls, Custom NR Mask usage, Mask Feather, processing memory path, and output scale.
• Improved settings preset compatibility with automatic migration of older presets to the new v8.0 controls, including Multi Pass, composition settings, Shimmer Suppression, and the new processing engine path.
• Removed legacy Neural Rendering preset and DLSS model preset controls that were not actually applied by the Neural Rendering engine, making the interface simpler and less confusing.
• Improved ProRes Proxy quality handling so higher quality selections now have a real effect on output quality.
• Added a real Preview workflow to RTX VSR Upscale, allowing the selected VSR quality and sizing settings to be tested before saving the final image.
• Updated default output naming with clearer _Neural_Rendering, _Frame_Interpolation, and _Upscale suffixes instead of the older technical naming.
• Improved application diagnostics with cleaner session logging and dedicated failure details, making processing errors easier to identify.
• Improved temporary file and preview management with an isolated application cache, automatic cleanup of stale files, and safer handling of previews left behind after interrupted sessions.
• Refreshed the Neural Rendering interface with cleaner control grouping, more stable Image and Video mode switching, improved preview behavior, and a more compact main navigation layout.