Veyra: Real-Time GPU Video Enhancement and Frame Generation for Windows
An overview of Veyra, an open-source C++ and DirectX 12 video enhancer bringing real-time super resolution, multi-layer denoising, and frame generation to local
Upscaling resolution or smoothing frame rates in real time when routing local video files or game console footage into a PC demands substantial hardware throughput. Veyra (GitHub: Likely7/Veyra-NRVideo) is an open-source GPU video enhancement utility built on C++ and DirectX 12 that processes local media files, hardware capture cards, and remote streaming feeds through a unified real-time pipeline.

Image source: https://github.com/Likely7/Veyra-NRVideo
Going beyond standard media playback, Veyra integrates AI-assisted Super Resolution (SR), layered Noise Reduction (NR), and Frame Generation (FG) under a single runtime engine. With release 2.1.5 available as of November 2026 and over 79,000 cumulative portable downloads recorded on GitHub Releases, the project has gained steady traction among console streamers and video enthusiasts.
Core Features and Input Ingestion Support
Veyra treats static video files and live external feeds identically inside its node-based image processing architecture.
The software accepts several distinct input pipelines:
- Local Video and Images: High-definition video containers (including MP4 and MKV) and image files can be loaded, filtered in real time, and exported through an editable transcode queue with subtitle and multi-audio support.
- Hardware Capture Cards: Users can route DirectShow capture cards directly from consoles like PS5 or Xbox, with automatic checks for driver resolutions, pixel formats, and timestamp consistency.
- Game Streaming Feeds: Alongside direct screen capture via Windows Graphics Capture (WGC), Veyra accepts streaming feeds from Moonlight and Sunshine servers, enabling real-time upscaling and smoothing for remote gaming sessions.
The user interface provides dedicated layouts including Cinema mode, a Professional List view, and an executable Node graph where filters, color controls, and scaling parameters can be wired and saved as session presets.
Super Resolution, Layered Denoising, and Frame Pacing
At the architectural core of Veyra is a GPU-accelerated graphics and inference pipeline designed to minimize per-frame latency.
First, the Super Resolution engine supports both NVIDIA RTX Video SR / DLSS and AMD FSR (FidelityFX Super Resolution). Starting with release 2.1.2, Veyra incorporates dual SR passes, allowing an initial upscaling step followed by controlled downsampling or secondary filtering. For edge cleanup, users can insert post-processing anti-aliasing (FXAA, SMAA, or MLAA) into the chain. Benchmark fixtures published in the repository note runtimes on an RTX 5070 of approximately 0.15 ms for FXAA and 0.30 ms for SMAA/MLAA in standard 4K workflows.
Second, the Noise Reduction subsystem supports hierarchical multi-layer controls with strength scales up to 5. Parameters govern neutral tint, chroma, shadow/highlight retention, and temporal stability across frames. Dedicated model paths are separated across NVIDIA architectures, AMD RX 9000, and experimental RX 7000 series profiles.
Third, frame generation and latency management work together to keep live streams responsive. The engine features full-frame deduplication to strip redundant frames from capture streams, adaptive VSync sync regulation, and NVIDIA Low Latency settings (Off, On, and On + Boost) to counter input lag during interactive gameplay.
Requirements, Distribution, and Practical Caveats
Veyra is distributed under the GNU General Public License v3.0 (GPL-3.0) directly via GitHub Releases as precompiled portable archives.
- System Compatibility: The tool strictly requires Windows 11 x64 and a modern DirectX 12 compatible graphics card.
- Architecture-Specific Packages: Release archives are split by vendor (for example,
Veyra-2.1.5-NVIDIA-win64-portable.zipversus AMD builds), requiring users to extract the exact binary distribution matching their installed GPU. - VRAM Overhead: Running simultaneous high-resolution dual-pass SR, multi-layer denoising, and frame generation places heavy demands on video memory. The author's primary test bench uses an RTX 5070; lower-tier GPUs will require lowering inference resolution percentages and disabling secondary passes to prevent frame drops.
For users seeking to enhance console capture streams or run customizable upscaling pipelines on Windows without proprietary licensing, Veyra offers a modular, high-performance open-source framework.
Sources
- Veyra GitHub Repository: Likely7/Veyra-NRVideo