Veyra: Real-Time GPU Video Enhancement for Capture Cards and Streaming
You've got a PS5 or Xbox hooked up to a capture card, and the footage coming through looks a little soft. Maybe you're streaming a PC game and want to clean up the picture without tanking your frame rate. Either way, you're stuck stacking filters in OBS or just accepting the quality you get. Veyra is an open-source Windows app that puts super resolution, denoising, HDR, and frame generation into a single real-time GPU pipeline for local video, capture cards, and game streaming.
What It Does
Veyra is a Windows 11 application written in C++ on top of DirectX 12, and it targets both NVIDIA and AMD GPUs. It takes video from local files, capture cards, or game streaming and runs it through a real-time enhancement chain on your GPU. The README frames the core idea simply: one engine, any source, real-time enhancement.
The pipeline covers super resolution (including a dual SR mode and SR-then-downsample), denoising, HDR handling, and frame generation, along with capture and audio features. It's open source under GPL-3.0, and the project ships as a portable package you download from GitHub Releases. There's also a fairly detailed release history — the README references versions 2.1.1, 2.1.2, and 2.1.6, each with specific fixes and additions, which tells you this thing is actively maintained rather than dumped and abandoned.
Why It's Cool
-
It treats your capture card as a first-class source. A lot of enhancement tools assume you're working with a file on disk. Veyra is built around the idea that the video is arriving live from a PS5, an Xbox, or a PC stream, and it handles capture alongside the enhancement chain.
-
The effect costs are actually published. The README lists measured numbers on an RTX 5070: FXAA around 0.15ms, SMAA/MLAA around 0.30ms at standard 4K anti-aliasing fixtures, and a final single-face mesh beauty pass at roughly 0.93ms for 1080p and 1.42ms for 4K. Those are honest, small numbers — and the project is upfront that costs vary with content and hardware, and that no local AMD speedup percentage was measured. That kind of restraint is rare.
-
The release notes are granular. The 2.1.1 notes mention processed-picture recording with audio, 90-degree playback rotation, per-layer NR percentage controls, event-driven refresh/VRR/HDR queries, and AMD HDR+NR handling. The 2.1.6 notes add instant replay on F9, microphone mixing with separate tracks, all-effect exclusion, and VFR dedup with 60-to-30 conversion. These are the details of a tool someone actually uses daily, not a demo.
-
It's portable and documented. You download a package, and there's an English wiki covering installation, capture card compatibility, and troubleshooting, plus a Chinese knowledge base. For a tool that touches capture hardware, that documentation matters more than the feature list.
How to Try It
- Head to the releases page and grab the latest portable package.
- Check the User Guide / Wiki before you plug anything in — it covers installation, software setup, and capture card compatibility, which is where most people will hit friction.
- You'll need Windows 11 and a DirectX 12-capable NVIDIA or AMD GPU. The README doesn't list a minimum GPU model, so check the wiki for specifics.
- If something breaks, the issue tracker is the place to report it, and the repo also has a Simplified Chinese README if that's useful to you.
The repository is at github.com/likely7/veyra-nrvideo.
Final Thoughts
Veyra is aimed at a specific kind of person: someone capturing console or PC gameplay on Windows who wants better picture quality without leaving their existing setup behind. It's not a general-purpose video editor, and it won't run on macOS or Linux. But if you're in that niche, the combination of live capture, GPU-side enhancement, and published performance numbers makes it worth a look. The steady stream of releases suggests the maintainer is still listening to users, which is usually the best sign for a project like this.
Follow @githubprojects for more developer tools and open source projects.