Skip to content

Repository files navigation

DLSS 5 Video Player

Process a photo, GIF or video, replay its cached neural result, and compare it with the original at the same timestamp. A native Windows player for local media and public YouTube videos, with optional DLSS Super Resolution during playback.

Get started · Usage guide · Build from source · Troubleshooting

Watch the 30-second DLSS 5 Video Player demonstration

Watch the 30-second demo — actual player footage from The Witcher IV. Compare the same paused frame with Neural Rendering Off and On, then watch uninterrupted playback with it left On. 1080p H.264 MP4; intentionally silent. Capture details and edit source.

v0.14.1 experimental release. Download the Windows build with recent-video caching, settings snapshots, MKV export and highest-bitrate YouTube selection. See the changelog.

Important

This is an experimental community project, not an official NVIDIA DLSS 5 integration. The optional neural runtime uses modified/unsigned third-party components. Hardware verification for v0.14.1 used an RTX 5090. See runtime details and notices.

Why use it?

  • Compare the same moment. Switch between original and cached neural video without changing the playback timestamp.
  • Replay recent videos. The last five distinct videos persist across launches. Acquired YouTube sources and neural renders are reused after validation.
  • Keep experiments consistent. Neural settings are saved with each render and included in its cache identity. A settings change triggers a new render.
  • Take the result with you. Export processed photos as PNG/JPEG, animations as GIF, or videos as MP4/MKV. MKV preserves available source audio, compatible subtitles and chapters without re-encoding.
  • Keep playback clear. Fullscreen hides the menu and controls until the mouse moves. Writable portable installs keep their cache beside the EXE.
  • Choose playback upscaling separately. Apply optional DLSS Super Resolution to either view at 1440p or 2160p. The neural cache retains source resolution.
  • Start with a game trailer. Six official trailers featuring human characters, each under three minutes, are available under File > Game trailers.

Get started

  1. Use Windows x64 with a D3D12-capable NVIDIA RTX GPU and a suitable NVIDIA driver. The experimental neural layout requires the separately supplied runtime described in setup.
  2. Build the current source, or download the v0.14.1 package if you have repository access. GitHub's source ZIP is not a runnable package.
  3. Extract a packaged build into a new folder and keep all helpers and neural-runtime/ intact. Launch DLSSVideoPlayer.exe.
  4. Open a local file (Ctrl+O), paste a public YouTube URL (Ctrl+L), or choose File > Game trailers.
  5. Let preparation finish. Press D to compare neural and original views; enable DLSS Upscaling separately if desired.
  6. Reopen through File > Recent videos, or choose File > Export processed media to save the prepared result.

The publishable core package has fewer capabilities than the complete experimental layout. Build inputs and package contents are explained in Building.

Everyday controls

Action Control
Open a local file / YouTube URL Ctrl+O / Ctrl+L
Play or pause Space
Compare original and neural views D
Seek / step a paused cached frame Timeline or Left / Right; . to step
Volume / mute Volume control or mouse wheel; M to mute
Fit or fill / fullscreen A / F11
Image adjustments Ctrl+E
Replay / export File > Recent videos / Export cached video

Volume, mute, fit/fill, comparison view, upscaling preference/output, YouTube quality and image adjustments are saved across launches.

Setting Fresh-install default
Neural Rendering On; prepare or reuse a validated cache
DLSS Upscaling Off; 1440p output selected, with 2160p available
YouTube source quality Auto: prefer exact 1080p, otherwise highest available up to 4K
Frame Generation Unavailable; no backend is implemented

Manual YouTube choices are 1080p, 1440p and 2160p. Each selects the highest advertised video bitrate at that resolution, across available codecs and containers. Source quality and playback upscaling are separate. See cache, settings and export details.

See the trailer list and runtimes for the released and upcoming AAA games in File > Game trailers.

Screenshots

Actual Windows captures of the v0.13.0 feature implementation on an RTX 5090. Images show interface states, not image-quality benchmarks. Open an image to inspect it at full size.

Recent videos and export

File menu with Recent videos and the cached-video export command

Matched original and neural face views

Cached neural view of Ciri in The Witcher IV daylight village sequence

Original video at the same paused timestamp as the neural view

Runtime upscaling is off in both comparison captures. These document a synchronized toggle, not a claim that every source gains visible detail.

Inspect the unscaled same-frame face crops.

Start screen

Start screen with local file and YouTube URL actions

Capture details and footage attribution.

Limits to know

  • Neural rendering prepares the complete video before playback. Processing time and results vary by source and hardware; this is not real-time neural rendering.
  • Motion and depth guides are estimated from video. Artifacts are possible; RTX 40 neural compatibility has not been hardware-verified for this build.
  • Export copies the cached 8-bit video. Playback adjustments and runtime upscaling are not baked in; export does not restore HDR or lost source precision.
  • Compatible source subtitles remain separate in export. In-player subtitle display, burn-in, queues, HDR processing and durable render resume are not included.
  • YouTube supports public, non-DRM videos without login. Availability and regional access can change. Local playback remains available.
  • Recent history is limited to five videos, not a disk-size quota. Large videos can consume substantial space; use Advanced > Clear Neural Cache when needed.

Development and help

The stack is C++20, Win32, Direct3D 12, FFmpeg and NVIDIA NGX. An isolated helper hosts the experimental neural runtime; playback upscaling runs in the player.

For a bug report, include the build/revision, GPU, driver, source dimensions and frame rate, reproduction steps and relevant log excerpts. Remove private paths and signed media URLs before sharing logs. Changes should preserve the source resolution and neural-validation contracts; run CTest and relevant hardware smoke checks before proposing a renderer change.

Acknowledgments

Built upon DLSS 5 Video Player by Jessica Natalia Mods. Credit to the original project and its contributors for the foundation this project builds on.

Project source is MIT-licensed. Third-party binaries, game footage and trademarks retain their own terms; see third-party notices.

About

Experimental Windows video player for DLSS 5-style neural rendering, cached original/neural comparison and optional runtime DLSS upscaling. Local files and YouTube. C++20 / Direct3D 12. Unofficial integration; NVIDIA component rights belong to NVIDIA.

Topics

Resources

Contributing

Security policy

Stars

39 stars

Watchers

1 watching

Forks

Releases

Packages

Contributors

Languages