2600th

2026 Case file

DLSS 5 Video Player

Open-source release

I built a free, open-source Windows player that puts any video through NVIDIA’s DLSS 5 neural renderer and checks every rendered frame against the original.

My role
Creator and engineer
Period
2026
Bands
Applied AI
Player UI and demo clip from the project. Game footage from 007 First Light © IO Interactive, shown for comparison.
On this page
Project briefOPEN-SOURCE TOOL

Show what the model changed, frame by frame.

I built DLSS 5 Video Player to answer a simple question honestly: what does a neural renderer actually do to a picture? Open a video, photo or GIF, press D, and within seconds the render plays while the rest of the video renders behind the playhead, nearest frames first. The original is always one key away, on the same frame.

Comparison is the product

Most tools stop at showing the result. This one is built around checking it. Split, Wipe, Difference (the change, amplified), Side by side and 2 × 2 views share one timeline, with a mix slider, zoom to 8x and a 4x loupe. Switching views never moves the playhead. The project’s own comparison stills include a frame where the model makes the picture worse.

Saving a comparison image writes exactly what the picture shows, with a footer that records the video, frame, view, neural settings and runtime, so anyone can check the comparison later.

Rendering behind the playhead

A session collects a four-second lead, then plays rendered frames while the render continues ahead of the playhead. Rendered frames play wherever they exist; elsewhere the original plays while the render catches up. A whole-video render is reused as long as the source, runtime and settings still match.

Honest about what it is

This is a community project, not an NVIDIA product. The player calls NVIDIA NGX Super Resolution, and a separate runtime add-on observes those calls to perform experimental neural rendering. It is not an official DLSS 5 SDK integration, and motion and depth derived from video are estimates.

The neural runtime is an unsigned community build, so the release documents how to verify every file or supply the runtime yourself. Each package has a checksum, and the core package also has a build attestation. Neural rendering needs Windows, an NVIDIA RTX card and driver 610.47 or newer; the README lists the tested GPUs and measured frame costs.