Upload a gameplay video, set the parameters while watching a single frame in preview, export. The rendering runs on the NVIDIA runtime, on real RTX cards: they sit here with us, powered on, and for three minutes one of them is yours.
This is not a filter that looks like DLSS: your frames are reconstructed by a real RTX 40 or 50 with the NVIDIA pipeline, and every render comes back with a report declaring feature 18.
Freeze a frame, move the sliders and see how it turned out right away. No ten-minute export only to find out at the end that the intensity was too high.
The GPU is yours for the whole slot: as many previews as you want, and one export. If they are all busy you wait your turn, because these are real machines in a room.
If you like a frame you can put it in the catalogue, with the original next to the DLSS 5 version and the exact parameters you used.
The pipeline does not know what it is looking at: it sees frames. This one comes from an animated series, not a game, and goes through the same runtime with the same result.
DLSS OFFDLSS 5 ONArcane (Riot Games / Netflix) · 1280×720 in and out: no upscaling, just the neural filter. · drag to compare
How much neural reconstruction to apply to the image.
Local contrast in shadows and highlights.
Micro-detail on foliage, fabric, dense surfaces.
Skin handled separately, so faces don't turn to plastic.
The weights the model starts from, before your corrections.
The character of the image at identical parameters.
The model variant used for the reconstruction.
From no upscaling to the most aggressive reconstruction.
Resolution and codec of the file you take home.
These comparisons were published by the people who came through the Studio. Every entry carries the source, the DLSS 5 version and the exact parameters it was produced with.
DLSS 5 only runs on NVIDIA RTX 40 and 50 GPUs. They are expensive, not everyone owns one, and for most people trying it would mean buying first and finding out afterwards whether it was worth it. On some games, setting up a serious test is awkward regardless of the hardware.
We have the GPUs, and it seems only fair to let you use them: upload a frame, move the parameters and watch what actually happens, buying nothing and installing nothing. It is a free project, for education and outreach. There is no paid tier and donating gets you nothing extra: what comes in keeps the cards powered and, when it adds up, buys another one.
DLSS5 Studio does not approximate DLSS: it sends your frames to a real RTX GPU running the NVIDIA Neural Rendering pipeline, and every render produces a report confirming that feature 18 ran. That is why capacity is finite — the GPUs are physical machines, and when they are all busy there is a queue.
When the slot expires your material is deleted from the machine: we do not keep it anywhere.
Take a slot