Open-source project
wilsjo2/OptiScaler-DLSSNR-PreSR-Multipass avatar
wilsjo2/OptiScaler-DLSSNR-PreSR-Multipass

OptiScaler Neural Rendering ships one command, a separate DLL, and three prereleases

A game mod that uses NVIDIA AI to change lighting, detail and colour, with controls for strength and performance.

889 stars87 forksC++GPL-3.0

At a glance

What is it?
A fork of OptiScaler that adds NVIDIA Neural Rendering to games, with multipass and pre-upscaling placement as its own additions. The install path is one batch file, the model runtime is a second download chosen per GPU, and the recent release history is three prereleases inside two days.
Who is it for?
Treat this fork as an experimental build rather than a release channel. The download button points at the latest release page while the three most recent tags are all marked prerelease, so check what you actually get before extracting it over a game folder.
Can I use it commercially?
Yes, with conditions. GPL-3.0 is a copyleft licence: if you distribute software that includes it, you must release that software's source code under the same licence. Running it internally without distributing it does not trigger that obligation.
Is it still maintained?
Yes. The repository last received commits 20 days ago.
What is it written in?
Mainly C++, according to GitHub's language statistics.

Answers come from the project's GitHub data, last synced on November 15, 2026, and from our analysis. They are not legal advice.

From daily snapshots

Star history

  • wilsjo2/OptiScaler-DLSSNR-PreSR-Multipass889 ★ · +230
Star history
Repository2026-09-212026-11-15
wilsjo2/OptiScaler-DLSSNR-PreSR-Multipass659889

230 Stars in the last 55 days, measured from the daily GitHub snapshot. Compare with other projects

Editorial analysis

The repository name is a feature list

OptiScaler-DLSSNR-PreSR-Multipass reads as three switches, and each one maps to something the README asks you to decide. DLSSNR is Neural Rendering, the NVIDIA model path added by Dagherbou's fork. PreSR is placing the effect before the game's own upscaling rather than after it, which the README offers as a choice and marks as still experimental when it sits ahead of Ray Reconstruction. Multipass is applying the effect more than once with different settings per pass, at more performance cost.

The README calls itself an experimental community version of OptiScaler and says results and game support vary. There is no homepage field and no documentation site of its own; the setup guide is a file in the tree, `INSTALL-DLSSNR.md`, and everything else points at upstream OptiScaler or at the release list. The C++ sources sit in `OptiScaler/` next to an `OptiScaler.sln` solution file and an `OptiScaler.ini`, with `external/`, `redist/` and a `.gitmodules` entry for the third-party code that is not carried in the repository.

The whole install path is one command and a second download

Five numbered steps, and only one of them runs anything:

1. Close the game and back up existing mod files. 2. Extract all files from the release ZIP beside the game's executable. 3. Add the model file to the same folder. 4. Run the setup script and answer NVIDIA when asked. 5. Start the game, select DLSS, press Insert, enable Neural Rendering, start with one pass.

The command in step 4 is:

bash
setup_windows.bat

Step 3 is where people get stuck. The model runtime is not in the ZIP: `nvngx_dlssnr.dll` has to be fetched separately, and which file you need depends on your GPU. The setup guide has a section called choose the correct runtime for picking it and for verifying what you have. Hardware is the other gate: an RTX 20, 30, 40 or 50-series card and a supported 64-bit game, with the README warning that older cards can be much slower.

Three prereleases in two days, and the download button ignores them

The three most recent tags land inside 36 hours and every one carries the prerelease label: v0.8.8 on 2026-09-22 for optional lighting and colour reconstruction, v0.8.9 on 2026-09-23 for peripheral spatial compression, and v0.8.91 on 2026-09-23 for a compressed model input preview. The numbering jumps from v0.8.9 to v0.8.91, so string sorting and version comparison will not agree about which is newest.

Against that, the README's first link points at the repository's latest release page rather than at a versioned file. Whatever a reader downloads from that button is decided by the release list, not by the README, and the three most recent entries are explicitly labelled prerelease. For a fork whose own README calls it experimental, that is the difference between testing a build and installing one.

Reporting a problem is more specific than most: an issue should carry the game, the GPU, the settings and the contents of `OptiScaler.log`.

A Linux setup script ships next to a Windows-only guide

The tree carries both `setup_windows.bat` and `setup_linux.sh`, along with `Streamlined_fetcher_windows.bat`, `get_streamline.ps1`, `get_hybrid_assets.ps1`, `build_mfg_unlocker.ps1` and `package_release.ps1`. None of the shell scripts appear in the README, and the only install section it has is titled for Windows. Every instruction given is a Windows instruction: extract beside the executable, run the batch file, pick NVIDIA, press Insert in game.

The fetchers are the clearest clue about what is not in the repository. One script exists to retrieve Streamline components and another to retrieve hybrid assets, both pointing at NVIDIA pieces a user is expected to supply. The hybrid mode files are described as included, and the two fetch scripts sit alongside them.

A `tests/` directory is present at the top level as well. The README says nothing about running tests, and nothing about which games are on the supported list, which is the one gap a reader cannot fill from the tree.

Two files in the tree have no counterpart in the README

`Spoofing.md` sits at the top level next to `Config.md`, `Features.md`, `Issues.md` and `Changelog.md`, and the README never mentions it. The install steps do not reference it either, so a reader following the guide step by step will not find out what it covers or whether they need it.

The second is `build_mfg_unlocker.ps1`, a build script whose name is not explained anywhere in the README. It sits with the other PowerShell helpers, none of which are documented, so what gets built, what it unlocks and what it depends on all have to be read out of the scripts themselves.

Both files point the same way. A tree of this size, with a solution file, a clang-format config, an ini, submodules and a test directory, is a buildable C++ project rather than a folder of loose DLLs, and the README describes it as though it were the latter. Anyone wanting to compile rather than extract should start from the tree, not from the guide.

Hybrid mode may look like a freeze on an RTX 50

Three warnings come out of the keep-in-mind section, and the second is the most practical. Neural Rendering costs performance and can produce flicker or other visual problems. Using it before Ray Reconstruction is still experimental. And the optional hybrid mode is for RTX 50 GPUs only, its files are already included, and loading it may pause the game in a way that looks like a freeze, with the instruction to wait.

That last warning is the kind that gets reported as a crash. It is also the reason the install steps tell you to start with one pass rather than the maximum: the settings you open the overlay with should be the ones you can turn off again.

Two more boundaries are worth noting. The effect is meant to be applied to the finished picture to help with green noise, and that is stated to work with frame generation on or off in native DirectX 12 games, including HDR. And the README tells you to avoid anti-cheat-protected multiplayer games, which is the only hard prohibition in it.

Passes and model resolution are the two performance dials

Six things are adjustable, and four of them move cost rather than appearance. The effect's strength, lighting, detail and colour are the creative controls, and separate settings exist for skin and scenery, so characters and environments can be tuned independently. Everything else trades frame rate for output.

Multipass is the expensive one: applying the effect more than once with different settings each time, where extra passes cost more performance. Model resolution is the cheaper one, since lowering it reduces the performance cost. Placement relative to the game's upscaler is the third, and it is the one the repository name advertises, with the pre-upscaler route carrying the experimental label.

What the README does not give is a frame time figure for any of these. No pass count is costed, no resolution is tied to a percentage of frame time, and the warning that older cards can be much slower is the only performance statement available. Budget by measurement on your own card.

Editorial conclusion

Treat this fork as an experimental build rather than a release channel. The download button points at the latest release page while the three most recent tags are all marked prerelease, so check what you actually get before extracting it over a game folder. Three things decide whether it is worth your time: the per-GPU model runtime, which is a separate download you have to pick by card; the number of passes you intend to run, since each extra pass costs more and the recommendation is to start at one; and whether your GPU is an RTX 50 if you want hybrid mode, which is the only optional mode and is allowed to look like a freeze while it loads. Read the GPL-3.0 grant and the bundled `Licenses/` directory before redistributing anything built from it, since the README says nothing about redistribution or about the terms of the games themselves, and it does rule out anti-cheat protected multiplayer titles. The last push was 2026-09-29, after the v0.8.91 prerelease on 2026-09-23.

Frequently asked questions

What GPU does OptiScaler Neural Rendering need?

An NVIDIA RTX 20, 30, 40 or 50-series card and a supported 64-bit game. The README notes that older cards can be much slower, and the hybrid mode is limited to RTX 50 GPUs.

Do I have to download an NVIDIA model file separately for OptiScaler?

Yes. The release ZIP does not contain the runtime, so `nvngx_dlssnr.dll` is fetched on its own and the correct file depends on your GPU. The setup guide has a section called choose the correct runtime that covers the choice and the verification.

Can I use OptiScaler Neural Rendering with frame generation and HDR?

Yes, in native DirectX 12 games. Applying the effect to the finished picture to help with green noise works with frame generation on or off, including HDR.

Is OptiScaler Neural Rendering safe to use in multiplayer games?

The README tells you to avoid anti-cheat-protected multiplayer games. Everything else it warns about is visual: flicker, other visual problems, and a pause while hybrid mode loads on an RTX 50.

How do I open the OptiScaler overlay inside a game?

Select DLSS in the game first, then press Insert to open OptiScaler. Enable Neural Rendering and start with one pass, since each additional pass costs more performance.

Official sources

  1. Issues
  2. License: GPL-3.0
  3. README
  4. Releases
  5. wilsjo2/OptiScaler-DLSSNR-PreSR-Multipass on GitHub
Add this badge to your README

If you maintain this project, the badge below links readers to this analysis and shows its maintenance status from the daily GitHub snapshot. Paste the markdown into your README; add ?metric=license or ?metric=stars to the image URL for a different field.

Add this badge to your README

markdown
[![Hysen Labs](https://hysenlabs.com/badge/wilsjo2-optiscaler-dlssnr-presr-multipass.svg)](https://hysenlabs.com/projects/wilsjo2-optiscaler-dlssnr-presr-multipass)

Star history chart for your README

markdown
[![Star history](https://hysenlabs.com/chart/wilsjo2-optiscaler-dlssnr-presr-multipass.svg)](https://hysenlabs.com/projects/wilsjo2-optiscaler-dlssnr-presr-multipass)

Was this useful?