Earlier this month, AMD rolled out its FSR Redstone suite of ML-enhanced features, including FSR Ray Regeneration, Neural Radiance Caching, and crucially, an upgrade to its Frame Generation solution courtesy of a new machine learning algorithm that aims to boost the quality of generated frames. However, just as with FSR 4 upscaling, FSR Redstone FG is only available for titles with FSR 3.14+ already integrated. This means that currently, there are only around 30 titles that support the new version of AMD's FG tech.
This number pales in comparison with the 175+ games with DLSS4 MFG support (as of August), and although AMD has shown in the past with FSR4 that it can rapidly scale game integration, RDNA 4 users who wish to enable ML FG are currently quite limited by the lack of support. Once again, it's up to the community to step up and sort this out - and step up they did, big time.
Just a day after AMD officially launched the full suite of Redstone technologies to the public, the Optiscaler devs announced they had added basic redstone support, thanks to an internal upgrade to the FFX 2.1 SDK. This means that any game with at least basic upscaling support (anything from FSR 2 / DLSS 2 onwards) can technically use the latest version of AMD's FSR FG, although you'll get the best results from sticking to games with DLSS FG inputs.
How to Install
The first thing you need to do is join the Optiscaler Discord, where you can find pre-release build 1215. The stable release on their GitHub page dates all the way back to August, and so obviously doesn't support Redstone features. Once you've done that, go ahead and install Optiscaler; you can find detailed documentation of the process here. Make sure to press 1 to spoof Nvidia inputs when prompted. If you can't be bothered to open that link, here's a quick recap of how to install Optiscaler:
- Extract the Optiscaler folder.
- Copy the contents into the game's directory (where the .exe file is)
- Run setup_windows.bat
- Repeatedly input 1 and press enter until the setup process completes.
Now, you're ready to launch your game. This is where your path diverges depending on what kind of Frame Generation technology the game supports natively:
Option 1: Game Supports DLSS FG, but only through Streamline SDK 1
If your game has native support for DLSS FG, but uses the older Streamline 1 SDK, you won't be able to use DLSS inputs for ML FG. Instead, you'll have to use OptiFG, which uses the same inputs as your upscaler. An example of a game like this is A Plague Tale: Requiem. The first thing to do is launch the game and enable DLSS Upscaling. Then, once you've entered an actual gameplay section, launch the Optiscaler menu by pressing Insert. You can then change your upscaler from the default, XeSS, to FSR4.
Once you've changed the upscaler, go ahead and select OptiFG input in the frame generation settings, with FSR FG as the output. You'll need to hit Save INI (at the bottom) and restart the game to see changes apply.
Then, change your FG model from FSR 3.1.6 to 4.0.0.
And you're done! In terms of visual quality, this method of implementing ML FG seems to deliver decent results, although truth be told, I was hard-pressed to find the difference between the FSR 4.0.0 (ML FG) and FSR 3.1.6 models. This is probably because of the relatively poor input data - OptiFG doesn't need any in-built frame generation support to run, and instead only relies on whatever data is available to the upscaler, which makes me think it doesn't use any of the specialized input data proprietary frame generation solutions have.
Short of slow-motion video capture at thousands of frames per second or a built-in refresh rate counter in my monitor, I don't have any concrete way to test frame pacing, so it's probably best if you experiment with each model to find the best option. However, to the naked eye, it doesn't seem like FSR 4.0.0 FG has any additional frame pacing issues compared to the FSR 3.1.6 implementation - although neither is completely smooth, I'd say they both fall well into the realm of the usable. I tried using CapframeX's MsBetweenDisplayCapture metric to see if some frame delay issues could be spotted with FSR 4.0.0 FG, but it turns out this isn't actually a reliable way to test FG frame pacing, as pointed out by the creator of the software himself.
Option 2: Game supports DLSS FG via Streamline SDK 2
If your game has DLSS FG implemented via Streamline SDK 2+ (but no FSR FG support), your best option is to use the game's DLSS inputs to produce an FSR 4.0.0 FG output. We'll use Star Wars: Jedi Survivor as an example for this.
The first step, as always, is to extract the contents of Optiscaler to the game's directory, where the .exe file is. For these kinds of titles, it's better to use Optipatcher to enable the use of DLSS inputs without any kind of spoofing. This removes performance penalties associated with spoofing, but more importantly, makes the whole process a lot smoother. You can grab Optipatcher here; make sure to grab the latest release from the right-hand side, and download the .asi file, not the source code. Next, make a folder called "plugins" in the game directory (same place you extracted Optiscaler.zip to) and put the Optipatcher.asi file into it. Head into the Optiscaler config file (Optiscaler.ini) and change the value of LoadAsiPlugins to true (auto defaults to false).
Head into the game. Once you get into a gameplay section, press Insert to bring up the Optiscaler menu. Change the upscaler from XeSS to FSR3.X/4. Note that you might have to use arrows to navigate the menu here, as the beginning section of Star Wars: Jedi Survivor has MB1 bound to camera zoom.
Then, change the FG source to DLSSG via Streamline, and the FG Output to FSR FG. Save INI and close.
Restart the game, and head into the game's graphics menu. Here, you should see DLSS Frame Generation is now available. Enable it, and then from the Optiscaler menu, change FG Model from FSR 3.16 to FSR 4.0.0.
I have to say, at least in Jedi Survivor, I was mightily impressed with the final result. Frame pacing felt smooth, HUD elements were completely excluded by the FG algorithm (probably thanks to using DLSS inputs), and visual quality was rock solid. Again, I wasn't quite able to notice a difference between FSR 3.1.6 FG and FSR 4.0.0 FG, but I'm sure users with keener eyes than mine will definitely notice an improvement in visual quality.
Follow Wccftech on Google to get more of our news coverage in your feeds.
