Frame generation means your GPU creates an extra frame between every two real game frames, making motion look noticeably smoother. Until now, that was mostly locked behind NVIDIA's DLSS Frame Generation. Now ANR (Apex Neural Rendering) from Apex Compute has arrived, and it pulls off the exact same trick with a neural network that weighs just 0.17 MB—working on older GPUs and non-NVIDIA cards too.
What do the numbers look like?
On an RTX 5090 at 4K, according to Apex's own benchmarks:
Speed: Generating each frame takes just 1.01 ms, compared to 1.26 ms for DLSS-G.
VRAM usage: It uses only 214 MB of GPU memory, versus 562 MB for DLSS-G.
Image quality: In 2X mode, it hits a PSNR of 33.84 versus 31.22, and scores 96.59 against 93.66 on the CGVQM-5 metric. That gap widens even further in the 3X through 6X modes.
Apex also shared side-by-side comparison clips from The Last of Us Part II, and TechPowerUp reports seeing ANR running smoothly in Hogwarts Legacy.
How do you install it?
ANR FG works as an injector built on top of OptiScaler, meaning it runs in pretty much any game that already supports DLSS Frame Generation:
Grab
anr-fg-fullfrom the GitHub Releases page (oranr-fg-addonif the game already has OptiScaler installed).Close the game and run the installer script (
install.cmdon Windows,install.shon Linux and Steam Deck), pointing it to the game's executable folder.Launch the game and simply turn on NVIDIA Frame Generation in the settings.
Supported hardware covers GTX 16, RTX 20 through 50, Radeon RX 5000 through 9000, Steam Deck, and Intel Arc. The project page even lists native timings on Apple Silicon—for instance, 4.26 ms per frame at 1080p on an M2 Max.
The repo also includes a handy tool called nr-record, which captures a game's raw frame generation inputs so you can replay the exact same sequence through ANR, DLSS-G, and others for side-by-side video and pixel-peeping comparisons.
What's coming next?
Frame generation (FG) is available right now in beta. Super resolution (SR) and ray reconstruction (RR) are actively in the works, and the team plans to open-source the code as well.
My take
What really stands out to me is that frame generation is no longer tied to a single GPU brand or the latest hardware generation. A model small enough to fit inside a couple hundred kilobytes can bring smooth frame generation to handhelds and older laptops too. It's going to be really interesting to see how the full suite—once SR and RR arrive—stacks up against DLSS 5.
