Frame Generation Explained: How DLSS Frame Gen and AMD FMF Actually Work
When a GPU maker claims a new graphics card "doubles" frame rates in a supported game, there's a good chance at least part of that gain isn't coming from the GPU rendering more real frames at all. It's coming from frame generation: a technique that inserts entirely new, AI-synthesized frames between the frames the GPU actually rendered, boosting the reported frame rate without a proportional increase in rendering work. It's one of the more genuinely clever — and more misunderstood — technologies to show up in modern graphics cards, and it works differently enough from traditional rendering that it's worth understanding before trusting a frame-rate number that includes it.
How traditional rendering works, for contrast
In conventional rendering, every single frame shown on screen is calculated from scratch by the GPU: geometry is processed, lighting is calculated, textures are applied, and the finished image is sent to the display. Doubling the frame rate the traditional way means the GPU has to do twice as much actual rendering work in the same amount of time, which is why frame rate has historically scaled closely with raw GPU horsepower. This is also why techniques like the ones covered in our DLSS vs FSR vs XeSS explainer became so important — rendering at a lower internal resolution and upscaling with AI cuts down the rendering workload while trying to preserve visual quality.
What frame generation actually does
Frame generation takes a different approach entirely. Instead of rendering every frame from the game engine, it renders real frames at a reduced rate and then uses a trained AI model to analyze two consecutive real frames — along with motion vector data the game engine provides — to synthesize a brand new frame that sits in between them, estimating what the in-between motion should look like. NVIDIA's DLSS Frame Generation, AMD's Fluid Motion Frames (FMF), and Intel's frame generation technology all work on this same basic principle, though each uses its own model and implementation details. The result on screen can be a genuinely higher displayed frame rate, with the caveat that a meaningful share of those frames were never calculated by the game engine at all.
Why this matters for input latency
This is the single most important thing frame generation does not fix: input latency is tied to how often the game engine actually processes a new real frame and registers player input, not to how many frames are displayed. A generated frame is inserted for visual smoothness, but it doesn't contain any new information from the game about where the player just moved the mouse or pressed a button. This means a game running with frame generation can display a much higher frame rate number while still feeling exactly as responsive — or occasionally slightly less responsive, due to the buffering needed to generate the in-between frame — as it did before generation was turned on. This is why frame generation is generally recommended for visual smoothness in single-player or less input-critical games, and treated far more cautiously for competitive, latency-sensitive gaming.
Where it works well
Frame generation tends to shine in exactly the situations where visual smoothness matters more than split-second responsiveness: cinematic single-player games, especially ones already running at a healthy base frame rate that frame generation can smooth out further, and situations where a high-refresh-rate gaming monitor is being fed a base frame rate too low to look fluid on its own. Generating extra frames to better fill a 144Hz or higher-refresh display can noticeably reduce perceived motion blur and juddering, even without any change to actual game responsiveness.
Where it struggles
Frame generation depends on having a clean, reasonably high base frame rate to work from — the AI model is interpolating between two real frames, and the lower and choppier that base frame rate is, the harder it becomes to generate a convincing in-between frame without visible artifacts around fast-moving objects or fine detail like text and UI elements. This is part of why it's generally recommended as a way to push an already-decent frame rate higher, not as a fix for a genuinely struggling one. It also interacts closely with ray tracing, since ray-traced games are often exactly the ones with base frame rates low enough that frame generation's limitations become more visible.
How to read frame-rate claims that include it
Because of all this, it's worth treating any frame-rate figure advertised with frame generation enabled as a different, separate metric from a traditionally rendered frame rate rather than a directly comparable number. Reviewers increasingly report both the "real" rendered frame rate and the frame-generation-boosted number specifically because the two measure meaningfully different things: one reflects genuine GPU rendering throughput and responsiveness, the other reflects display smoothness with an AI-assisted boost layered on top. A GPU that shows a big generational leap almost entirely because of new frame generation capability, rather than more raw rendering power, is a very different upgrade than one delivering the same frame-rate gain through actual rendering improvements.
Multi frame generation and diminishing returns
Newer implementations have pushed the idea further, generating multiple synthesized frames between each pair of real frames instead of just one. This can push the displayed frame rate up even more dramatically, but it doesn't change the underlying tradeoff — every one of those additional generated frames still carries zero new input information, and the base real frame rate the technique is working from matters just as much as before, arguably more, since the AI model has to bridge a wider gap convincingly with each extra inserted frame. Latency-sensitive players generally get the least benefit from pushing multi-frame generation further, while people who mainly care about a display looking smooth get the most.
The bottom line
Frame generation is a genuinely useful tool for smoothing out already-decent performance on a high-refresh display, not a substitute for real rendering horsepower, and definitely not something to lean on for latency-sensitive competitive play. Understanding that a generated frame carries no new player input, only interpolated motion, is the key to knowing when to turn it on — and when a frame-rate number that includes it is telling a slightly different story than it appears to be.
