The Illusion of Detail
The prevailing wisdom among players tuning upscalers is straightforward: if the image looks soft, add sharpening. This advice is not malicious, but it is fundamentally incomplete because it treats visual quality as a linear scale where "more detail" is always better. In reality, sharpening is a signal processing operation that amplifies high-frequency components in an image. When you apply a sharpening kernel to an upscaled frame, you are not revealing hidden data that was lost during the downscaling process. You are mathematically exaggerating the contrast between adjacent pixels. This creates a perceptual illusion of crispness that the human visual system interprets as detail. The problem is that this operation is blind to context. It does not know the difference between a genuine edge on a rock and a noise spike from temporal filtering. It amplifies both with equal indifference.
This mechanism explains why a heavily sharpened image can look superior in a still frame while appearing degraded in motion. In a static scene, the amplified edges remain fixed relative to the screen. The eye registers the increased local contrast as clarity. However, in a dynamic scene, the camera or objects move. The sharpening kernel is applied to a new set of pixel values every frame. If the underlying low-resolution data has any temporal instability, which is inherent to most upscaling techniques that rely on feedback buffers, the sharpening process magnifies those fluctuations. The result is not just a sharper image, but a noisier one. The "detail" you are adding is often actually amplified error.
Ringing and Edge Artifacts
The specific artefact that accompanies aggressive sharpening is known as ringing. When a high-contrast edge, such as a bright window against a dark wall, is processed by a sharpening filter, the mathematical operation causes the pixel values on either side of the edge to overshoot their natural range. On one side, the pixels become brighter than they should be; on the other, darker. This creates a faint, ghost-like halo around the edge. In a single frame, this halo is often subtle enough to be ignored, or even mistaken for a lens effect. However, the halo is not uniform. Its intensity depends on the exact alignment of the edge with the pixel grid. As the image moves, the edge shifts by fractions of a pixel. The ringing pattern shifts with it, pulsing in and out of visibility.
This pulsing is what players describe as shimmer. It is a temporal artefact caused by the spatial operation of sharpening. The condition under which this stops being a problem is when the image is perfectly static. In a paused menu or a still screenshot, the ringing is consistent and unobtrusive. The moment the camera pans, the ringing moves, and the human eye, which is highly sensitive to motion, picks it up. Most online guides recommend specific sharpening values based on screenshots. This is a methodological error. A screenshot captures the spatial state of the image at one instant. It does not capture the temporal stability of that state. By tuning for the screenshot, you are optimising for a condition that rarely occurs in actual gameplay. You are trading temporal stability for spatial contrast, a trade that becomes increasingly expensive as the sharpening amount rises.
The Motion Penalty
The cost of this trade is most visible in high-contrast, high-motion scenarios. Consider a scene with a bright sky, dark trees, and a camera moving quickly. The upscaler is already struggling to reconstruct the fine details of the leaves against the sky. The temporal filter smooths out the noise by blending the current frame with previous ones. This smoothing introduces a slight lag, or ghosting, behind moving objects. When you apply sharpening, you are sharpening this ghost. The halo around the moving object becomes a trailing smear of exaggerated contrast. This is not just ugly; it is disorienting. It breaks the visual continuity of the scene. The brain expects edges to be stable. When they oscillate in brightness and position, it creates a sense of visual fatigue that accumulates over time.
The mechanism here is the interaction between temporal filtering and spatial sharpening. Temporal filters reduce noise by averaging over time, which blurs moving edges. Spatial sharpeners increase contrast, which sharpens those blurred edges. But they do so by amplifying the very noise that the temporal filter was trying to suppress. The result is a feedback loop where the upscaler tries to clean the image, and the sharpener tries to dirty it again. The condition under which this stops being true is when the upscaler is so good that it does not need temporal smoothing. DLSS 3 and 4 use transformer models that are trained to handle motion data explicitly. They are less reliant on simple temporal averaging. Therefore, the penalty for sharpening is lower in these systems. However, it is not zero. Even the best neural upscalers produce some level of temporal variance. Sharpening will always amplify that variance. The only question is whether the variance is below the threshold of perception.
Tuning for the Moving Image
If you want to tune your upscaler correctly, you must abandon the screenshot as your primary reference. Instead, you need to observe the image in motion. Set your sharpening to zero. Play a scene with high contrast and significant camera movement. Gradually increase the sharpening. Watch the edges. You are not looking for the point where the image looks sharpest. You are looking for the point where the edges start to vibrate. This vibration is the ringing artefact manifesting in time. The optimal setting is just below this threshold. It will likely be lower than the setting that looks best in a paused frame. This is the trade-off. You are accepting a slightly softer image in exchange for a stable one.
This approach is counterintuitive because our visual preference for detail is usually assessed in stillness. We pause games to admire the art. We take screenshots to share. But we play games in motion. The visual system is tuned to detect change. A stable image, even if slightly soft, is less fatiguing than a crisp one that shimmers. The condition under which this advice fails is if you are playing a game with very low motion, such as a turn-based strategy or a visual novel. In these cases, the temporal penalty is negligible, and you can push the sharpening higher. But for real-time action, shooters, or open-world exploration, the motion penalty is the dominant factor. You are not tuning for a photograph. You are tuning for a video.
The Limits of the Slider
The sharpening slider is a blunt instrument. It does not allow you to control the amount of ringing separately from the amount of contrast. This is a design limitation of most upscaling implementations. They use a simple unsharp mask or a similar kernel. A more sophisticated approach would be to apply sharpening only to edges that are stable over time. This would require the upscaler to have a motion vector map and to apply the sharpening selectively. Some advanced upscalers, like DLSS, do this implicitly through their neural networks. The network learns to sharpen static edges and ignore moving ones. However, the user-facing sharpening slider in most games is still a global control. It applies the filter to the entire frame. This means that even in a game with a good upscaler, the user is forced to make a global compromise.
The implication for players is that the "best" setting is not a number. It is a state of equilibrium. It is the point where the benefit of increased clarity is balanced by the cost of increased shimmer. This equilibrium point varies by game, by scene, and by monitor. A high-refresh-rate monitor will make the shimmer more noticeable because the temporal resolution is higher. A lower-resolution upscaling mode will require more sharpening to look acceptable, which means the shimmer will be more pronounced. There is no universal setting. The advice to "just set it to 50%" is lazy and incorrect. You must observe the image in motion. You must accept that a slightly softer image is often the better choice. The goal is not to maximise detail. The goal is to minimise visual noise. Sharpening is a tool for adding contrast, not for removing noise. Use it accordingly.