The Illusion of Detail
The prevailing advice for tuning FSR or DLSS sharpness is straightforward: increase the value until the image looks crisp. This guidance assumes that the primary failure mode of upscaling is softness, and that the remedy is to re-introduce edge definition. In a static frame, this logic holds up. A sharpening kernel, typically a simple unsharp mask or a more complex edge-aware filter, increases the local contrast at boundaries between pixels of differing luminance. By boosting the high-frequency components of the signal, the eye perceives a higher level of detail. However, this is not the addition of information; it is the amplification of existing frequency data. When the source resolution is lower than the target display resolution, the upscaler has already made a guess about what the missing pixels should be. Sharpening does not correct a bad guess; it simply makes the edges of that guess more aggressive. The result is an image that feels sharper to the human visual system because the transition from light to dark is steeper, but the underlying geometric accuracy remains unchanged. The danger lies in equating this perceptual sharpness with actual fidelity.
This mechanism creates a fundamental trade-off that is rarely discussed in casual tuning guides. The same mathematical operation that steepens an edge also amplifies any error in the position of that edge. In a perfect reconstruction, this would be harmless. In an upscaled image, where edges are approximated, the amplification causes a specific type of artefact known as ringing. This manifests as faint, ghostly halos around high-contrast boundaries, such as a bright window against a dark sky or a thin wire against a bright background. In a still image, these halos are often subtle enough to be ignored, or even mistaken for a stylistic choice. The visual system interprets the increased local contrast as clarity. However, this interpretation is context-dependent. The "crispness" you are dialing in is actually the intensity of the error signal. You are not revealing detail; you are highlighting the boundaries of the upscaler's uncertainty. The condition under which this stops being true is when the source image contains genuine high-frequency detail that the upscaler failed to preserve. In that rare case, sharpening can help. But in the majority of real-time rendering scenarios, the source is already aliased or blurred, and the sharpening is purely cosmetic.
Ringing as a Visual Error
The artefact introduced by aggressive sharpening is not random noise; it is structured error. Ringing occurs because sharpening filters are essentially high-pass filters. They subtract a blurred version of the image from the original to isolate edges, then add that result back in. When the image contains sharp transitions, the filter overshoots, creating pixels that are brighter than the highlight or darker than the shadow on either side of the edge. This overshoot is the halo. The severity of this halo is directly proportional to the sharpness setting. A low setting might produce a halo that is indistinguishable from normal anti-aliasing. A high setting produces a halo that is clearly visible, appearing as a white fringe on dark objects or a dark fringe on bright ones. This is not a bug in the implementation; it is an inherent property of linear convolution filters operating on discrete pixel grids. The only way to avoid it entirely is to use a non-linear, edge-aware filter that stops processing at the boundary, but even these have limits. Most real-time upscalers use a balance of techniques, but the sharpening stage is often a simple, fast kernel that does not have the computational budget to perform complex edge detection for every pixel.
The consequence of this is that the image becomes unstable in its representation of edges. In a static scene, the halo is fixed in space. Your eye adapts to it. You stop seeing the halo and start seeing the "sharp" edge. This is a form of visual adaptation that works against you. It trains you to prefer the artefact. When you look at a screenshot, you are seeing the peak of the halo, which creates a strong contrast boundary. But in motion, the halo moves with the object. This is where the problem becomes acute. The eye is much more sensitive to moving artefacts than static ones. A static halo is background; a moving halo is foreground. It catches the peripheral vision and disrupts the perception of smooth motion. This is why a setting that looks perfect in a paused frame can look terrible during gameplay. The "crispness" you tuned for is now a shimmering, vibrating outline around every moving object. The mechanism is the same, but the perceptual weight changes entirely. The condition under which this is acceptable is when the motion is slow and the edges are not high-contrast. In fast-paced action with high-contrast lighting, the ringing becomes a significant visual pollutant.
The Motion Penalty
The critical omission in most tuning advice is the distinction between static and dynamic image quality. Screenshots are frozen in time. They allow the eye to settle, to focus, and to ignore the subtle inconsistencies in the image. When you pause a game to take a screenshot, you are effectively removing the temporal dimension from the equation. The upscaler's temporal filtering, which uses data from previous frames to stabilize the image, is no longer relevant to the final output. The sharpening is applied to a single, isolated frame. In this context, the ringing artefact is a minor blemish. It might look like a slight over-exposure or a digital artifact, but it does not detract from the overall impression of sharpness. Many users tune their settings in this mode, comparing paused frames side-by-side. They choose the setting that produces the stepest edges, unaware that they are optimizing for a scenario that does not exist during actual play. The "crisp" look is a static illusion. It relies on the absence of motion to hide the fact that the edges are not real.
In motion, the temporal filtering of the upscaler becomes the primary stabilizer. The upscaler blends the current frame with previous frames to reduce flicker and jitter. This blending smooths out the high-frequency noise, including the ringing artefacts, to some degree. However, if the sharpening is applied after this blending, or if the sharpening is so aggressive that it overwhelms the temporal smoothing, the artefacts persist. More importantly, the motion of the object causes the ringing to move across the screen. The eye tracks the object, but the ringing trails behind or leads in front, creating a shimmering effect. This is often described as "shimmer" or "boiling." It is a direct result of the high-pass filter interacting with the temporal changes in the image. The condition under which this stops being a problem is when the upscaler's temporal stability is high enough to completely suppress the ringing. This is rare, especially at high sharpness settings. Most upscalers struggle to maintain stability in fast-moving scenes, and the sharpening exacerbates the instability. The result is an image that is technically sharper in a single frame but visually noisier over time.
Tuning for the Wrong Metric
The standard method for tuning sharpness involves adjusting the slider and observing the change in a static scene. This is a flawed methodology because it ignores the temporal component of the visual experience. A better approach is to tune in motion. Find a scene with fast-moving objects and high-contrast edges, such as a character running past a bright window or a vehicle speeding along a road. Observe the edges as they move. If you see shimmering, halos, or boiling, the sharpness is too high. The goal is not to maximize edge steepness; it is to minimize temporal instability. This requires a different visual skill. You are not looking for detail; you are looking for consistency. A slightly softer edge that remains stable in motion is preferable to a sharp edge that vibrates. This is a counter-intuitive lesson for many users who are accustomed to traditional image editing, where sharpening is always seen as an improvement. In real-time rendering, sharpening is a risk. It is a tool that can improve perceived quality, but it can also degrade it, depending on the context. The condition under which static tuning is acceptable is when the game is slow-paced and the player spends a lot of time pausing. For fast-paced games, this is rarely the case.
The implication of this is that the "best" sharpness setting is not a single number; it is a range that depends on the game, the scene, and the player's sensitivity to artefacts. There is no universal setting that works for all games. A setting that is perfect for a slow-paced RPG may be unbearable for a fast-paced shooter. Players need to understand that they are not just adjusting a filter; they are adjusting the balance between spatial detail and temporal stability. This requires a willingness to accept a slightly softer image in exchange for a smoother, more stable one. It is a trade-off that is rarely acknowledged in marketing materials, which focus on the "crisp" look of static images. The reality is that the human visual system is more sensitive to motion artefacts than to static softness. We can live with a slightly blurry image; we cannot live with a shimmering one. The condition under which this trade-off fails is when the upscaler is so poor that even a low sharpness setting produces significant artefacts. In that case, the only solution is to lower the resolution or disable upscaling entirely.
The Cost of Crispness
The final point to consider is the computational cost of sharpening. While sharpening is a relatively cheap operation, it is not free. It requires an additional pass over the image, which consumes GPU resources. In most cases, this cost is negligible compared to the cost of the upscaling itself. However, in some implementations, the sharpening is applied to the full resolution image, which can add a small but measurable overhead. More importantly, the sharpening can interact with other effects in the pipeline. For example, if the game uses a bloom effect, the sharpening can amplify the bloom, causing bright areas to become even brighter and more distracting. This can lead to a washed-out look, especially in scenes with high dynamic range. The condition under which this is a problem is when the game has a lot of bright, high-contrast elements. In such scenes, the sharpening can make the image look harsh and fatiguing. Players should be aware of this interaction and adjust their settings accordingly. If the image looks too bright or too contrasty, lowering the sharpness may help, even if the edges look softer.
In summary, the claim that "more sharpening makes upscaled images look better" is incomplete. It is true only in the context of static images and only up to a certain point. Beyond that point, the artefacts introduced by sharpening outweigh the benefits of increased edge definition. The key to tuning upscalers is to understand the trade-off between spatial detail and temporal stability. It is not about making the image look sharper; it is about making the image look more stable. This requires a different approach to tuning, one that prioritizes motion over stillness. By understanding the mechanism of sharpening and the nature of the artefacts it introduces, players can make more informed decisions about their settings. They can choose a setting that provides a good balance between detail and stability, rather than blindly maximizing sharpness. The result is an image that is not just crisp in a screenshot, but comfortable and clear in motion. This is the true measure of quality in real-time rendering.