A positive vignette fades the edges of a photograph toward white. In one of our Lightroom comparisons, Chiado turned a corner into a flat patch of RGB (255, 255, 255). Lightroom kept small variations around 249. The difference was visible, and it suggested a straightforward correction: approach white smoothly instead of stopping abruptly at it.
We tested that correction. It removed much of the flat white region. We still decided not to adopt it.
The corner was a clue, not a specification
Our first check was whether Lightroom simply used a dimmer white. Fifty-four existing references covered nine development photographs, four positive strengths and two additional cropped cases per photograph. They ruled out a fixed cap. At stronger settings, Lightroom approached 255; the corner values also depended on the starting image.
A second idea changed the entire blend to approach white gradually. We fitted its parameters using only the original three training photographs. On a separate, already-consumed six-photo development set, it worsened eight of twelve preview comparisons. Reducing clipped values did not rescue the broader mismatch.
Change only the approach to white
The narrower candidate preserved the existing blend through most of its range. Above a fitted transition point, it approached white exponentially. Its value and slope joined continuously at that point. This preserved the central image and avoided an abrupt mathematical corner in the blend function.
We built the candidate separately from the app. Across 72 preview renders and 36 native cropped renders, all 72 zero or negative vignette control cases remained pixel-identical. Tests for center preservation, repeatability and regional consistency passed. Those checks established that the change was contained; they did not establish that it matched Lightroom.
The averages hid an important split
For positive vignette over neutral, detail and grain backgrounds, the preview comparisons split evenly: twelve improved and twelve regressed. All twelve native vignette comparisons improved. Native grain measurements were mixed, with a slightly worse mean error.
Another 144 paired renders checked four strengths and cropped behavior. Across the original six validation photographs, now treated as consumed development data, average response error fell by only about 0.5%. At +50, five of six full-frame cases and four of six cropped cases worsened.
Visual review explained the tradeoff. The candidate preserved corner content that the current renderer flattened. But in several scenes, Lightroom applied a stronger white fade than the candidate. Preserving more detail can be desirable in isolation while moving away from the behavior we are trying to reproduce.
Keep both findings
We rejected both candidates and left the app unchanged. That does not make the original white plateau correct. It remains a documented limitation, alongside the weaker negative vignette and unresolved grain differences.
The experiment answered a narrower question: these two corrections did not offer a convincing overall replacement. Further work needs a different explanation supported by isolated reference measurements, rather than more coefficients fitted to familiar photographs.
Related: a grain improvement that failed on native crops, and why repeated development checks are not independent validation.