Difference Evaluators
Pixel Eagle supports multiple difference evaluators. You can configure the evaluator in your project settings.
Pixel XOR
Compares screenshots pixel by pixel using XOR. Differences are highlighted in red on a transparent background.
Fast and precise - best for catching any pixel-level change.
ꟻLIP Paid plans
A perceptual comparison algorithm based on NVIDIA's ꟻLIP metric . It models the human visual system to detect differences that would actually be noticeable to a viewer.
Best for ignoring sub-pixel rendering differences while catching meaningful visual changes. Can be enabled for all screenshots or only specific ones.
Settings
Fine-tune how comparisons behave through your project settings.
Tolerance Paid plans
Controls how much two pixels must differ before they are marked as different. A tolerance of 0 (the default) means any change counts. Higher values ignore small color variations.
With Pixel XOR, tolerance is compared against the normalized RGB difference of each pixel. A pixel is only flagged when its difference exceeds the tolerance value.
With ꟻLIP, tolerance adjusts the perceptual sensitivity of the algorithm and the percentile used to compute the final score, making it less sensitive to subtle differences.
Threshold
A final gate applied after comparison. If the overall difference score falls below the threshold, the comparison is treated as identical (score set to 0).
Use this to suppress noise from runs that have only a handful of barely-different pixels. The default threshold is 0.02%.
Diff dilation Paid plans
Expands each differing pixel in the diff image by a given radius, making isolated pixel changes easier to spot when visually reviewing a comparison.
Dilation only affects the generated diff visualization - the difference score itself is computed before dilation is applied. The default radius is 0 (disabled).
Tolerance vs Threshold: Tolerance decides which pixels count as different during comparison. Threshold decides whether the final result is different enough to matter. Use tolerance to ignore per-pixel noise (e.g. anti-aliasing), and threshold to ignore small overall scores.
Evaluate view: Each comparison has an Evaluate view that lets you sweep one setting (tolerance, threshold, or diff dilation) across a range of values and see how it affects the diff image and the final score in real time. Use it to dial in the right value for a given project before saving it to your settings.
Visualizations
Every comparison produces two complementary images to help you understand what changed.
Red diff mask
A binary overlay where every pixel that exceeds the tolerance is painted solid red on a transparent background, and every other pixel is left transparent.
This mask reflects exactly which pixels contributed to the score, and is what the diff dilation setting expands. Use it to locate the regions that need your attention.
Error-map visualization
A heatmap rendered with the magma color map, going from dark purple (no difference) through red and orange to bright yellow (large difference). It shows the full per-pixel error, not just a pass/fail flag.
With Pixel XOR, the map encodes the normalized RGB distance for every pixel. With ꟻLIP, it encodes the perceptual error produced by NVIDIA's algorithm.
Red mask vs error map: The red mask is tolerance-aware and binary — a pixel is either flagged or not. The error map is tolerance-agnostic and continuous — it shows how different each pixel actually is. Use the red mask to see what the current settings catch, and the error map to spot subtle differences that fall under the tolerance and to decide whether to loosen or tighten it.