At 150 percent, every source row is supposed to become one and a half rows tall. The output grid can store only whole rows. With nearest-neighbor sampling, some source rows therefore get copied into one output row and others into two. Hard edges and the original colors can survive while equally thick source stripes come out with unequal widths or heights.
This difference is visible without motion or blur. Compare a small image with evenly spaced stripes at its original size, 150 percent and 200 percent. Inspect the stored output pixels, because a second resize in the viewer can make a correct file look uneven or soften the boundaries.
Six rows must fill nine target rows
Take six source rows labeled A through F. Scaling the image height by 150 percent produces nine rows. A nearest-neighbor resize gives each destination row the color of one source row. At least some colors have to repeat to fill those nine positions. In one common center-based mapping, the destination rows are A, B, B, C, D, D, E, F and F. A, C and E take one row each. B, D and F take two.
This example uses the centers of the destination cells and maps them back to the source. A center that falls in a particular source row uses that row's value. At exact boundaries this example selects the later source row. A different program can choose the earlier row, place the grid a little differently or round the size differently. That changes which rows repeat. It does not change the reason an even set of six source rows cannot all become exactly one and a half discrete rows.
Now scale the same six rows to 200 percent. Each row can occupy exactly two destination rows if the grid lines up. The source and destination need no uneven distribution for that integer enlargement. The difference is arithmetic before it is visual. At 150 percent, the extra three rows cannot be spread equally across six source rows using only whole cells.
There is an easy way to check the example by hand. Number the destination rows from zero to eight and put each sample at its center. Divide each center coordinate by 1.5, then choose the source row containing it. The middle of destination row one maps exactly to the boundary after source row A. Choosing the later row explains why the second output entry is B in this example.
The same allocation affects columns and connected shapes
The mapping runs along both axes. If the width is also scaled to 150 percent, some source columns become one column and some become two. A source pixel can end up as a one-by-one, two-by-one, one-by-two or two-by-two block depending on where it lands. That is why a repeated pattern may look heavier in some parts even though the art was not stretched in one direction more than the other.
A good example is a one-pixel checkerboard or a sprite with a narrow border. A repeated row turns one stripe into a thicker stripe. In a checkerboard, the extra width and height can make some squares become rectangles. A diagonal edge can gain uneven steps because row and column repeats interact. The outer edge can still be sharp at every step.
OpenCV's geometric transform documentation describes inverse mapping and nearest-neighbor interpolation. The exact allocation in this article is an example of that family of methods. It does not predict the tie-breaking choices of every editor, operating system or graphics device. For your image, count the destination pixels using the actual export rather than assuming it follows a particular phase.
Hard edges and equal thickness are separate outcomes
MDN documents the Canvas setting imageSmoothingEnabled. Disabling smoothing keeps pixel art from looking blurry when a canvas image is scaled. It does not give the canvas fractional output rows. A crisp 150 percent result can still have uneven stripes because it keeps hard edges and has to distribute whole target rows.
With smoothing enabled, interpolation can blend values from nearby source rows. That can make the transition look more even from a distance or make the pattern seem softer. It may introduce intermediate colors where a sharp version kept the original palette. To tell the two effects apart, zoom into the output without further smoothing. Repeated full-color rows show allocation. New intermediate edge values show blending somewhere in the path.
The CSS image rendering guide describes image-rendering choices that can affect the way a browser displays a raster at a new size. Some values can use more than one scaling step. A page preview is therefore not always the same raster an editor exported. Check the file's actual dimensions and pixel values before changing the source artwork.
Use a known-size output to isolate the extra rows
Make a small source with six contrasting one-row bands. Export one copy with nearest-neighbor scaling to nine rows, one with smoothing to nine rows and one with nearest-neighbor scaling to twelve. Give them different names. Open each output at native size or enlarge it by a whole number for inspection. If the first file repeats some pure rows, you can see the distribution. If the second shows blended edges, you can see the filtering. The twelve-row copy is the integer comparison.
Our guide to what makes a pixel art GIF file large explains how canvas dimensions affect the size of a stored animation. Here you can use a still image first so the changed row counts are easy to inspect. Compare one scale operation at a time. If the native file looks right but a preview changes when you resize the window, that preview is another transform.
The Divoom Pixoo-64 II provides a known 64 by 64 grid for preparing a preview. Use the destination grid for the artwork you intend to show on it or another pixel display. If you need equal row thickness at a size that would require a 150 percent resize, a redraw at the target size can be more predictable than hoping the app's scaling chooses the rows you want.
Divoom Pixoo-64Ⅱ WiFi 64x64 Pixel Art Display LED & Contador de mídia social
Divoom Pixoo-64 WiFi pixel art display with 64x64 LED panel. Social media counter, custom animations, clocks, and community art. App controlled frame.
$113.99 USD
$149.99 USD
com o código BF26PIXOO64
Buy now
For a hand-tuned result, fix the part that actually reads unevenly. You may be able to redraw just the border or stripe rhythm at the target size while leaving the rest of the artwork alone. Inspect it again at native size. A correction that looks balanced at large zoom can shift the silhouette at normal viewing size.
If your source has an odd height, multiplying by 1.5 can also produce a fractional total size. Check the exported height in pixels before comparing individual rows. A tool that rounds upward and a tool that rounds downward can both report a 150 percent setting while producing different actual target grids.
FAQs About Uneven Rows in 150 Percent Pixel Art Scaling
Why are there one-row and two-row stripes in the same output?
A 150 percent scale averages one and a half output rows per source row. A nearest-neighbor result has to allocate whole rows.
Does turning smoothing off fix uneven widths?
It keeps hard edges in a typical canvas resize. It does not divide an output pixel into fractions, so the uneven allocation can remain.
Why does another program repeat different rows?
It may use a different coordinate convention, tie-breaking rule or rounding method. Compare the actual stored pixels in each output.
Is 200 percent always better for a native grid?
An aligned integer enlargement can replicate each source row uniformly. Choose a scale that also fits the destination and meets the visual purpose.
Can the file look correct and the browser preview look uneven?
Yes. Browser layout and screen mapping can rescale the exported image again. Inspect the file separately from a fitted or zoomed preview.
Conclusion
A 150 percent resize has to fit fractional source-cell sizes into a whole-pixel target. Count the rows in the exported file to separate hard-edged repetition from smoothing. Use an integer scale where it fits or redraw the critical parts at their final size.
