TL;DR / Summary: Sprite sheets behave when tile grids follow powers of two: 8×8, 16×16, 32×32 frames pack cleanly and slice without fractional pixels. Pick a frame size with a 2ⁿ habit, compute sheet dimensions before exporting, and verify awkward totals with tothepowerof2.com so your atlas importer stops complaining about off-by-one crops.
Fractional tiles are a pipeline tax
Art tools will happily export a 47×47 frame. Engines and texture packers will just as happily introduce seams, bleeding, and UV math that only fails on one device. Powers of two will not fix every filtering issue, but they remove an entire class of “why is there a one-pixel hairline” bugs that appear after compression.
Indie teams feel this hardest when a contractor delivers sheets sized for print mockups instead of runtime atlases. Agreeing on 2ⁿ frame sizes early is cheaper than rewriting crop metadata the week before a build.
Choose the frame first, then the sheet
Decide whether characters live on 16×16, 32×32, or 64×64 cells. That choice is your exponent story: 2⁴, 2⁵, 2⁶. Animation length then becomes “how many cells,” not “how many arbitrary rectangles.” A walk cycle of eight frames on 32×32 cells wants a strip 256 pixels wide—another power of two—which packers love.
If you need nine frames, you can still place them on a 2ⁿ cell grid and leave empty cells. Empty cells are cheaper than non-aligned frames that fight every mipmap level.
Compute sheet width as columns × frame size
Write the formula on a sticky note: sheet width = columns × frame width; sheet height = rows × frame height. When columns and frame width are both powers of two, width usually is too. Mixed cases need a quick check. Example: 12 columns of 16-pixel frames → 192 pixels wide. 192 is not a power of two, which may still be fine for an atlas page, but you should know you left the comfort zone.
Some engines prefer the entire texture page at 2ⁿ even if the logical grid is irregular. In that case, pad to 256 or 512 and document the used rectangle. Padding is intentional; surprise non-power sizes are not.
Mipmaps, compression, and why exponents keep showing up
GPU mipmap chains repeatedly halve dimensions. Starting from a power-of-two texture keeps those halvings integer for longer. Compression formats often assume block sizes like 4×4—again powers of two. Your sprite sheet does not need to be sacred geometry, but aligning with those constraints reduces “works in the editor, sparkles on device” nights.
Students can practice by listing 2⁰ through 2¹⁰ and matching common game sizes: 1, 2, 4, 8, 16, 32, 64, 128, 256, 512, 1024. That list should be muscle memory before the first atlas export.
Workflow that survives handoffs
Put frame size, columns, rows, and padded texture size in the art README. When a new animator joins, they should not reverse-engineer a PSD layer name. Add a one-line note: “Verify 2ⁿ with the team calculator if you change frame size.” Tools exist so arguments stay short.
Version control diffs on giant PNGs are painful. Prefer regenerating sheets from source frames with a script that encodes the power-of-two rules. Humans eyeballing crop boxes will eventually miss one pixel.
When breaking the rule is acceptable
UI nine-slices, photographic backgrounds, and some modern atlases intentionally avoid strict 2ⁿ pages. That is fine when the engine path is tested. What is not fine is mixing strict tile gameplay sprites with accidental 47-pixel frames in the same importer profile. Separate pipelines; do not hope.
Behaviour when frames are not square
Characters rarely need square cells. A tall wizard sprite in a 32×64 cell is still power-of-two on both axes, and the same arithmetic applies. The mistake is mixing shapes within one sheet, where a wide portrait frame sits beside a tall banner frame and the importer has no single grid to slice.
Group frames by aspect ratio into separate sheets. Two atlases with consistent cells are easier to maintain than one atlas where every rectangle is a special case, and the art pipeline stays predictable as the roster grows.
If a boss needs an oversized frame, give it its own sheet at the next exponent up rather than squeezing it into the character grid. Scaling irregular frames during slicing is where bleeding and off-by-one crops originate.
Memory budgets on mobile
Texture memory grows with area, not with frame count, so doubling a frame from 32 to 64 quadruples the space each cell consumes. Powers of two make that growth obvious and predictable, which helps when a budget conversation happens late in a project.
When you are near a ceiling, the cheapest saving is usually a smaller exponent on the least-visible assets rather than reallocating the whole atlas. Splitting into a high-resolution sheet for close-up characters and a lower one for background props is often the realistic compromise.
Document the chosen exponent per sheet in the README. The next person optimizing memory needs that number before they need the artwork.
A five-minute preflight before export
Ask: Is frame width 2ⁿ? Is frame height 2ⁿ? Is the padded page 2ⁿ if the engine requires it? How many frames fit without fractional cells? If any answer needs a calculator, open one. Guessing under deadline is how hairline seams ship.
Pack frames on exponents you can explain
Sprite pipelines stay boring—in the good way—when cell sizes follow powers of two and sheet math is written down. Confirm awkward products on the powers-of-two tool, pad deliberately, and hand importers rectangles that divide cleanly. Players notice polish; they should never notice your crop metadata.


