Why emotes live or die at 28 pixels

An emote is not really an image. It is a 28-pixel-wide piece of typography that has to read inside a sentence, repeatedly, at speed.

This guide is about the specific properties that survive the downscale, and the ones that always fail — so you can stop guessing which details matter.

What downscaling actually does

Reducing a 1024px image to 28px averages roughly 1300 source pixels into each output pixel. Anything smaller than that averaging window does not shrink — it blends into its neighbours.

A 2px line inside a 1024px artwork becomes a fraction of one output pixel. A soft gradient over 40 pixels becomes a single flat value. Fine hatching becomes a uniform mid-tone. None of these are "preserved at lower quality"; they are arithmetically destroyed.

The four things that survive

• Silhouette — one contiguous shape. If the emote breaks into disconnected pieces, each piece gets a fraction of the already-tiny pixel budget and the whole thing turns to noise.
• Contrast against the chat background. Twitch chat is #18181B in dark mode and near-white in light mode. An emote must separate from both; a mid-luminance subject with a rim in a different value is the standard solution.
• Hard edges. Flat fills and heavy outlines downsample to clean stair-steps. Soft shading and anti-aliased gradients become semi-transparent mush, because almost every edge pixel ends up partially transparent.
• Framing. A subject that fills a healthy share of the square has more pixels to spend. One that floats small in the middle reads as a speck.

The four things that never survive

• Text. Five letters across 28 pixels is about five pixels per glyph. Two large characters is the practical maximum — "GG", not "nice one".
• Thin linework. Hair strands, whiskers, hatching, fine outlines: all of it averages away.
• Gradients and soft shading. They become flat mid-tones, which often reduce contrast rather than adding depth.
• Small emotional detail. Individual teeth, eyelashes, tiny tears. The expression has to live in the eyes and mouth at a scale that is far larger than looks correct at full resolution.

Why the eyes and mouth decide everything

Chibi proportions are not a fashion. A large head with oversized eyes allocates the most pixels to the parts that carry meaning. On a 28px canvas a face might get 20 pixels across, and the eyes have to be a meaningful share of that to register as an expression at all.

This is also why so much good emote art oversizes features relative to "correct" anatomy: at 112px it looks exaggerated, and at 28px it looks right.

Concept prompts built for 28px

Checking yours

The reliable test is not a squint, it is the actual downscale. Our checker resamples to 28×28 with a premultiplied area filter, then measures silhouette coherence, contrast against both chat themes, edge softness, canvas fill and border bleed — and returns a score with specific fixes.

Run the readability checkerSee it in a simulated chat line

Questions