Most people think of emojis as digital stickers. That framing undersells them.
📦 Some emoji may appear as blank boxes on your device
This post references emoji from Unicode 14.0–15.1 (2021–2023). If your OS or browser font hasn't been updated recently, newer characters will render as empty rectangles. That's tofu. We'll cover it in detail later.
Emojis are a rigorously engineered symbolic layer on top of text, governed by international standards, shaped by corporate interests, and loaded with cultural context most Western users never think about. The body that standardises them, the , includes representatives from Apple, Google, Microsoft, and Meta. Every emoji on your keyboard survived a vote, a naming debate, and a usage-frequency argument before it got there.
This post covers the corporate Easter eggs hiding in plain sight, the emojis you've been misreading for years, why the pregnant man exists, why no new flags are coming, and how the rendering pipeline actually works. There are interactive bits along the way. Try them before reading on.
Vibe check + thoughts
How was this read?
No sign-in needed to react — only comments require an account.
Comments
Crickets. Loud ones.
Be the first to say something — good, bad, or unhinged.
Emojis are a rigorously engineered symbolic layer on top of text, shaped by corporate interests, Unicode committees, font format politics, and geopolitical decisions. Here's how the whole thing actually works.
Before Unicode, before Apple, before any of this, there was Shigetaka Kurita, a 25-year-old designer at NTT DoCoMo in Tokyo working on a mobile internet platform called i-mode.
The year was 1999. i-mode let subscribers send short messages, but the text interface was sterile and cold. Weather forecasts used words. Emotional tone was invisible. Kurita's job was to make the interface feel human without increasing character limits, which the network couldn't support.
His solution: A 12×12 pixel grid for each symbol. He drew 176 of them by hand, pulling visual vocabulary from manga conventions (the snot bubble for sleep, the sweat drop for anxiety), Japanese weather maps, and street signage. The original emoji set wasn't expressive by accident; each symbol was designed to carry a specific, unambiguous meaning within the constraints of a tiny monochrome screen.
NTT DoCoMo's competitors, SoftBank and au by KDDI, immediately released their own incompatible emoji sets. By 2000, Japan had three parallel emoji standards with no interoperability. A message sent from one carrier's phone to another rendered as garbage characters. This was the original emoji compatibility problem. It never fully went away. It just scaled up to a global level.
When Apple launched the iPhone in Japan in 2008, they needed emoji support to compete. The SoftBank emoji set shipped on that device, which is why several original Apple emoji designs trace directly back to SoftBank's early 2000s graphics. When Google added emoji support to Android to reach the Japanese market, they mapped their own set. The Unicode Consortium's standardisation effort in 2010 was, in large part, a cleanup operation for this inherited chaos.
Kurita's original 176 emoji are now in the collection of the Museum of Modern Art in New York. The set that started as a workaround for a pager network is now considered a legitimate artefact of design history.
Open your calendar emoji 📅 on iOS, Google, or Samsung. It shows July 17. Not a default or placeholder, a deliberate choice made over two decades ago.
On July 17, 2002, Apple unveiled iCal at the Macworld Expo in New York their first desktop calendar app. When Apple designed their original emoji set, designers chose that date as a quiet memorial to the product's launch. The original iCal product manager confirmed the date was tied directly to the announcement day.
In 2014, Jeremy Burge - founder of Emojipedia, declared July 17 as World Emoji Day, specifically because of the Apple icon. Once it became a global event, other platforms had reason to fall in line. Google switched in 2016. Samsung and Twitter followed to avoid user confusion on the day itself.
Not everyone complied. WhatsApp shows February 24, the day the company incorporated in 2009. Facebook traditionally shows May 14, Mark Zuckerberg's birthday.
The same logic applies to Apple's clock emoji 🕐. It consistently shows 10:09, the hands arranged in the V-shape that most watch advertisements use, because symmetric positions look balanced on a circular face. That's not an emoji-specific decision; it's a century-old watchmaking convention that Apple carried over without modification.
Apple embeds nearly unreadable text into several of its symbols.
Apple's Page Facing Up emoji (📄) contains a full Easter egg, a letter addressed "Dear Katie," signed "Take care, John Appleseed," carrying the text of Apple's 1997–2002 Think Different ad campaign. According to Emojipedia, the same hidden letter appears across the Scroll (📜), Clipboard (📋), and Open Book (📖). At normal emoji size it's indistinguishable from generic scribbles. You need significant zoom to read it.
John Appleseed is widely believed to be the alias of Mike Markkula, the second CEO of Apple Computer, Inc., who used the name while publishing software for the Apple II. Apple has never officially confirmed this connection, but it has been treated as an open secret in the industry for decades. Apple has used the name as a placeholder in software demonstrations and demo accounts ever since.
This pattern isn't unique to Apple. Google's original Noto emoji font embedded a tiny "G" inside several punctuation glyphs in early versions as an internal tracking marker. Samsung's older emoji set had watermarks in the transparency layer of several face emojis, invisible unless you extracted the raw PNG from the font file.
Emojis originated in Japan in the late 1990s. The original set was culturally Japanese, so many carry meanings that Western users consistently get backwards.
A 2015 YouGov survey found 780 out of 1,000 Americans read the Sleepy Face emoji as "sad" or "crying", making it the most cross-culturally misread emoji in that study.
The droplet is not a tear. It comes from Japanese manga and anime, where a snot bubble signals that a character has fallen asleep, often at an awkward or comedic moment. The bubble swells and shrinks with the character's breath and pops on waking. Sleep, not sadness.
The official Unicode name is "Face With Look of Triumph." The steam represents the derisive snort of someone winning, pride after a hard task, not hostility. Western users read it as rage. The character originated from Japanese manga shorthand for a smug, victorious exhale.
"Person Bowing" depicts dogeza - a deep, prostrating bow used in Japanese culture for extreme apology or to ask someone a very large favour. Western users frequently read it as exercise, meditation, or deep focus.
The poop emoji originated in Japan, where feces had become associated with humour after an anthropomorphized poop appeared in Dr. Slump, Akira Toriyama's popular manga. Kurita created it in 1997 for phones sold by J-Phone. In Japan it carries luck, the word for poop (unko) starts with the same sound as the word for luck (un).
When Google prepared to add it to Gmail, internal teams resisted. A direct appeal from Google's Japanese product manager, explaining its cultural popularity, eventually cleared it.
The popular theory that the poop emoji and soft-serve ice cream share a swirl isn't far off, in Dr. Slump itself, poop characters meet a Soft-Serve-Ice-Cream Boy they mistake for a poop. The visual confusion is built into the origin.
In Western usage: hello or goodbye. On WeChat in Chinese digital culture, sending this wave can carry the meaning "I don't want to be your friend anymore." Same picture, entirely different message depending on who's receiving it.
In 2019, reports from distributed teams (including Slack engineering discussions) showed that globally distributed teams had significantly higher rates of emoji-related misunderstandings than single-region teams. The wave, the folded hands (🙏, which reads as prayer in Western contexts and "please" or "thank you" in Japanese), and the OK hand (👌, which has offensive connotations in several countries) topped the list.
In 2015, Unicode 8.0 added skin tone modifiers, five shades derived from the Fitzpatrick scale, a dermatological classification system developed in 1975 by Thomas Fitzpatrick to measure how human skin responds to UV radiation.
The modifiers are code points U+1F3FB through U+1F3FF, representing five skin tones from light to dark. When appended to a compatible human-form emoji, they shift the rendering. No modifier means the default, which Unicode specifies should be a non-realistic yellow, following the convention of avoiding any particular ethnic association.
The adoption of the Fitzpatrick scale was contested from the start. Critics pointed out that five points is medically arbitrary, the scale was designed for clinical sunburn prediction, not digital representation. It doesn't map cleanly to the diversity of human skin tones in practice. The Unicode Consortium's position: the scale was an existing, documented, neutral standard, which made it defensible in a way that an invented five-point gradient wouldn't be.
Hair colour modifiers, including red hair (🦰), curly hair (🦱), white hair (🦳), and bald (🦲), were added in 2018 via Unicode 11.0. These combine through more complex ZWJ sequences, which many platforms still don’t fully support. A curly-haired, medium-dark-skinned woman technologist emoji is: U+1F469 (woman) + U+1F3FE (medium-dark skin) + U+200D (ZWJ) + U+1F4BB (laptop), the curly hair can only be achieved by using a separate hair-component sequence, which most platforms don't fully support yet.
The gap between what the standard allows and what platforms actually render is wide. Unicode defines the sequence. Whether your device has a font that draws it is a separate question entirely.
The Unicode Consortium accepts emoji proposals from the public. The bar is higher than most people expect.
A submission requires a completed proposal document covering: evidence of distinctiveness (it can't be too similar to an existing character), expected usage frequency backed by data, multiple use cases across diverse communities, compatibility with existing emoji design principles, and examples of how it would look at 72×72px.
The Consortium evaluates proposals against a published checklist. Characters with high "compatibility" scores, meaning they work as parts of sequences, have gender or skin-tone variations, or fill a gap in an existing category, score better. Characters that are "too specific" (a single food from a single country, a niche sports term) tend to fail.
The review cycle runs annually. From proposal submission to appearing on your keyboard typically takes two to three years: one year in the Unicode pipeline, then six to twelve months for platform vendors to implement and ship.
The most-rejected category is food. Proposals for specific national dishes arrive constantly and almost always fail on distinctiveness grounds. The most successful recent proposals have been ones filling genuine sequence gaps, like the handshake skin tone combination (🫱🏿🫲🏻), which required a new encoding approach because existing handshake sequences couldn't carry two independent skin tone modifiers.
The most consistently rejected category besides food is logos and brand marks. Several tech companies have submitted proposals for brand-adjacent emoji (a specific laptop design, a specific phone shape). All rejected. The Consortium has a firm rule against proprietary visual references in a universal standard.
Some rejections are instructive. The bagel was rejected twice before finally being accepted in 2018, initial designs showed a plain bagel, which testers found indistinguishable from a donut at small sizes. Only after Emojipedia ran a public campaign and the proposal was revised to include cream cheese did it pass.
The middle finger (🖕, Unicode 7.0, 2014) made it only after significant internal debate. The deciding argument: the character had already existed in SoftBank's pre-Unicode Japanese emoji set, so rejecting it would create a compatibility break. Legacy obligations won.
In 2019, the Unicode Emoji Subcommittee (ESC) adopted the Gender Triplet model: every human-form emoji needs three variants, a woman, a man, and a gender-neutral person. The neutral variant is the intended default. According to Unicode Technical Standard #51, human-form emojis should be gender-neutral unless a specific gender is desired.
The pregnancy category was the outlier. Pregnant woman (U+1F930) had existed since 2016, added before the triplet model, leaving pregnancy as the only human-form emoji category stuck with a single gender. The ESC proposed the man (🫃) and neutral (🫄) variants in 2021 specifically to fix that.
Other emojis went through the same correction. Santa Claus got supplemented by Mrs. Claus and Mx. Claus. "Man in Tuxedo" became "Person in Tuxedo." "Bride with Veil" became "Person in Wedding Dress."
During the 2020–2021 review cycle, feedback proposed naming these emojis "man with swollen belly" and "person with swollen belly", arguing the names were more descriptive and accommodated use cases like food baby or general bloating.
The ESC rejected this on data integrity grounds. In their formal response (L2/21-055), the committee noted that pregnancy and a swollen belly aren't synonymous, and naming them differently from the existing "pregnant woman" would semantically detach the new characters from it, breaking historical data comparisons across the triplet. The names stayed: PREGNANT MAN and PREGNANT PERSON.
The food baby use case wasn't ignored. The CLDR, the project that assigns search keywords to every emoji in every locale, added "bloated," "full," "overeat," and "belly" to the emoji's keyword list. The name is a stable identifier; keywords are the semantic surface area.
An iOS user sending what they saw as a playful squirt gun was being received as a realistic firearm on Android. A jokey message could read, on the recipient's screen, as something threatening. The same code point, two different meanings and no technical fix short of platform convergence.
Then in July 2024, X reverted to a realistic firearm, specifically a semi-automatic M1911, not the original revolver, rolling out on July 18, 2024, the day after World Emoji Day. The semantic gap is open again for anyone using X's web client.
When a platform changes what a code point looks like, they change what a message means in transit. Emoji design decisions are not cosmetic.
As of March 2022, the Unicode Emoji Subcommittee officially announced a moratorium on new flag emojis. No new national or regional flags will be accepted as proposals.
Today, flags are already the largest emoji category, over 200 flags out of roughly 3,600 total emoji. But ISO 3166-2 recognises over 5,000 geographically-valid regions including states, provinces, and autonomous territories. Adding sub-national flags at any scale would require a different encoding architecture entirely, with an effectively unlimited long tail of rarely-used icons.
There's also a harder problem: borders and political identities change. Once a Unicode character is added, it cannot be removed ever. A flag emoji added today represents a political reality that may not exist in ten years, but the code point stays in the standard permanently.
Subdivision flags compound the problem. A flag for Catalonia creates immediate pressure for Northern Ireland, Kurdistan, and dozens of other contested regions. There's no principled stopping point short of the moratorium itself.
The ESC's position: flags are better handled through stickers, GIFs, and image attachments, decentralised distribution that doesn't require international standards consensus to expand.
Flag encoding itself is architecturally unusual. National flags use Regional Indicator Symbols, a pair of code points corresponding to the ISO 3166-1 country code. 🇮🇳 is U+1F1EE (Regional Indicator I) + U+1F1F3 (Regional Indicator N). Subdivision flags use a completely different mechanism: a tag sequence starting with U+1F3F4 (Black Flag), followed by subdivision code tag characters, terminated with U+E007F. The Scottish flag 🏴 is seven code points. At 5,000+ regions this becomes unworkable.
Everything above describes what emojis mean. This section is about how they exist. If you're not interested in the mechanics, the post ends cleanly above.
Unicode is an international encoding protocol that assigns a unique numeric value, a code point, to every character in every writing system on earth. It's the reason you can send a message in Malayalam, receive a reply in Arabic, and have both render correctly on the same screen.
Emojis are text characters, not images injected into a chat. When you send 🔥, you're sending the code point U+1F525. Whatever device receives it is responsible for rendering it. Unicode defines characters, not glyphs. What you see is an implementation detail. This is why the fire emoji looks different on iOS vs Android vs Windows. You sent the same code point; each platform drew its own picture.
Most modern emojis live in the Supplementary Multilingual Plane (SMP), specifically U+10000 to U+1FFFF, a block designed for complex pictographic data that exceeds the capacity of the legacy Basic Multilingual Plane. That said, not every emoji lives there. Legacy symbols like the anchor (⚓, U+2693) and warning sign (⚠️, U+26A0) have been part of the BMP since long before the emoji era, they were repurposed into emoji use via Variation Selector 16. It instructs a renderer to display the character as a coloured emoji glyph rather than a monochrome symbol.
Variation Selector 16 (VS-16, U+FE0F) is the mechanism that forces emoji-style rendering. Without it, many characters default to text presentation. For example, a heart without VS-16 renders as ♥ (text), while with VS-16 it becomes ❤️ (emoji).
This placement has a practical consequence for older software. JavaScript's String.prototype.length counts UTF-16 code units, not Unicode code points. An SMP emoji requires two UTF-16 code units, a surrogate pair, which is why "🔥".length === 2 in JavaScript, not 1. For correct grapheme counting in 2026, Intl.Segmenter with granularity: 'grapheme' is the most reliable native approach, simple array spreading still fails on complex ZWJ sequences like multi-person families:
const seg = new Intl.Segmenter(undefined, { granularity: "grapheme" });console.log("🔥 length:", "🔥".length);// 2 - counts UTF-16 code units (surrogate pair)console.log("🔥 spread:", [..."🔥"].length);// 1 - counts Unicode code points (correct for simple emoji)console.log("family spread:", [..."👨👩👧👦"].length);// 7 - counts code points: 👨 + ZWJ + 👩 + ZWJ + 👧 + ZWJ + 👦// not user-perceived charactersconsole.log("👩💻 graphemes:", Array.from(seg.segment("👩💻")).length);// 1 - counts grapheme clusters (what users see as one emoji)console.log("family graphemes:", Array.from(seg.segment("👨👩👧👦")).length);// 1 - entire family sequence rendered as a single grapheme
Most people assume one emoji equals one code point. For simple emojis like 😊 that holds. A huge number of emojis you use daily are sequences of multiple code points assembled invisibly.
The 👩💻 (Woman Technologist) emoji can consist of up to five code points:
U+1F469 - Woman
U+1F3FB - Skin tone modifier (Fitzpatrick scale)
U+200D - ZWJ, the invisible glue
U+1F4BB - Laptop
VS-16 - Variation Selector 16, which tells the renderer to display this as a coloured emoji rather than a black-and-white symbol
The ZWJ (U+200D) is the critical piece, a character with zero visual width that instructs a compliant renderer to merge surrounding characters into a single combined glyph. When your device supports the sequence, it draws one image. When it doesn't, you see each piece separately: a woman, a laptop, side by side.
This modular approach lets emoji designers produce hundreds of skin tone, gender, and profession combinations without exhausting unique code points. Instead of one code point for "woman with medium skin tone fixing a lightbulb," you get: woman + skin modifier + ZWJ + lightbulb.
The pregnant man emoji (🫃) is an atomic character, its own unique code point (U+1FAC3) rather than a sequence, specifically to match the pregnant woman character, which predated the sequence-based gender system. Skin tone variations are still sequences, built as U+1FAC3 + the relevant Fitzpatrick modifier.
When your device doesn't support a newer emoji, you see a small blank rectangle, an empty box or a box with a question mark. This is tofu, named after the plain white block of soybean curd.
Tofu appears when the sender's device uses a newer Unicode version than the recipient's font supports. The code point arrived correctly; the device just has no picture to match it against.
Google's emoji font is called Noto, short for "No Tofu." The project's goal is to support every character in the Unicode standard so no user ever sees a blank box. Noto currently covers over 1,000 languages and scripts, making it one of the largest typeface design projects ever undertaken.
The version gap is more significant than most users realise. Unicode 15.1 (2023) added 627 new characters. iOS ships new Unicode support roughly annually. Android's update cadence depends on the device manufacturer. A new emoji sent from an iPhone 15 to an older Android on a carrier that hasn't pushed a system update in 18 months will show as tofu, not because the sender or recipient did anything wrong, but because the fragmentation is baked into the distribution model.
Standard fonts are monochrome vector drawings. Emoji fonts need full colour, multiple resolutions, and transparency, which required special font table extensions. Different companies implemented these differently, which explains why the same emoji looks so distinct across platforms.
The four main formats:
sbix (Apple) - Stores each emoji as a stack of embedded PNG images at different resolutions. When your iPhone renders an emoji, it selects the bitmap size that matches the display's DPI. This produces the glossy, high-detail look characteristic of Apple Color Emoji. On macOS, where assets include high-res bitmaps optimised for 5K displays, AppleColorEmoji.ttc exceeds 160MB, on iOS the same assets are significantly more compressed and optimised to conserve storage.
CBDT/CBLC (Google, Android original) - Uses compressed PNG data embedded inside the font file. Same concept as sbix, different implementation. Also produces large font files.
COLR/CPAL (Microsoft, now Google) - A vector-based format that defines layered colour fills. Google has moved toward COLRv1, which supports gradients and complex blending while keeping file sizes small. Noto Color Emoji in COLRv1 format is roughly 9MB, a 94% reduction from the equivalent bitmap format. More scalable at the cost of rendering complexity.
SVG-in-OpenType (Adobe/Mozilla) - Embeds a full SVG document per glyph. Most flexible for graphic complexity and infinite scalability, heavier on file size. Used primarily in Firefox.
When the same emoji looks "cuter" on Samsung and more "realistic" on iOS, that's partly aesthetic choice and partly the rendering architecture. Apple's sbix bitmaps are drawn by hand at multiple sizes in a specific illustrative style. Google's COLRv1 vector layers are mathematically defined and scale perfectly, the art direction differs because the designers and toolchains differ.
There's also a temporal dimension. Older versions of an emoji can persist in the wild for years. If a platform updates an emoji's appearance, which Apple, Google, and Samsung do regularly, that change only applies to new renders. Screenshots, cached images, and older devices continue showing the old design. Emojipedia maintains a full version history of every emoji on every platform specifically because the "same" emoji isn't always the same.
When you type "fire" and 🔥 appears in suggestions, your keyboard runs a search algorithm against a database of emoji names and keywords.
Early keyboards used lexical search, exact matching against the official Unicode character name. The limits are obvious. The official name for 🔈 is SPEAKER LOW VOLUME. Search "audio" and you get nothing.
Modern keyboards use semantic search backed by machine learning. Google's Gboard represents emojis as numerical vectors in a high-dimensional space, mapping visual appearance and descriptions mathematically. The system understands that 🐶 relates to "puppy" even though "puppy" doesn't appear in the official Unicode name.
The CLDR keyword system bridges both approaches. CLDR maintainers manually curate keyword lists for every emoji in every locale. The pregnant man emoji has "bloated," "full," "overeat," and "belly" in its keyword list, not because those are Unicode names, but because the committee knew users would search that way.
CLDR also handles localisation. The 🍚 (Steamed Rice Bowl) emoji searches differently by locale: in Japanese locales it surfaces for "gohan" (the word for both rice and meal); in Indian locales it surfaces for "chawal." The English keyword list is completely separate from the Japanese one, maintained independently. There are currently CLDR keyword sets for over 60 locales, each curated by native speakers.
When you tap 👩💻 and it appears on your screen, here's the full sequence:
Input - The keyboard sends the code point sequence to the OS text input system.
Storage - The sequence is stored as UTF-8 or UTF-16 in the text buffer, typically around 15–20 bytes in UTF-8 depending on modifiers.
Layout - The text layout engine (CoreText on iOS, HarfBuzz on Android, DirectWrite on Windows) reads the sequence, identifies the ZWJ boundaries, and determines this is a single grapheme cluster requiring a single glyph slot.
Font selection - The layout engine queries the font fallback chain. If the primary font doesn't contain this glyph, it walks the chain until it finds one that does.
Glyph lookup - The font engine looks up the glyph ID for this sequence in the font's GSUB table, which maps character sequences to single combined glyphs.
Rendering - Depending on the format (sbix, COLR, SVG), the engine rasterises or composites the layers at the appropriate size and DPI.
Display - The resulting pixels are composited into the frame buffer and shown on screen.
Steps 3–5 happen in under a millisecond on a modern device. The full round-trip from tap to pixel is typically under 16ms to hit a 60fps frame. Emoji rendering is fast not because it's simple, but because the entire stack, from Unicode spec to font format to layout engine, has been optimised over 15+ years of international standardisation, beginning in earnest with the Unicode 6.0 cleanup in 2010.
The emoji you tap is a small piece of a very large system: a standards committee that's been meeting since 2000, naming debates conducted in formal proposal documents, font format competition between Apple and Google that plays out in file sizes, and geopolitical decisions about what counts as a country.
Next time you send 😤 expecting frustration and it lands as a flex, you're not looking at a misunderstanding. You're looking at the edge of a system that was never designed to carry meaning universally and still tries anyway.