Chinese, Japanese and Korean Text in Unity Without Huge Font Files

Last reviewed October 31, 2026

The usual options

• The size of the fonts. A full CJK font is large: Noto Sans CJK Simplified Chinese Regular is about 16 MB per weight, Source Han Sans about the same, and even the regional subsets run from about 4.5 to 8.4 MB.
• TextMesh Pro static atlases. Thousands of characters are baked into Alpha8 textures, about 1 MB at 1024² and 4 MB at 2048², split across a main asset and fallbacks, from character lists the project has to maintain.
• TextMesh Pro dynamic atlases. Dynamic font assets add glyphs at runtime but include their source font in the build. Dynamic OS font assets read a system font instead and are the way to avoid shipping one; they are added to fallback lists by hand.
• One shape for all three languages. TextMesh Pro looks glyphs up by code point only and has no language setting, so the regional form depends on which font asset in the fallback chain holds the character. Users report Chinese text drawn with Japanese glyph shapes, and the usual workaround swaps font assets per locale.
• Line breaking. TextMesh Pro breaks CJK text between characters, guided by the leading and following character lists in its settings, rather than the Unicode line-breaking algorithm.

How UniText handles CJK

• System fonts first. Characters no font in the stack covers come from the operating system's font for that run's language, through DirectWrite, CoreText, AFontMatcher or fontconfig. Nothing ships for CJK on desktop and mobile.
• Right glyphs for the language. The Language property or <lang=ja>, <lang=zh-Hans>, <lang=ko> ranges turn on the OpenType locl feature and prefer a font family declared for that language, so the same code point is drawn the Japanese, Chinese or Korean way.
• No baked atlas. Glyphs are rasterized when text first shows them, so only the characters actually displayed take atlas space, and player names or chat never meet a missing glyph.
• Smaller shipped fonts. Font bytes are Zstd-compressed into the font asset, the Font Subsetter keeps only the characters a text corpus or character set needs, and Font Memory Mapping reads large files on demand instead of loading them whole.
• CJK line breaking. Chinese and Japanese break between characters by the Unicode rules; an optional CJK dictionary adds word selection and word-by-word animation.

WebGL has no access to system fonts, so a WebGL build assigns CJK fonts, ideally subset, or downloads font bytes at runtime with UniTextFont.CreateFontAsset(byte[]).

Questions