Glyph-ByT5 Fonts

This is the text-layout node for the Glyph-ByT5 side of the pack. Glyph-ByT5 is a Microsoft research project - a character-aware text encoder fine-tuned from ByT5 that, bolted onto SDXL, produces "Glyph-SDXL" and can render whole paragraphs of accurate text, in around ten languages, at a spelling accuracy the original SDXL couldn't touch. UL_Image_Generation_Glyph_ByT5_Font is where you specify what that text says, in what font, and in what color - up to six independent lines.

It's the most-searched node in this pack by a wide margin, which makes sense: multilingual paragraph text was a genuinely hard problem in the SDXL era, and this was one of the few open answers.

How it works

No model runs here - it's a parameters builder. Each of the six slots is a triple: font_textN (the actual text, multiline, so line breaks are fine), font_colorN (a named color), and font_nameN (the typeface). The font list is huge - thousands of entries, tagged by language (Chinese, Japanese, Korean, English, Russian, Portuguese and more) - which is the point, since Glyph-ByT5's whole claim is multilingual glyph accuracy. The node packs all six triples into one Glyph_SDXL_Font_Params object that the Glyph-ByT5 sampler consumes.

The inputs and outputs that matter

• font_text1 … font_text6 (multiline STRING) - the text for each line. Leave a slot's text as-is or blank if you don't need all six. The defaults are a showcase of languages on purpose.
• font_name1 … font_name6 - the font per line, chosen from a language-tagged list of thousands. Match the font's language to your text (a Latin font won't render Chinese glyphs).
• font_color1 … font_color6 - named color per line, ~137 options.

Output is font_params (Glyph_SDXL_Font_Params) - wire it into UL_Image_Generation_Glyph_ByT5.

How to install it

ComfyUI Manager, search ComfyUI_Anytext, or:

cd ComfyUI/custom_nodes
git clone https://github.com/zmwv823/ComfyUI_Anytext

then restart. Glyph-ByT5 needs its own weights: the Glyph-SDXL checkpoints go in ComfyUI/custom_nodes/ComfyUI_Anytext/Image_Generation/Glyph_SDXL/checkpoints, and google/byt5-small goes in ComfyUI/models/text_encoders (or use the Auto_Download option). The fonts referenced here ship with the Glyph-SDXL font set.

Common issues & troubleshooting

A line renders as boxes or blanks. Font/language mismatch. Each font is tagged by language for a reason - pick a Chinese font for Chinese text, a Latin font for English. The huge font list is there precisely so you can match glyphs to script.

This is the layout node, not the generator. By itself it produces nothing visible. It only builds font_params; the actual image comes from the Glyph-ByT5 sampler downstream.

The bigger caveat: Glyph-ByT5 is the rough part of this pack. The author flags it as "TEST FAILED ON MY PC," and the model has real limits - output resolution is locked to 1024×1024 with a 1:1 aspect ratio, and the background quality can come out worse than plain SDXL. On top of that, Microsoft pulled the original Glyph-ByT5 weights (trained on a private dataset, pending review), so sourcing a working checkpoint is itself a hunt. Set expectations accordingly: this is an impressive research idea wrapped in a fragile, hard-to-provision node, not a dependable production path.