Omost
lllyasviel/Omost · Python · Apache-2.0
LLM-composed canvases for structured concept art
この解説はまだ日本語がありません。英語の原文を表示しています。
Omost lets an LLM plan a canvas then fill regions — useful when game concept art needs layout control, not just a single prompt.
スコア
導入の要点
git clone https://github.com/lllyasviel/Omost.git && pip install -r requirements.txt 深読み
Omost is lllyasviel's compositional image generator: a coding LLM writes code that draws your image on a Canvas — every element placed, sized and layered explicitly instead of hoping a diffusion model guesses the layout.
That code-first approach makes it unusually good at structured scenes — multiple characters, precise object placement, UI-like compositions — which is exactly where pure prompt-based generators struggle.
It runs as a Hugging Face Space and a local Gradio app, with demo videos in the repo showing the code-to-image loop in action.