APNext Architecture

If you generate buildings, interiors, or concept structures and you're tired of typing "modernist glass villa" for the thousandth time, this is the shortcut. APNext Architecture is a dropdown prompt builder loaded with architecture-specific word lists: real named buildings (the Gherkin, Aqua Tower, the Acropolis Museum), 200-odd architectural styles, 158 named architects from Gaudí to Alvar Aalto, plus materials, interior design schemes, room types and some genuinely fun extremes like "built into a glacier."

You wire it in front of your text encoder, flip the categories you care about to Random or a specific pick, and it hands you an assembled architectural prompt. It's a member of the APNext family, so it works exactly like its siblings - same controls, different vocabulary.

How it works

Each dropdown is a category file. None sits it out, Random draws one entry, Multiple Random draws several, or you pin an exact value. Your typed prompt is the base; the picks get appended and joined by the separator. The seed makes a given combination repeatable, which matters here - when you land a building-plus-style-plus-architect combo that renders well, you want to lock it.

Naming a specific architect or a real building is the highest-leverage thing this node does, and it's the same reason naming a physical camera works for photos: these are concrete referents the model has actually seen captioned, not vague quality words. "in the style of Zaha Hadid" moves an image; "beautiful architecture" does almost nothing.

Inputs and outputs that matter

• prompt (multiline, required) - your base scene. "a lakeside house at dusk", then let the categories dress it.
• architect / buildings / styles - the three that change everything. An architect name sets the whole aesthetic; a named building borrows a real silhouette; the style list is your broad genre dial.
• materials and interior_design - reach for these when you're doing interiors or want to control surface and texture.
• seed - lock the winners, roll for more.
• attributes (boolean) - folds in the extra descriptor tags some entries carry.

Outputs are prompt (base plus assembled architecture terms, ready for CLIP Text Encode) and random (just the randomly chosen additions, if you want to reuse only those).

Installing it

ComfyUI Manager: search comfyui_dagthomas, install, restart. Or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/dagthomas/comfyui_dagthomas
cd comfyui_dagthomas
pip install -r requirements.txt

The full pack pulls a large dependency list for its LLM and image-FX nodes, but this prompt node needs none of it - no API keys, no model downloads, just the word lists that ship in the repo.

Common issues

Turn on Multiple Random across buildings, styles, homes, and interior all at once and you'll get a contradictory prompt - a brutalist bungalow that's also a glass cathedral. The model averages it into mush. Pick one primary style driver (architect or named building or style) and keep the rest tight.

Worth knowing which model you're feeding: these lists are written in the SDXL tag idiom, occasionally with weighted parentheses. That lands well on SDXL, Illustrious and Flux. On the newest LLM-encoded checkpoints the weighting is ignored and the model would rather read a sentence - so on those, treat this node as an idea generator and rephrase its output into prose. Architecture and interiors are genuinely a strong subject for it; the vocabulary is deep and the named references are what make renders look designed instead of generic.