Superside Architectural Style Dial

If you've ever tried to prompt an interior-design LoRA and watched it render a room with three chandeliers and a table melting into the sofa, this node is the fix - and it's not the kind of fix you'd guess. It's a prompt driver, not a generator: no API key, no model call, nothing leaves your machine. You pick three things, it hands you a ready-to-use prompt fragment plus matching settings, and you wire those into your existing inpaint or LoRA workflow.

The design philosophy is what makes it work, and it's worth internalizing because it applies to any style LoRA you'll ever use. The node describes the style through general material, fabric, surface, metal, palette, and light categories - the character of transitional, traditional, or modern - and pointedly does not list furniture. The LoRA already learned what a transitional living room looks like; if you also cram "wingback chair, coffee table, floor lamp, credenza, throw pillows" into the prompt, you're asking the model to place eight specific objects and it dutifully invents clutter, duplicates, and malformations. The room selector adds only a minimal anchor - "a bedroom with a bed and nightstands" - just enough to set the scene without over-constraining it.

The inputs are three dropdowns plus three toggles:

• style - transitional, traditional, or modern (transitional is the default).
• room - any, living_room, bedroom, kitchen, dining_room, bathroom, hallway. Default any skips the scene anchor entirely.
• level - five realism presets, from "very subtle" to "extreme", defaulting to medium. This one drives the other two outputs, so it's the dial you'll actually turn.

Optional extras: trigger_word (prepend a LoRA trigger token verbatim, if your LoRA was trained with one), include_base (append the shared "warm evening real-estate photography" look - turn it off if your own base prompt already sets the scene), and include_room_anchor (the minimal scene anchor; ignored when room is any).

The three outputs are what you wire into the rest of the graph:

• prompt_fragment (STRING) - the composed prompt. Feed it into a combine node like Superside Combine Prompt's part2.
• lora_scale (FLOAT) - the weight your style LoRA should run at for that level.
• strength (FLOAT) - the denoise/inpaint strength to match.

The vocabulary it uses is maintained in modules/architectural_styles_glossary.txt in the repo, split into two layers: Layer A is the general descriptors the node actually uses, and Layer B is a detailed per-room catalog kept as reference for training captions - explicitly never to be dumped into a single inference prompt. That's a genuinely good habit to steal for your own LoRA training, and it's why this node produces cleaner rooms than whatever you were typing before.

Install - it's part of the Superside pack, so ComfyUI Manager (search "comfyui-superside-nodes") or:

cd ComfyUI/custom_nodes
git clone https://github.com/Superside/comfyui-superside-nodes
pip install -r requirements.txt

Restart ComfyUI, look under the Superside category. No key, no model downloads, no new dependencies beyond what the pack already needs. If your output still has weird artifacts after wiring this in, the usual suspects apply: your LoRA scale is too high for the level you picked, or you left include_room_anchor on while also prompting a specific room layout yourself - double anchoring is how you get the model to fight you.