🔱 Artha Gemini Style

Artha Gemini Style is the 🔱 Artha pack's answer to one of the most boring parts of prompting: describing a visual style in words. Instead of typing "golden hour, film grain, impressionist brushwork" and hoping the model agrees with you, you assemble a style from five categorized dropdowns - or, if you have a reference image, let Gemini look at it and write the style description for you.

It's part of the pack's "Gemini as your prompt engineer" layer. Style, Subject, and friends produce structured outputs that feed into other Artha nodes, so you can compose a full image or video prompt without juggling forty comma-separated fragments. If you've ever envied people who write gorgeous long-form prompts, this is the closest thing to outsourcing it.

How it works

The five combo inputs - traditional, modern, photographic, named, inspired - are each populated from a bundled catalog (compose.json) of style options grouped by category. Pick one from each (or leave them at NONE), and with use_image off the node just packages your selections into a style dict (the ARTHASTYLE custom type) plus a markdown string that renders the whole catalog as a reference cheat-sheet.

Flip use_image on and feed an image in, and it stops being a local picker. The node converts your tensor to a PIL image, sends it to Gemini with a bundled "describe this style" system prompt, and returns Gemini's prose description of the reference's style. Anything non-style gets filtered out, so you get a clean paragraph you can drop into a prompt.

Inputs that matter

• traditional / modern / photographic / named / inspired - the five style axes. All default to NONE; the dropdowns contain real option names from the bundled catalog.
• use_image - the big switch. Off = you pick from dropdowns; on = Gemini reads the image.
• image - optional IMAGE input, used only when use_image is on.
• api_key, model, max_tokens, temperature - the standard Artha Gemini block; model defaults to gemini-2.5-flash.

Outputs are style (ARTHASTYLE) and markdown (STRING). The style output is what wires into the rest of the Artha prompt-builder chain; the markdown is handy for a quick visual reference widget.

Installing it

It ships in the one-pack install, same as all Artha nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Cyrostar/ComfyUI-Artha-Nodes
pip install -r ComfyUI/custom_nodes/ComfyUI-Artha-Nodes/requirements.txt

then restart ComfyUI. ComfyUI Manager users can just search "ComfyUI-Artha-Nodes". The heavy lifting here is google-genai plus the usual torch/numpy/Pillow stack - no model files to download. You do need a Gemini API key from aistudio.google.com, stored in the pack's api.json (gemini_api_key), as a GEMINI_API_KEY env var, or pasted into the node field (which the pack itself warns is visible plain text).

Common issues

• Empty style output in image mode: missing or bad API key again. The node prints the error to the console.
• use_image on but no image wired: nothing to analyze - wire the image input or flip the toggle off.
• Cloud quota: image description calls use the same free-tier metering as everything else; a busy batch of reference images can burn through it.
• The dropdowns look sparse: options come from the pack's own compose.json, so the vocabulary is what Cyrostar curated, not an exhaustive art dictionary. You won't find every niche style - that's what the inspired free-form vibe is for.

The honest take: the use_image mode is the reason this node exists - reference-based style description is genuinely useful when you have a look you can't name. The dropdown-only mode is nice, but it's really a structured alternative to typing, not a magic upgrade.