🔱 Gemini Style

Naming an art style in a prompt is usually where people get vague - "cinematic, detailed" means nothing to a diffusion model and everyone knows it. Gemini Style turns that into a menu. One dropdown for traditional art movements (Impressionism, Cubism, Art Deco, Pop Art…), one for modern looks (Cyberpunk, Vaporwave, Anime, Pixel Art…), one for photographic treatments (Cinematic, Polaroid, Black and White, HDR…), one for named studio styles (Ghibli, Pixar, Tim Burton, Elden Ring…), and one for artist-inspired styles (Van Gogh, Klimt, Hokusai, Warhol…). Pick one per category and you've got a style that a text encoder can actually chew on.

It's from the 🔱 Artha-Gemini pack (Artha/LLM/GEMINI category), and it sits on the composition side of the family. Its style output feeds Gemini Compose, the pack's scene/video counterpart to Gemini Portrait. Style, Camera and Subject all plug into Compose; Face, Body, Form, Makeup and Cloth plug into Portrait. If you're building a still image with a character in it, you'd typically want both halves.

How it works

Same two-mode pattern as the rest of the pack. Leave use_image off and no API call happens: the node collects your selections into a dict, labels each with a STYLE_… key, and pushes it out the style output. Turn use_image on, pick an image, and it sends the photo to the Gemini Vision API with a short agent prompt - "analyze this image and determine the art style" - and returns what it thinks the style is. The agent file is deliberately minimal, just the role and task; Gemini does the rest.

One extra output worth knowing about: markdown. Both Gemini Style and Gemini Camera emit a second STRING that is the entire option list rendered as a markdown reference document - every category, every choice, with its stored description. The pack ships a display node (Gemini Markdown) that shows it on the canvas, so you can scroll the full catalog without tabbing back to the GitHub page. It's genuinely handy given how many choices there are (14 traditional, 19 inspired).

The three fields you'll actually touch

• inspired - the artist-styles dropdown, usually the highest leverage for a distinctive look.
• photographic - "Cinematic" here beats three paragraphs of lighting adjectives for a still.
• traditional / modern / named - pick the one that matches your reference, leave the rest NONE.

Everything after that is the pack's standard cluster: api_key, model (gemini-2.5-flash default), max_tokens, temperature, use_image and the image file dropdown.

Install & gotchas

Same pack as the others - install once and every 🔱 Gemini node appears:

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

Restart ComfyUI after. ComfyUI Manager can also grab it by searching "Artha-Gemini". The only heavy requirement is a Gemini API key from Google AI Studio, and that's only for use_image mode - the dropdown mode never touches the network.

Watch the image input: it's a file picker, not an IMAGE socket. Wire a Load Image node into it and nothing happens; select or upload a file instead. And remember image mode is a billed API call every run, so only flip use_image when you actually want style extraction. One small quirk from reading the source: when the manual dict is built, one of the keys has a trailing space in it (STYLE_TRADITIONAL ). Harmless - Gemini Compose treats these as opaque bags of text - but if you ever inspect the dict and see a weird key, that's what you're looking at.