🔱 Gemini Scenery

Gemini Scenery is the pack's scene-settings node, and it's essentially a curated location database with a prompt generator bolted on. You pick a setting from a dropdown - mountains, a rainy jungle, a neon city street, an enchanted forest - and the node produces a detailed, ready-to-use scene description for an image or video prompt. No more summoning "mossy forest, fog between trunks, god rays" from memory; it's already in the menu.

It's part of the Artha-Gemini pack (the 🔱 Gemini nodes), and it slots into the pack's character/scene assembly pipeline: Subject, Scenery, Camera, Light, and Style each output a typed description, and Gemini Compose merges them into a full composition prompt. Scenery is the one that decides where the whole thing happens.

How it works

The node ships with a JSON library of scene descriptions organized into seven categories, each appearing as a dropdown:

• landscapes - mountains, forests, deserts, beaches, coasts, rivers, lakes, arctic…
• urban - city streets, plazas, and built environments.
• interior - rooms and indoor spaces.
• fantasy, futuristic, abstract - the three "other worlds."
• miscellaneous - everything else.

Each dropdown defaults to NONE, and every selection is a fully-written scene sentence ("Dense tropical jungle with sunlight filtering through thick green canopy, with vines, ferns, parrots"). Pick one per category you care about and leave the rest at NONE.

There's a second mode via the use_image toggle: flip it on, upload a reference image, and the node switches from menu assembly to asking Gemini to describe the scene in the picture - turning a photo into a reusable scene prompt. And the node outputs a markdown reference sheet of every scene option, meant for the pack's Gemini Markdown display node, so you can browse the whole library in the UI.

Inputs and outputs

• The seven category dropdowns.
• use_image - boolean, switches to image analysis mode.
• image - upload widget (only used with use_image).

Shared API block (api_key, model, max_tokens, temperature) plus outputs: scene (ARTHASCENERY, wired into Compose) and markdown (STRING, the reference sheet).

Note: the menu mode still makes an API call - the selected scene sentence goes through Gemini's scene agent, which expands it into the final description. Only the dropdown options are local.

Installing it

Via ComfyUI Manager (search Artha-Gemini) or:

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

No model downloads. You need a Gemini API key from Google AI Studio.

Where people get burned

The classic mistake is treating the dropdowns like a multiple-choice test where every answer matters. They don't - NONE is the right call for anything you don't care about, and stacking one pick from every category produces a contradictory mess ("misty alpine meadow… inside a neon nightclub interior"). The node assembles from all non-NONE selections, so restraint is a feature.

Second, use_image isn't free or local - it's a vision API call, so it costs money and needs a valid key, same as everything else here. And remember the scene output is a description, generated fresh each run, so the exact wording varies between runs even with the same dropdowns. If you need byte-identical prompts, that's what the seed on Gemini Compose is for.

Pack-wide key hygiene applies as always: api.json (gemini_api_key) or the GEMINI_API_KEY env var, never the plain-text workflow field.