🔱 Gemini Translate

Gemini Translate is the pack's translation node: you give it text, pick a source and target language from two dropdowns, and it returns the translation as a string you can drop anywhere in your workflow. It's the pack's most international-feeling node, and its real use case is one you'll hit more often than you'd expect: models and prompt libraries don't speak your language, and neither does your training data.

It's part of the Artha-Gemini pack (the 🔱 Gemini nodes), the family of cloud-LLM nodes that outsource their thinking to Google's Gemini API. Translate is text-only - no image, no audio, no GPU load - and it always makes a live API call.

How it works

Straightforward. You provide text, choose lang_from and lang_to, and the node builds a system instruction that reads "Translate the given text from [lang_from] to [lang_to]. Begin your output directly without any introductory sentence or summary phrase." That last clause is the important bit: it stops Gemini from adding "Here is your translation:" and other preamble, so you get clean text out - ready to feed into a CLIP encoder or a text file without post-processing.

The language lists are fixed dropdowns covering the world's major languages: English, Chinese, Spanish, Hindi, Portuguese, Russian, Japanese, Turkish, German, French, Italian, and Bengali. Defaults are Chinese → English, which hints at where the author's user base skews, but every pair is one click.

Inputs and outputs

• text - the source text.
• lang_from / lang_to - the two dropdowns.
• api_key, model, max_tokens, temperature - the shared API block, gemini-2.5-flash by default.

Output: response (STRING), the translation. Wire it into a text encoder, a Save Text node, or the pack's Gemini Response display.

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; dependencies are google-genai, Pillow, numpy, torch, and soundfile. You need a Gemini API key from Google AI Studio.

Where people get burned

The main trap is translation drift on prompt-specific text. This node is a general translator, not a prompt translator, so technical terms, artist names, and prompt vocabulary can come back literally rather than idiomatically - "photorealistic, 8k, intricate details" can flatten into something generic in the target language. If you're translating a prompt you care about, check the output; the KB's prompt-engineering notes on how much exact wording matters to modern models make this a genuine concern, not pedantry.

Second, the dropdowns are the interface, and it's easy to set them backwards - there's no auto-detect, so if lang_from is wrong, you get garbage that looks perfectly grammatical. Keep the pair straight.

And the pack's usual gotchas: keep the key in api.json (gemini_api_key) or the GEMINI_API_KEY env var rather than the plain-text workflow field; remember every run is a billable API call; and a failed call returns a response starting with Error: with the real message in the ComfyUI console. For a node this simple, that's about all that can go wrong - and most of it is the API key.