Translator 🐑

There's no dedicated translation API or model bolted onto this node - it just asks whichever LLM you've connected to translate the text for you. That's the whole design, and it's the right call in 2026: modern chat models are genuinely good multilingual translators, good enough that a purpose-built translation service is no longer the obvious default for a workflow that already has an LLM connector sitting around for other jobs like prompt rewriting. If you're already running one of this pack's Set*LLMServiceConnector nodes for anything else, this node is essentially free to add.

Why you'd reach for it, in context. The pack's own example workflow shows the pattern clearly: generate a caption, run it through an LLM to compose a context-aware prompt, then display the result in both English and Chinese at once using a show/log node. This is the node that produces the second language in that pair. If you're writing prompts in your native language and want an English version for a model that expects English, or you're publishing bilingual captions alongside generated content, Translator sits right at that junction - and because it routes through whatever connector you feed it, quality and cost scale with whichever provider you picked, from a fast free-tier SiliconFlow model to a flagship Gemini or GLM.

Inputs that matter. llm_service_connector is required and it's the whole reason this node has no API key field of its own - it takes an LLMServiceConnector from any of the pack's connector nodes (Bailian, Gemini, ZhiPu, Kimi, SiliconFlow, GitHub Models, MiMo, whatever you've set up) and uses that to make the actual call. text is the string to translate, multiline, so paste in anything from a single line to a full paragraph. target_language is a dropdown of ten common languages - zh, en, es, fr, de, ja, ko, ru, it, pt - defaulting to zh (Chinese), which tells you something about who this pack was originally built for.

Output. translated_text - a single string, ready to feed into a prompt field, a Show And Save Anything node for logging, or a StringConcat/StringFormat step if you're assembling a bilingual label.

Install. Search "ComfyUI_MieNodes" in ComfyUI Manager, or cd ComfyUI/custom_nodes && git clone https://github.com/MieMieeeee/ComfyUI-MieNodes and restart ComfyUI. This node itself needs nothing beyond a working connector upstream - no separate translation-service signup, no model download. The dependency is whichever LLM provider you've already configured for the connector feeding it.

Troubleshooting. If the run errors before it even reaches translation, check llm_service_connector is actually wired - this node can't do anything without one, and it's easy to leave the socket empty while you're rearranging a graph. If the translation quality is disappointing, remember the result is only as good as the model behind the connector: a fast, cheap model tuned for short chat replies may handle a one-line caption fine but flatten nuance in a longer paragraph, so if quality matters more than cost for a particular run, swap the connector for a stronger model rather than assuming this node is doing something wrong. And if target_language doesn't cover a language you need, there's no custom override here the way the connector nodes have for models - you're limited to the ten in the dropdown, so for anything outside that list you'd need to instruct the translation via a prompt-generator node instead, feeding your own instruction ("translate this into Thai") through something like BerniniPromptGenerator rather than this dedicated node.