🔃 Translate

Translate does one thing: it runs a chunk of text through Google Translate and outputs the result as a string. Most diffusion models were trained overwhelmingly on English captions, so if you think and write in another language, your prompts land better in English - and this is the node that gets you there without alt-tabbing to a translation tab and pasting back and forth.

There's not much mechanism to explain. text is your input (multiline, so full prompts are fine), translate is a picker of 108 target languages, and the STRING output is the translated text. Wire that output into a CLIP Text Encode and you've got a prompt in the model's best language while you kept writing in yours. That's the whole node.

Where it's genuinely handy: the pack builds this translate capability into a lot of its bigger nodes already - the CLIP encoder, Auto Generate, the chatbot all have a translate option baked in - but having it as a standalone node matters when you want to translate something that isn't a prompt. A caption you're about to save, text pulled from an image's metadata, a string headed into a filename or an overlay, a value you want readable in two places. Anywhere a piece of text needs to change language mid-graph, this is the drop-in.

A realistic take on quality, because it's worth setting expectations: Google Translate is fine for the gist and perfectly good for straightforward prompt text, but it's not flawless on idiom or on the compressed, comma-salad style prompts sometimes take. For prompt work that's usually not a problem - diffusion models aren't parsing grammar, they're keying off concepts, and "old wooden house, morning fog, cinematic" survives translation cleanly. If you're translating something where nuance actually matters, treat the output as a draft rather than gospel. Some people in the wider community also prefer DeepL over Google for certain languages; this node uses Google, so if a translation reads oddly, that's the engine, not a bug.

Install comes with the pack. ComfyUI Manager, search SDVN_Comfy_node; or cd ComfyUI/custom_nodes && git clone https://github.com/StableDiffusionVN/SDVN_Comfy_node, then pip install -r custom_nodes/SDVN_Comfy_node/requirements.txt from your ComfyUI root and restart. No weights - it's a text utility.

Troubleshooting is short. Because it calls Google Translate, it needs network access: if it errors or returns your text unchanged, the first suspect is connectivity (an offline or firewalled machine can't reach the service). A very long block of text can occasionally get truncated or rate-limited by the service, so break huge inputs up if you hit that. And make sure the translate picker is set to the language you actually want as the target - leaving it on None passes the text through untranslated, which is the most common "it's not doing anything" cause. Otherwise it's about as reliable as a one-input, one-output node gets.