🐵 GF Deep Translate

🐵 GF Deep Translate is a single-purpose node that translates text inside your ComfyUI graph using Google Translate's free web endpoint. The name is a bit of a lie - there's no API key, no account, no model download. You feed it a string, it hands back the translated string. That's the entire trick, and it's more useful than it sounds.

Why you'd reach for it

The most common reason: you're not a native English speaker, and your best prompts come out in your own language. Most Stable Diffusion models were trained mostly on English captions, so a Russian or Spanish prompt tends to land noticeably worse than the English version of the same idea. Drop this node in front of a CLIP Text Encode, set the destination to English, and type freely. It's also handy for workflow hygiene - translating node titles, embedded notes, or any text you want to inject into a prompt dynamically.

How it works

Under the hood it's the deep-translator Python library, which scrapes Google's free translate endpoint. That means it is not offline - every translation is a network round trip to Google, and there's no key because you're using the same endpoint a browser uses. Fine for a prompt here and there; don't build a bulk pipeline on it.

The inputs that matter:

• text - the string to translate. It's a multiline box, and the default placeholder is "Введите текст" (Russian for "enter text"), which tells you where the author is from.
• src_lang - source language, defaults to auto so Google detects it.
• dest_lang - target language, defaults to English. There are 135 options to pick from.

Output: translated_text, a plain STRING. Wire it into your CLIP Text Encode's positive or negative prompt, or into any node that accepts a text string.

Installing it

ComfyUI Manager: search "GF_translate" and hit install. Or do it by hand:

cd ComfyUI/custom_nodes
git clone https://github.com/gorillaframeai/GF_translate

then restart ComfyUI. The only dependency is deep-translator, which Manager installs for you (or pip install deep-translator yourself). No models, no VRAM, no GPU - it's one small Python file doing network calls.

Gotchas

The thing that'll bite you: this node doesn't raise errors, it returns them. If a translation call fails - network down, Google throttling you - it quietly hands back Ошибка при переводе: ..., Russian for "error while translating," as the output string. Your workflow keeps running and your prompt silently becomes an error message. If generated images suddenly look confused, glance at what this node is actually emitting.

Also respect the rate limit on that free endpoint. A hundred calls in a row will get you throttled; one prompt per generation is nothing.

One last note, the trust one: this is a tiny MIT-licensed pack from @GorillaFrame - a single Python file, no real community footprint. It does nothing scary (the only network traffic is Google Translate, and there's nothing to download), but it's still "install small third-party packs via Manager and skim the source if you're picky" territory. For translating your prompts into the language your model understands best, this is about the simplest path I know.