Text Translate Manual (All langs) [LP]

Translation matters more in ComfyUI than it looks like on paper. A growing number of text encoders - Qwen3-based ones especially - read prompts natively in Chinese, and non-English prompters generally get better adherence writing (or translating into) the language a model was actually trained closest to. This node is the pack's blunt tool for that: pick a source and target language from a dropdown, hit run, get translated text back, no API key required.

What it does

The README lists "Google Translate" plainly among the pack's "Other nodes," and this is one of two nodes implementing it - the "All langs" version, which exposes the full raw ISO-639-1-style language code list (109 source codes, 108 target codes: af, am, ar, az... and on). The translator choice between GoogleTranslator and MyMemoryTranslator matches the class names from the deep-translator Python package almost exactly, so that's very likely what's powering this under the hood - both are free, keyless translation backends that don't require you to sign up for anything, which is exactly why a node like this can exist without asking you for credentials.

"Manual" in the name refers to how you pick the languages, not how the translation happens - you set source_lang/target_lang from a fixed list by hand rather than the node inferring them automatically. If you don't speak the target language and don't know its two-letter code offhand, this "All langs" version asks you to know or look one up; its sibling TextTranslateManual-LP trades the full code list for a smaller, human-readable set of spelled-out language names, which is the friendlier starting point if you're only working with a handful of common languages.

Inputs and outputs that matter

• text (required, multiline, defaults to "text") - what you want translated.
• source_lang (required, dropdown, default auto) - the language text is written in; auto lets the backend detect it.
• target_lang (required, dropdown, default en) - the language to translate into.
• translator (required, dropdown: GoogleTranslator / MyMemoryTranslator, default GoogleTranslator) - which free backend does the work.

Output: text - the translated string.

Installing it

Through ComfyUI Manager, search "ComfyUI Level Pixel" or "ComfyUI-LevelPixel." Manual route:

cd ComfyUI/custom_nodes
git clone https://github.com/LevelPixel/ComfyUI-LevelPixel.git

Restart ComfyUI. This node needs internet access at run time - it's calling out to a hosted translation service, not running anything locally, so it won't work offline or fully airgapped.

Common issues & troubleshooting

Translation fails or times out with no clear reason. Since this hits a live web service rather than a local model, network issues, rate limiting, or the free backend being temporarily unhappy are the realistic failure modes - not a bug in the node itself. Switching the translator dropdown from GoogleTranslator to MyMemoryTranslator (or back) is a cheap first thing to try if one backend is acting up.

MyMemory stops translating after a while in the same session. Free, keyless translation APIs like MyMemory typically cap out on daily volume for anonymous requests - if you're translating a lot of text in one workflow run, hitting that ceiling and falling back to GoogleTranslator is a reasonable workaround rather than something to debug further.

Translation quality feels off, especially for non-Western-language prompts feeding a text encoder. This is a known soft spot for Google Translate specifically in image-gen prompting circles - community discussion around Qwen-based encoders (which read Chinese natively) has pointed to DeepL producing noticeably better results than Google Translate for that particular conversion. This node doesn't offer DeepL as an option, so if translation quality into the encoder's native language matters more than convenience, a manual pass through a better translator before pasting the result in may be worth the extra step.

Don't know the language code you need. With 100+ raw codes on offer here, this "All langs" node assumes you already know (or will look up) the code you want. If you're only translating between a handful of common, well-known languages, TextTranslateManual-LP's spelled-out language list is less error-prone.