Nodes/ComfyUI Level Pixel/CLIP Text Encode Translate [LP]
ComfyUI Node

CLIP Text Encode Translate [LP]

Prompt in your language, encode in English

By LevelPixel·Created 2 years ago·Updated 5 months ago· 31
CLIP Text Encode Translate [LP]
  • clip
  • CONDITIONING
text

This is a normal CLIP Text Encode node - text and a CLIP model in, CONDITIONING out - with one thing bolted on: it translates your text to English before encoding it. Instead of chaining a separate translate node into a standard CLIPTextEncode, you get both steps in one node.

Why translate before encoding at all

Almost every popular checkpoint - SDXL, Flux, Illustrious and its derivatives - was trained overwhelmingly on English captions and tags. Prompting in another language works to some degree, but it reliably underperforms the same idea expressed in English, because the model has far less training signal tying non-English words to visual concepts. If you naturally think and prompt in your own language, translating on the way in is a small, close-to-free step that avoids leaving quality on the table - and folding it directly into the text-encode step means one fewer node cluttering your graph.

How it works, and one thing worth flagging

This pack also ships a standalone TextTranslate-LP node, and that one exposes a translator dropdown letting you pick between GoogleTranslator and MyMemoryTranslator. This node's own schema doesn't expose that same choice - there's no translator selector here, just text and clip. That likely means it defaults internally to one backend (the README's own framing of the pack's translation feature as "Google Translate" suggests that's the one), but the exact behavior isn't documented, and it's worth knowing you don't get the same fallback flexibility the standalone translate node offers.

The inputs and outputs that matter

  • text - a multiline STRING, your prompt in whatever language you're comfortable writing in.
  • clip - the CLIP model to encode with, same as a standard CLIPTextEncode.
  • Output - CONDITIONING, ready to plug straight into your sampler the same way any CLIPTextEncode output would.

How to install it

ComfyUI Manager: search "ComfyUI Level Pixel" or ComfyUI-LevelPixel, install, restart - auto-updates via Manager's "Update ALL." Manual:

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

Restart ComfyUI. Like this pack's other translation node, this one needs a working translation library under the hood - a normal Manager or git clone install should pull in the pack's dependencies automatically, but if this specific node fails to load while the rest of the pack works fine, that's the first thing to check.

Common issues & troubleshooting

The node errors or hangs. Translation happens over the network on every single run - it needs live outbound internet access. If you're on a cloud GPU box with restricted outbound access (common on some rental platforms), this node won't be able to reach its translation backend.

You wanted to pick a specific translation backend and can't find the option. It's not exposed here. If you need control over which backend is used - say, one is rate-limited and you want to switch - use the standalone TextTranslate-LP node followed by a plain CLIPTextEncode instead of this combined node, since that path gives you the translator dropdown this one doesn't.

You're prompting with Danbooru-style tags rather than natural language, and translation is mangling them. Machine translation is built for natural sentences, not a comma-separated vocabulary of exact tag tokens - translating 1girl, blue_hair, looking_at_viewer risks rephrasing tags the checkpoint expects verbatim. If your prompting style is tag-based rather than descriptive, a plain (non-translating) CLIPTextEncode is usually the safer choice.

CategoryLevelPixel/Text

Inputs (2)

NameTypeDefaultDescription
textSTRING
clipCLIP

Outputs (1)

NameTypeDescription
CONDITIONINGCONDITIONING