✨Prompt Translate

If you think in one language and generate with another, you've felt the pain of trying to write good English prompts from memory when English isn't your thing. ✨Prompt Translate is the pack's answer: a node that auto-detects what language your text is in and translates it to English or Chinese. It's one of the quieter nodes in ✨Prompt Assistant, and honestly one of the most used - every prompt you feed a model has to survive the translation, and prompt syntax is exactly where machine translation chokes.

So the interesting design choice is that this node has two translation paths, and you pick per-use.

How it works

The node sniffs the source language first - it scans for CJK characters and plain ASCII, then decides whether it's Chinese, English, or "auto." If the source is already the target language, it returns your text untouched instead of pointlessly round-tripping it. That's a small thing that saves you from watching a queue slot get wasted on a no-op.

Then it routes to one of two services:

• 百度翻译 (Baidu) - the machine-translation path. Fast, free tier of around 5 million characters a month, but the README is honest that the quality is "一般" (mediocre). Fine for a quick prompt; the pack's own description says it well.
• An LLM service - Zhipu (智谱) free GLM flash models by default, or Ollama, or any configured service. This is the quality path: the pack has dedicated ZH/EN translation system rules that tell the model to preserve prompt syntax, weights, and structure rather than translate them like prose.

Which one you reach for depends on whether you want speed or fidelity. For a prompt you're actually about to generate with, the LLM path wins - machine translation has a habit of mangling tag syntax and (weight:1.3) style markup. For bulk translating node docs or notes, Baidu is plenty and costs nothing.

The inputs that matter

• source_text - the text to translate.
• target_language - English or Chinese, that's the whole dropdown. Defaults to English, which is the direction most people want: write in your native language, generate with English prompts.
• translate_service - 百度翻译 by default, or any 服务名/模型名 LLM option from your configured services.

Output is a single translated_text string, ready to wire into a prompt encode or another node.

The README also mentions a built-in translation cache to avoid re-translating the same text and drifting the wording on re-runs - so repeat executions with unchanged input shouldn't churn the API.

Install

Same pack, same drill. ComfyUI Manager → search Prompt Assistant → install → restart. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/yawiii/ComfyUI-Prompt-Assistant.git

Baidu requires an app ID and secret key in the pack's configuration (apply via Baidu's translation API portal). LLM services need their API key as usual.

Where people get burned

Baidu's free tier is generous but the README warns that using it behind a VPN/proxy can cause requests to fail - the API doesn't like being routed oddly. And remember the content-review gotcha that applies to the whole pack: Zhipu's free models return empty results for content that trips their filters, which can look exactly like a broken translation. It's not. The seed widget works the same as the other nodes here - fixed seed plus unchanged input won't re-fire the call, so toggle randomize when you want a fresh translation pass.