DD Qwen-MT

DD Qwen-MT is a translation node that talks to Alibaba's Qwen-MT cloud service - 92-plus languages, three translation modes, real model names. The name and the model dropdown make it look like a local AI model node, and it isn't: this is a paid API call to Alibaba Cloud's DashScope. No weights to download, no VRAM spent, and no free lunch - you need a DashScope API key and it costs money per request. Once that's clear, it's one of the most complete translation nodes in the ecosystem.

How it works

Under the hood it's the OpenAI SDK pointed at DashScope's OpenAI-compatible endpoint (https://dashscope.aliyuncs.com/compatible-mode/v1), calling the qwen-mt-turbo or qwen-mt-plus model. The API key isn't an input on the node - there's a configuration panel (a small gear UI the pack registers) that saves the key to qwen_mt_config.json inside the extension folder. It validates that the key starts with sk-, which is the DashScope format.

The inputs are the interesting part:

• 文本 (Text) - what you want translated.
• 源语言 (Source Language) - includes an "auto" option plus the full language list (English, simplified/traditional Chinese, Russian, Japanese, Korean, Spanish, French, German, and 80 more).
• 目标语言 (Target Language) - same list, minus auto.
• 翻译模式 (Translation Mode) - 通用翻译 (General), 术语翻译 (Terminology), or 领域翻译 (Domain).
• 模型 (Model) - qwen-mt-turbo (fast/cheap) or qwen-mt-plus (stronger).
• 模式配置 (Mode Config) - mode-specific extra input. For Terminology it wants a JSON dictionary like [{"source": "术语", "target": "terminology"}]; for Domain it wants a prompt like "The text is from IT domain. Pay attention to technical terminologies." General mode ignores it.

The single output, 翻译结果 (Translation Result), is a STRING you can wire into a text display node, a prompt builder, or a T5/CLIP encoder for translated prompts.

When you'd reach for it

When your prompts live in one language and your models understand another, or you're building multilingual workflows (batch-translating a list of prompts, captioning datasets with translations). The terminology and domain modes are genuinely useful for keeping technical terms consistent across a batch - that's the feature that separates this from a bare API call.

Installation

Part of the pack:

cd ComfyUI/custom_nodes
git clone https://github.com/Dontdrunk/ComfyUI-DD-Nodes
cd ComfyUI-DD-Nodes && pip install -r requirements.txt

Restart, or ComfyUI Manager → "ComfyUI-DD-Nodes". The openai package (needed here) is in the pack's requirements. Then get a DashScope API key from Alibaba Cloud and paste it into the config panel.

Where people get burned

Expectation management is the whole game. It won't work at all until you configure a key, and every call is metered - a big batch of prompts can quietly run up a bill. "Auto" source-language detection is handy but can misdetect on short, ambiguous text. And the pack's labels default to Chinese; flip ComfyUI's language setting for English. If you want local translation there are GGUF-based LLM nodes that run offline - this one is for when you want cloud-quality multilingual output wired directly into a workflow.