Local Ollama Translator for ComfyUI

A local-first English ↔ Simplified Chinese translator node for ComfyUI.

This Civitai upload is a demo workflow. The actual custom node is installed from GitHub:

https://github.com/InfernusIntraMe/ComfyUI-Local-Ollama-Translator

The node itself is model-agnostic. It can be used with Krea 2, Flux, Z-Image, Qwen Image, SDXL, or any workflow that accepts a STRING prompt input. Krea 2 may be used in showcase images, but it is not required for the translator node.

A local-first English ↔ Simplified Chinese translation workflow for ComfyUI.

This workflow uses a custom ComfyUI node that sends prompt text to a locally running Ollama model, receives a translation, shows a GUI preview, and outputs the translated text as a STRING for use in ComfyUI workflows.

Useful for:

- English → Simplified Chinese prompt translation

- Chinese → English translation

- Local/private translation inside ComfyUI

- Prompt testing with Qwen, Krea, Z-Image, Flux-style workflows

- Translating image prompts before feeding them into a model text encoder

Privacy:

Default endpoint:

http://127.0.0.1:11434

That means the node talks to Ollama running on your own machine.

It does not call Google Translate, DeepL, ChatGPT, or any external translation API.

Remote endpoints are blocked by default unless allow_remote_endpoint is manually enabled.

Requirements:

- ComfyUI

- Ollama installed and running

- qwen2.5:7b or another multilingual Ollama model

- The ComfyUI-Local-Ollama-Translator custom node

Install the recommended Ollama model:

ollama pull qwen2.5:7b

Install the custom node:

cd /d ComfyUI\custom_nodes

git clone https://github.com/InfernusIntraMe/ComfyUI-Local-Ollama-Translator.git

Restart ComfyUI.

How to use it:

For translation only, use the node by itself and queue the workflow.

For image workflows:

Local Ollama Translator: translated_text

↓

CLIP Text Encode: text

Your workflow still needs its normal model text encoder:

Load CLIP / text encoder

↓

CLIP Text Encode: clip

This node translates prompt text. It does not replace your model’s text encoder.

Recommended settings for image prompts:

direction: English → Simplified Chinese

style: Image Prompt

model: qwen2.5:7b

endpoint: http://127.0.0.1:11434

keep_alive: 0

allow_remote_endpoint: false

debug_logging: false

Notes:

The included workflow is only a demo/test workflow.

It does not include Ollama, Qwen, or any image model files. You must install Ollama and pull the model yourself.

If you use a different Ollama model name, change the model field in the node.

The workflow uses a local endpoint by default and is intended for privacy-first local use.