Version 1.1.4 expands the original English/Simplified-Chinese translator into a four-node multilingual toolset while preserving the original V1 node for existing workflows.

Nodes

Local Ollama Translator

The original V1 node, preserved for backward compatibility.

English to Simplified Chinese
Simplified Chinese to American English
Auto English/Chinese direction
Natural, Image Prompt, Literal, and UI Text styles

Local Ollama Translator Multilingual

The new main translator.

Auto-detect source language
30 target languages
Separate source and target selectors
Natural, Image Prompt, Literal, and UI Text styles
Preserve-terms field for model names, LoRA triggers, file paths, URLs, brands, and technical tokens
Optional Ollama thinking trace
Automatic one-time refusal retry
Automatic final-only pass if a thinking model spends its entire output budget reasoning
translated_text, source_text, and thinking_text outputs
Resizable Translation Preview and Thinking Preview panels

Local Ollama Translation Prompt Forge

Optional system-prompt control for the multilingual translator.

Profiles:

Hardened Translator
Image Prompt Translator
Lyrics / Poetic
Literal Archivist
Natural Native Voice
Custom

The Forge can preserve structure and intensity, add refusal resistance, and append custom translation rules.

Local Ollama Model Selector

Reads locally installed Ollama models from http://127.0.0.1:11434/api/tags and outputs the selected model name.

Restart ComfyUI after pulling a new Ollama model so the dropdown refreshes.

Privacy

The default endpoint is:

http://127.0.0.1:11434

Remote endpoints are blocked unless allow_remote_endpoint is deliberately enabled. The node does not call Google Translate, DeepL, ChatGPT, or another cloud translation API.

Translation and thinking preview text are session-only. Version 1.1.4 excludes them from both workflow JSON and prompt/API serialization. Preview heights may be saved as harmless numeric layout properties.

Remember that ordinary ComfyUI workflow JSON can still contain the source prompt and normal widget settings. Clear private source text before sharing a workflow.

Requirements

ComfyUI
Ollama installed and running
A multilingual Ollama model installed locally

No additional Python packages are required.

A practical public starting model is:

ollama pull qwen3:8b

Check your installed names with:

ollama list

The translator does not download models automatically. You may use any compatible multilingual Ollama model by entering its installed name or wiring the Model Selector into model_override.

Installation

ComfyUI Manager / Registry

Search for Local Ollama Translator in ComfyUI Manager and install it, then restart ComfyUI.

Git

From your ComfyUI custom_nodes folder:

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

Restart ComfyUI and hard-refresh the browser with Ctrl+F5.

Manual ZIP

Extract the included ComfyUI_Local_Ollama_Translate folder into:

ComfyUI\custom_nodes\

The final path should contain:

ComfyUI\custom_nodes\ComfyUI_Local_Ollama_Translate\__init__.py
ComfyUI\custom_nodes\ComfyUI_Local_Ollama_Translate\web\local_ollama_translate.js

Restart ComfyUI and press Ctrl+F5.

Quick start

The nodes appear under:

Local/Ollama

Recommended wiring:

Local Ollama Model Selector: model
        ↓
Local Ollama Translator Multilingual: model_override

Local Ollama Translation Prompt Forge: system_prompt
        ↓
Local Ollama Translator Multilingual: system_prompt

Local Ollama Translator Multilingual: translated_text
        ↓
CLIP Text Encode: text

The translator only provides text. Your image or video workflow still needs its proper model-specific text encoder connected to the clip input of CLIP Text Encode.

Recommended fast defaults

source_language: Auto Detect
target_language: your target language
style: Image Prompt
system_prompt_mode: Built-In + External Rules
thinking_enabled: false
retry_on_refusal: true
endpoint: http://127.0.0.1:11434
keep_alive: 0
num_ctx: 8192
temperature: 0.0
top_p: 0.7
timeout_seconds: 180
allow_remote_endpoint: false
max_output_tokens: 4096

timeout_seconds is a streamed response-inactivity window, not a hard total-generation timer.

Turn thinking on only when you want to inspect the model's reasoning. Fast mode is the default because translation normally does not need visible reasoning.

System prompt modes

Hardened Built-In

Uses the package's translator system prompt.

Built-In + External Rules

Uses the hardened translator and appends the Prompt Forge or another external STRING input. This is the best general-purpose mode.

External Prompt + Runtime Language Directive

Uses the external prompt as the main instruction while retaining the requested language direction at runtime.

Supported languages

English (American)
Chinese (Simplified)
Chinese (Traditional)
Japanese
Korean
Spanish
French
German
Italian
Portuguese (Brazil)
Russian
Ukrainian
Polish
Dutch
Turkish
Arabic (Modern Standard)
Persian
Hindi
Bengali
Vietnamese
Thai
Indonesian
Malay
Swedish
Danish
Norwegian (Bokmål)
Czech
Greek
Hebrew
Romanian

Preview controls

Drag the bar beneath either preview to resize that panel from 110 to 900 pixels.
Double-click a preview drag bar to reset its height.
In fast mode, a blank Thinking Preview is not created.
The Thinking Preview appears after a thinking-enabled run returns a trace.

Troubleshooting

The model is missing from Model Selector

Run ollama list, confirm Ollama is running, then restart ComfyUI.

Connection refused

Start Ollama and verify the endpoint remains http://127.0.0.1:11434.

A thinking run is slow

Set thinking_enabled to false. The node's default is already fast mode.

The model reasons but never produces a final translation

Version 1.1.4 automatically performs a second final-only pass with thinking disabled. If that pass also reaches the output ceiling, increase max_output_tokens or use a more instruction-efficient model.

Old workflow contains preview text

That workflow was saved before the v1.1.3 privacy fix. Install v1.1.4, reload it, clear any private source text, and save a fresh copy.

Upgrade notes

Existing V1 workflows continue using Local Ollama Translator unchanged.
Existing multilingual workflows keep their saved thinking_enabled value.
Newly added multilingual nodes default to thinking off.
Version 1.1.4 retains the v1.1.2 streamed backend and automatic finalization behavior.

License

MIT