A Swiss Army knife for ComfyUI prompts

What it does

ComfyUI Prompt Assistant embeds a floating utility panel into ComfyUI text nodes, letting you call LLMs for translation, prompt expansion, and image-to-text captioning without leaving the canvas. It supports cloud providers like Zhipu, Gemini, and Silicon Flow, as well as local Ollama instances, and adds tag presets, prompt history with undo/redo, and Markdown translation to the mix.

The interesting bit

Instead of treating language models as external tools, the plugin mounts context-aware assistant buttons directly onto nodes—even inside subgraphs—and intelligently routes Ollama requests through either native or OpenAI-compatible APIs depending on the URL. It also caches translations to keep repeated edits from drifting away from the original meaning.

Key highlights

• Multi-backend support: cloud APIs (Zhipu, Silicon Flow, Gemini, Baidu, 302.AI, custom OpenAI-compatible) plus local Ollama with smart API routing.
• Reverse-prompts images and videos (beta) into text, with styles ranging from natural language to comma-separated tags.
• CSV-based tag presets with favorites, custom categories, and one-click insertion into any text field.
• Sentence-level prompt history with undo/redo across nodes, plus translation caching to prevent semantic drift.
• Rebuilt for ComfyUI V3 API and Node 2.0 (Vue) frontend; UI supports multiple languages.

Caveats

• Video reverse prompting and node document translation are explicitly marked beta.
• Requires ComfyUI 0.3.27 or newer; V2.x targets the ComfyUI V3 API.
• Cloud features need API keys, though the plugin keeps them server-side rather than exposing them in the frontend.

Verdict

Grab this if you spend more time wrestling with prompt wording than tuning denoising strength. Skip it if your prompt workflow is already locked down and never needs translation, expansion, or reverse captioning.