ComfyUI Prompt Manager

A comprehensive prompt and recipe toolkit for ComfyUI

ComfyUI Prompt Manager is a prompt toolkit for ComfyUI. It helps you manage, save, and reuse prompts along with their LoRAs and trigger words. It can generate prompts using a local LLM through llama.cpp or Ollama, or generate them directly with ComfyUI’s own CLIP/text encoders. It also extracts metadata from images, videos, and JSON files, so you can turn them into reusable prompts — and it includes a prompt browser and save/load tools to keep everything organized.

See Installation Guide

See Node Reference

See latest update

What This Provides

• Prompt Manager:
  • For saving and loading Prompts.
  • These saved prompts can include LoRA stacks, trigger words, as well as thumbnails.
• Prompt Generator:
  • To run a local LLM (llama.cpp or Ollama), or directly use ComfyUI's CLIP/text encoders.
  • Also supports image analysis when using Vision Capable models.
• Prompt Browser:
  • For viewing or editing all data types used by the add-on.
  • This node is how one modifies System Prompts used by Prompt Generator.
• Prompt Extractor:
  • Read metadata from images/videos/JSON and output the prompt and LoRAs they use.
• Lora Preview integration:
  • When Lora-Manager is installed, LoRAs can be previewed on hover.
• Advanced media loading:
  • Extractor nodes can also act as advanced image loaders, but I'd recommend FBNodes for better solutions.
• And more:
  • See the Reference guide for a complete list.

---

Prompt Creation With Lora and Trigger Words

Generating Prompts based on Images

Advanced Prompt Generation using Multiple Images

Recipe extraction and modification (Adding a Style LoRA)

Easily find saved Prompts & Recipes using the Built-in File Browser

Toolset Overview

Prompt Toolset

• Prompt Manager: Supports Prompts, LoRA stacks, trigger words and thumbnail for workflows.
• Prompt Manager (Basic): Simple no-frills basic version (The OG).
• Prompt Generator: prompt creation and enhancement using llama.cpp, Ollama, or a connected ComfyUI CLIP/text encoder.
• Prompt Extractor: Reads metadata from images/videos/JSON and outputs prompt + LoRA + recipe context.
• Prompt Compose: A tool to apply re-usable prompts to your prompts, be it expressions, actions, etc.
• Prompt Browser: A node meant to allow access and write prompts for:
  • System Prompts used by Prompt Generator.
  • Compose Data used by the Prompt Compose.
  • Prompt Data used by our Prompt Manager.

Recipe Toolset (Experimental)

These tools were an experiment, with some of the tech now used in other parts of the add-on.

• Recipe Extractor: normalizes extracted metadata into reusable recipe_data.
• Recipe Builder: edit and validate recipe_data in a cleaner authoring surface.
• Recipe Renderer: execute recipe_data through built-in generation templates.
• Recipe Hub: merge/append model blocks into one recipe_data and expose each model block as its own output.
• Recipe Relay: edit one model block (model_data) with seed and LoRA stack overrides.
• Recipe Model Loader: resolve and load models from recipe_data.
• Recipe Manager: save and reuse recipe entries.

Workflow Examples

Workflow examples are provided to help understand the basics.

Preferences And Settings

• Addon settings are available in ComfyUI Preferences (Settings) under Prompt Manager.
• This is where you set model/backend defaults, NSFW visibility defaults, view preferences, and more.
• Prompt Generator backend choices (llama.cpp, Ollama) and related options are configured there.
• The Prompt Generator's system prompts are fully user-customizable. They're stored in user/default/prompt_generator_data.json and can be created and edited with the Prompt Browser — your custom prompts then show up directly in the Prompt Generator node. (Press r for refresh if not visible)

Documentation

• Detailed node reference: docs/feature-reference.md

Latest update

• First draft of a new Prompt Compose system is now included.
• This first draft is functional and usable today, though the provided default prompts still need refinement.
• Prompt Compose replaces the Expression Selector node and expands on it significantly: it still handles expression-style prompt appending, but also adds a broader prompt composition workflow with category/prompt management and editing tools. Prompt Compose also supports random generation, letting you select multiple prompts or randomly select one.
• New Prompt Compose Manager is provided to add new prompts and categories.

Use Prompt Compose to generate Prompts from preset fragments

Prompt Browser now allows for direct Editing

The new Prompt Browser lets you create and edit all three prompt libraries from a single interface:

• System prompts: the instructions that steer the Prompt Generator's LLM
• Compose prompts: Prompt Composer fragments
• Prompt Manager prompts: your saved user prompts

System prompts now feed straight into the Prompt Generator, so you can author your own and pick them right in the node. Sample system prompts ship in the node's /prompts folder to get you started. If new prompts are added and you don't have them, you can use the Prompt Browser's Import JSON feature to add them.

Installation

1. Navigate to your ComfyUI custom nodes directory:

   cd ComfyUI/custom_nodes/
   

2. Clone this repository:

   git clone https://github.com/FranckyB/ComfyUI-Prompt-Manager.git

3. Install dependencies:

   cd ComfyUI-Prompt-Manager
   pip install -r requirements.txt

4. Install llama.cpp:
   • Windows:

     winget install llama.cpp

   • Linux:

     brew install llama.cpp

5. Place custom .gguf models in models/gguf.
   • Or use preferences to set a custom path.
6. Restart ComfyUI.

Using the CLIP/text encoder backend: No additional setup is needed. Connect any ComfyUI CLIP output to the Prompt Generator node's clip input. When clip is connected, the node bypasses llama.cpp/Ollama and generates prompts directly through the text encoder.

Data Location All data generated by the nodes is saved under user/default folder. The data is saved in JSON format and the thumbnail data is also included, saved as base64 data URL. The node creates a backup folder that keeps 5 daily copies in the hope of preventing any mishap. As the node supports exporting the JSON, be safe and make backups from time to time.

Requirements

• huggingface_hub
• llama-server, Ollama or Comfy's text encoders (Much Slower)