Sign in

Nodes Browser

ComfyDeploy: How sml-comfyui-prompt-expansion works in ComfyUI?

What is sml-comfyui-prompt-expansion?

Prompt Expansion for Stable Diffusion, using Deepseek API.

How to install it in ComfyDeploy?

Head over to the machine page

  1. Click on the "Create a new machine" button
  2. Select the Edit build steps
  3. Add a new step -> Custom Node
  4. Search for sml-comfyui-prompt-expansion and select it
  5. Close the build step dialig and then click on the "Save" button to rebuild the machine

sml-comfyui-prompt-expansion

Stable Diffusion Prompt Expansion using Deepseek API

Installation

Navigate to where you have installed ComfyUI. For example:

cd ~/dev/ComfyUI/

Go to the custom nodes folder:

cd custom_nodes

Clone this repo

git clone https://github.com/smlbiobot/sml-comfyui-prompt-expansion

Go inside the repo folder

cd sml-comfyui-prompt-expansion

Install the requirements

pip install -r requirements.txt

Copy the example config config.ini.example to config.ini, then edit the config.ini with the actual Repliate API token.

cp config.ini.example config.ini

Start ComfyUI.

Parameters

workflow-0.png

  • system_prompt. The system prompt to use.
  • prompt. The prompt to expand.
  • min_char. Minimum number of characters to generate.
  • max_char. Maximum number of characters to generate.

Setting min 1000, max 3000 will generate between 1000 and 3000 characters. This prompt length is generally good for models like Flux and SDXL. If you’re working with SD 1.5, you may wish to set this to a lower value.

This prompt expansion node is not intended for SD 1.5, since SD 1.5 uses a different style of prompt.

Methodology

The DeepSeek engine is used to generate the expanded prompt. It has a caching mechanism that would output the same result given the same user prompt. To ensure that the same prompt will in fact output different results, I’ve employed a few tactics through salting:

  • Character count. By using a min and max character count, it asks the LLM to output a prompt with a character length that’s different every time. Thus you should set a range for this.
  • If it detects that you are using the same prompt as last time, it will fetch a few random sentences from the previous output and put that into the prompt you typed. This is mainly to avoid repetition by forcing the LLM to output different results.

Examples

workflow-1.png

This shows a very simple prompt expanded into a complex paragraph.

workflow-flux-pro-ultra.png

A real example for Flux Pro Ultra output.

Suggestions

Please open an issue if you have any suggestions or questions.