Inspiration

I love automation, I love collecting datasets. I love comfyUI. I wanted a way to easily build new datasets for Lora training. Kontext is the perfect tool for copying styles and it does it very fast. I feel that with most AI models, they are not very random. I needed a better way to create random prompts.

I was also inspired by your Presets that BFL created a few months ago. I looked into how it works and replicated the functionality into a comfyUI node.

What it does

I have created a set of custom nodes that allow you to create lists of items, these lists can be combined together to either; pick a random set of values from each list or loop through each and every combo of all possibilities - complex chains can be created to generate millions of unique prompts in seconds.

The kontext-presets node will generate a suitable prompt to be used with a vision LLM and a reference image. It also gives examples and is easy to build upon using JSON files. Taking advantage of long context windows in modern LLM's, the system prompt was generated using the BFL official blog post for best practices, this can be modified in the comfyui workflow.

How we built it

Python and JavaScript with help from ChatGPT to generate all of the different options.

Challenges we ran into

Everything is a challenge - I love a challenge!

Accomplishments that we're proud of

I'll be proud if I win a prize.

What we learned

Keep trying - never give up!

What's next for Kontext Prompt Helper Tools

Share the code and Open Source it.

Built With

Share this project:

Updates