Skip to content

samuel-s-marques/thisdatedoesnotexist-server

Repository files navigation

thisdatedoesnotexist-server

Welcome to the backend documentation of "This Date Does Not Exist", a cutting-edge project that combines Stable Diffusion and LLM to create an AI-driven chat and matchmaking experience.

Installation

You will need Firebase (for authentication), MySQL database, OneSignal, a Text Generation backend (Kobold or Ooobabooga), Comfy UI, and profile-suggester.

Comfy UI and the Text Generation backend are both used to generate characters. To chat with them, it will also be used a LLM.

To install and use this project, you need to follow these instructions:

  1. Clone the project
git clone https://github.com/samuel-s-marques/thisdatedoesnotexist-server.git
  1. Follow profile-suggester setup
  2. Navigate to its directory
cd thisdatedoesnotexist-server
  1. Install dependencies
npm i
  1. Fill the .env-example with values and rename it to .env
  2. Update and seed the database
node ace migration:fresh --seed
  1. Start the server
node ace serve

You need to install and download profile-suggester, otherwise, the characters won't like, dislike nor match the user. The user won't see the character's cards either.

Speech to Text / Audio Messages

The project allows the users to send audio messages to characters. The audios are transcribed using WhisperCPP, locally.

If you want your users to send audio messages in the app, you need to clone WhisperCPP and move it to the project's root. Then, you need to download a model and put the model's name in config/app.ts, in the WhisperConfig section. After that, you need to change the enabled option above the model key.

It is this way because I have no experience or knowledge about using multimodal LLMs.

Image Generation

This project uses ComfyUI to generate character portraits, based on data coming from character-forge - npm.

You need to run ComfyUI and add its URL to the .env file. Also, you need to put your model name in config/app.ts.

The project also uses:

Download them.

Example character

Text Generation / Character responses

This project uses a text generation API through Kobold or Oobabooga to generate character responses.

You need to run one of them and add their URL to the .env file. You also need to define the API name in config/app.ts. Currently, the project uses kobold as the default API.

You'll also need to define your prompt format in the config. The project accepts only some prompt formats, like chatml, openchat, alpaca, mistral, user-assistant-newlines, and metharme. For default, it uses user-assistant-newlines.

The demo video uses kobold and chatml, with NousResearch/Nous-Hermes-2-SOLAR-10.7B as the model.

Support

If you encounter any bugs or issues, please report them in the issue tracker.

Related

Here are some related projects

ThisBotDoesNotExist - procedurally generated characters powered by AI

Authors

License

This project is licensed under the GNU GPLv3 License - see the LICENSE.md file for details.

About

No description, website, or topics provided.

Resources

License

Stars

Watchers

Forks