cnnmmd

background

◯

the purpose

The purpose of this tool is to enable users to create a system that allows them to easily converse with their favorite characters. [※1]

◯

background

Character generators and image generation/speech synthesis models make it easy to create characters, while language generation and sentiment analysis models allow you to have natural conversations with your characters.

In addition, the language generation model also generates network/server settings and code (programs) in various languages that connect characters and generative models.

◯

problem

However, applying these codes in the real world can be quite a challenge -- the real world environment is not as clean as language generation models assume.

Furthermore, to be able to reuse that code, you need a certain amount of knowledge -- while there is a lot of overlap in the codes for different character types, the codes are often subtly different, making it difficult to reuse them.

To begin with, there is the problem that many people are reluctant to deal with problems that come with installing an app or to write code in general.

◯

correspondence

This tool [1] uses containers to solve the problem of conflicts between apps. And [2] connects containers via a network, allowing the configuration to be changed flexibly. [※2]

In addition, [3] all tools can be customized, and they can be created and published as [4] plugins. [※3]

Conversations and the associated triggers/actions can be created as visual workflows [5], but also written as handwritten code [6] to implement complex flows (sequential, branching, and iterative).

All of these plugins, including containers and workflows [7], can be created, deleted, started, and stopped in the same way.

*1

There are many so-called AI chat apps and services, and some of them support character customization and replacement -- but what users really want is to talk with the face and voice of their favorite character. At least for personal use, it is possible to create the appearance and voice of your choice by using generative models -- so the remaining problem is to introduce a mechanism for conversation.

※2

This includes not only the flow configuration, but also the server deployment and functions called from each node -- the deployment of each server corresponding to a node can be local or in the cloud, and the server functions can be CPU/GPU or API calls.

*3

If you generate code for this environment from a language generation model, the environment itself is clean because it is separated by a container. You only need to generate individual code that is not shared, so the effort is minimal.