Your own chatbot and image generating AI service with the easiest possible setup: a website available from everywhere you need, and a computer that handles processing the prompts. Poor Man's AI features:
- Easy to install Website for interacting with your models
- Easy to configure AI Processors with support for many
- Chat popup on your website
- Enhancing chatbots with your documents
- Active swapping between models based on your rules
This is how an example installation looks like, you'll have something similar after a very easy installation process.
The home page hosts the AI selector, currently chat and image generation. Users can register, and registration or full access can be opened on demand. Admins can change the server from high quality chat-only mode to image generation and a bit downsized chat on the side.
The chat is split to multiple models, called by friendly names, and switching between them is instant. You can even swap "personalities" mid-conversation.
Image generation is fast and easy. Models are swapped based on prompt (see Poor Man's AI's MoA technology), resulting in better prompt following than standalone models.
Poor Man's AI has a few tricks up its sleeve either first introduced in this software or rarely found in the wild.
The image generation part is based on allowing model swapping on demand. Prompts can contain specific keywords or selectors that guide the image generator to specific models, like the "photo" keyword could trigger the use of a model trained on photos only to produce perfectly photorealistic results. When using Stable Diffusion 1.5 models, this could mean that even terabytes of models can result in a maximum VRAM usage of 6 GB with very fast generation speeds that include model loading.
The knowledge of the chatbot doesn't have to be present in the model: you can take your text files and increase its knowledge with factual information that will be loaded when a corresponding prompt is asked.
Poor Man's AI is split into two parts: the Website and the Processor. They can be downloaded, handled, and updated separately. The Website runs on any web hosting service, including free ones. Users can prompt the AIs there. You can handle who has access, you could close it for yourself or open it to anyone, or just for friends. The Processor is the computer that handles your prompts and replies with chat messages or generated images. It needs to be a mildly powerful (mid-range gamer) PC that runs this software. Everything is nearly plug and play, the default settings already provide a pretty powerful setup, you only need to configure some accounts.
Poor Man's AI requires the following:
- SSD storage: 40 GB - HDDs don't provide sufficient performance
- GPU memory: 6 GB
- System memory: 8 GB
The downloadable Website is a zip file containing what you will eventually need to upload to your chosen web hosting provider. But first, you have to prepare a website. This tutorial is for the easiest method using a web hosting provider and not for complete self hosting.
- Get a web hosting account with PHP and MySQL, free ones are fine. If you want to run completely locally, get XAMPP as a web hosting software.
- Download and unzip the Website.
- Upload every folder and PHP file from the unzipped download to the root of your web hosting server.
- Go to /install.php, fill in the login data for your MySQL server that came with your web hosting account, provide an admin username and password, then click "Install". If it's successful, it will redirect you to the installed website, that is currently offline. Set up the Processor to make it online.
The Processor requires the Website already set up, as it only works while connected to one.
- Download and unzip the Processor on the PC you want to run it on.
- Go to the Configuration folder, and choose one file that matches your GPU, or is a bit lower in VRAM. Open that file with Notepad. What each setting does is written in the file. When you're done, don't forget to save the changes to this file.
- Start the Processor application (PoorMansAI.exe). At first launch, it might
take an hour to download and extract every dependency. This only happens at
the first launch, subsequent startups are fast. When you see
[INFO] Ready in ... mode., you're ready to use your own AI service through your own website.
Note: extraction of Stable Diffusion WebUI can slow down over 99%. This is normal, and large files are being extracted. The full extraction should not take longer than 10 minutes.
- Website: Delete sql.php from the server and navigate to install.php again.
- Processor: Every model and setting is stored in the unzipped folder that might have broken your installation. Just delete and redownload or unzip again to fix issues caused by misconfiguration or half-done downloads. Make sure to backup your settings.
Special use cases like adding a chat popup to your website, giving extra knowledge to the chat, design settings, conversation starters, and addon development are documented in this file.
By downloading the software and/or its source code, you are accepting these terms. The source code, just like the compiled software, is given to you for free, but without any warranty. It is not guaranteed to work, and the developer is not responsible for any damages from the use of the software. You are allowed to make any modifications, and release them for free under this licence. If you release a modified version, you have to link this repository as its source. You are not allowed to sell any part of the original or the modified version. You are also not allowed to show advertisements in the modified software. If you include these code or any part of the original version in any other project, these terms still apply.