AITVwiki 美人百科

AI Painting Tutorials 01: Principles of AI Painting(AI繪畫教程01 : AI繪畫原理)

Added 2023-05-24 14:24:47 +0000 UTC

Introduction:

Currently, there are relatively few systematic learning tutorials available online. In this series of articles, I will share with you the principles and processes of AI drawing based on my own experiences and knowledge. Starting from scratch, I will present the key concepts in a concise manner to help you quickly grasp the art of AI drawing. I hope to provide everyone with the joy of AI drawing. Please note that this tutorial is primarily based on my own learning, and I welcome any corrections if there are any errors.

一、Principles of AI Drawing:

AI drawing is a method that utilizes artificial intelligence techniques to generate visual artworks. It is based on deep learning models, often employing a framework called Generative Adversarial Network (GAN). While this technology has been studied by scholars for a long time, significant progress has been made in the past couple of years, allowing for the generation of images that conform to human aesthetics after receiving specific instructions.

One key difference between AI drawing and human drawing is that the process of human drawing involves sketching, outlining, composition, and coloring, while AI drawing starts by diffusing and adding noise to the original image. Then, based on new logical instructions, an algorithm is applied to reverse the diffusion and remove the noise, resulting in the generated image. The example below illustrates this process:

The left image represents the original image.
The middle image shows the diffusion process with added noise.
The image on the right is the generated image after removing the noise.

For a detailed algorithmic process, please refer to the following image. The text prompts we set and the Lora model play a role in this process of generating images. (We will discuss text prompts and the Lora model in future articles.)

Based on this deep learning algorithm, there are currently various AI drawing products available, such as NovelAI and Midjourney. Additionally, I will introduce Stable Diffusion (referred to as SD). The advantages of SD are that it is currently free and can be downloaded and used on your own computer. There are also many open-source Lora models available for drawing purposes. However, SD has higher requirements for graphics cards. It is recommended to prioritize Nvidia graphics cards when choosing computer equipment, preferably from the 3060 series or above. Otherwise, generating each image will be time-consuming.

The main software used in SD at present is developed by a brilliant individual known as AUTOMATIC1111 in the GitHub open-source community.

The current mainstream SD interfaces are based on his code. This is what people commonly refer to as SD WEB UI.

Below is the download link for SD WEB UI:

https://github.com/AUTOMATIC1111/stable-diffusion-webui

Regarding how to perform local installation, there are many detailed tutorials available on YouTube. It is difficult to explain through written instructions, so I suggest following their video tutorials directly. If you do not want to purchase a new graphics card, you can also use Google Colab for installation, but the free version has limitations on usage. However, it is a great option if you just want to try AI drawing.

In the next article, we will introduce the features and functions of the SD WEB UI interface. If you find this article helpful, please give it a thumbs up to encourage me to continue writing.

前言

目前網路上比較少系統性的學習教程

這系列文章我會依照我自己目前所學到的經驗和知識

將一些AI繪畫的原理跟製作過程分享給大家

從零毫無基礎開始

以重點節略的方式讓大家能快速上手學會AI繪畫

希望能讓大家一起體會AI繪畫的樂趣

當然本教程目前還是主要依據我學習到的知識

如果有錯誤請見諒並歡迎指正

一、AI繪畫的原理

AI繪畫是一種利用人工智能技術來生成視覺藝術作品的方法。

它的原理基於深度學習模型，通常使用稱為生成對抗網絡（GAN）的架構。