What you will learn
- Understand the basic of Stable Diffusion to create new images
- Learn how to use Stable Diffusion parameters to get different results
- Create images using other models provided by the Open Source community
- Learn about Prompt Engineering to choose the best keywords to generate the best images
- How to use negative prompts to indicate what should not appear in the images
- Use fine-tuning to create your custom model to generate your own images
- Send initial images to condition image generation
- Use inpainting to edit images, remove unwanted elements or swap objects
Requirements
Programming logic and Python basics are desirable but not required
It is possible to follow the course without having technological skills
Description
The generation of images using Artificial Intelligence is an area that is gaining a lot of attention, both from technology professionals and people from other areas who want to create their own custom images. The tools used for this purpose are based on advanced and modern techniques from machine learning and computer vision, which can contribute to the creation of new compositions with high graphic quality. It is possible to create new images just by sending a textual description: you ask the AI (artificial intelligence) to create an image exactly as you want! For example, you can send the text “a cat reading a book in space” and the AI will create an image according to that description! This technique has been gaining a lot of attention in recent years and it tends to growth in the next few years.
There are several available tools for this purpose and one of the most used is Stable Diffusion developed by StabilityAI. It is Open Source, has great usability, speed, and is capable of generating high quality images. As it is open source, developers have created many extensions that are capable of generating an infinite variety of images in the most different styles.
In this course you will learn everything you need to know to create new images using Stable Diffusion and Python programming language. See below what you will learn in this course that is divided into six parts:
- Part 1: Stable Diffusion basics: Intuition on how the technology works and how to create the first images. You will also learn about the main parameters to get different results, as well as how to create images with different styles
- Part 2: Prompt Engineering: You will learn how to send the proper texts so the AI understands exactly what you want to generate
- Part 3: Training a custom model: How about putting your own photos in the most different environments? In this section you will learn how to use your own images and generate your avatars
- Part 4: Image to image: In addition to creating images by sending texts, it is also possible to send images as a starting point for the AI to generate the images
- Part 5: Inpainting – exchaning classes: You will learn how to edit images to remove objects or swap them. For example: remove the dog and replace it with a cat
- Part 6: ControlNet: In this section you will implement digital image processing techniques (edge and pose detection) to improve the results
All implementations will be done step by step in Google Colab online with GPU, so you don’t need a powerful computer to get amazing results in a matter of seconds! More than 50 lessons and more than 6 hours of videos!
Who this course is for
- People who want to learn how to create images using Artificial Intelligence
- People who want to create their own avatars
- Beginners in Computer Vision
- Undergraduate and graduate students who are taking courses on Computer Vision, Artificial Intelligence, Digital Image Processing or Computer Graphics
Course Content
Ratings and Reviews
Review posted on Udemy
An excelent course! One of the best I've taken so far, the content and knowledge is 5-stars rating to me, taking this course your are able to make real whatever image you have in your imagination.However some minor details made me consider 4,5. For instance, some videos use a model of stable difussion uploaded by runwayml on huggingface , which is now not available (I had to search it on other repository) and a more natural communication from the teacher would hve been better (sometimes it seemed very structured to what has been written in the presentation or notebook). However, if you are looking for a way to create your own custom models and the different techniques of stable diffusion this is by far the best course! Very enjoyable
Review posted on Udemy
In the truth words, I will recommend who want to understand deeply and clearly about SD aka stable diffusion. (but so deep in math for researching) The course has covered all the things you need to know such as theory about SD, structure of it, etc. More over, he has provided code in colab notebook and clearly description about it before each line. I hope in the near future, there should be a course deeply into math for researcher. Sum up, it's worth. Thank you
Review posted on Udemy
Your course is great, and your accent is fine for me. I think the part that can be improved is the part about stable diffusion basic, which has many technical terms. Although I googled it, I still don't understand it well. I think you can let chatgpt generate more popular text or metaphors to make it easier for learners to understand. Thank you very much for your course.
Review posted on Udemy
The course offered valuable content and reference links, providing a solid introduction to various aspects of Stable Diffusion. While it serves well as a foundational course, it might benefit from diving deeper into advanced topics to become a mastery-level course. I found that significant time was spent by the presenter on reading the content aloud, which I felt could be streamlined/avoided to add more value. Skipping those parts helped me stay focused on the most beneficial sections and saved time.
Review posted on Udemy
# udemy_study_stable_diffusion [Udemy] Master AI Image Generation using Stable Diffusion 저는 한국인이고 아래 리뷰는 한국어로 입력되어 있습니다. 한국인들이 이 강의를 많이 듣기를 바라며 강의 내용에 대한 설명을 추가합니다. 아래 리뷰들을 구글번역기로 번역을 할 경우 최대한 자연스럽게 번역이 될 수 있도록 입력하였습니다. [섹션1] 대본을 크롬 번역기능을 이용해서 보고 있습니다.(만약 수강중이라면 "자막"이 아닌 "대본"을 클릭해보세요.) 이해가 안 되었던 부분은 허깅페이스 공식문서를 찾거나 구글링하면 쉽게 찾을 수 있는 내용입니다. 만약 검색 결과가 만족스럽지 않더라도 일단 스킵하시고 넘어가도 문제는 없어보입니다. [섹션2] Stable Diffusion에 대한 전반적인 설명이 많습니다. 들으면 재밌고 흥미롭습니다만, 만약 수강중이고 이미지 생성하는 것을 먼저 하고 싶다면 "7. Stable Diffusion - limitations of use" 부터 듣고 오는 것이 더 나을 것 같습니다. 이왕이면 "섹션3", "섹션5", "섹션6"에서 여러 프롬프트를 입력해보고 나서 섹션2으로 돌아오는 것이 더 재밌을 수 있어요. [섹션3] 강의 그대로 프롬프트 엔지니어링에 대해 강의하며 제일 흥미를 유발할 수 있는 강의일 것입니다. 만약 프롬프트 엔지니어에 관심이 있다면 이 섹션이 가장 재미있을 것 같습니다. [섹션4] jax 라이브러리를 접해볼 수 있는 기회이므로 이번 기회에 경험해보면 좋을 것 같네요. fine tuning은 강사분께서 최대한 잘 따라할 수 있도록 예제를 정성스럽게 준비하셨습니다. 다만 저처럼 colab이 아닌 wsl에서 작업을 하는 사람에게는 몇가지 팁을 드리려고 합니다. 1. root디렉토리가 아닌 현재 workspace상에 폴더를 만들어주세요. 2. cuda버전을 꼭 확인하고 jax를 설치해주세요. (터미널에 /usr/local/cuda/bin/nvcc --version 를 입력하세요) 3. wget으로 몇가지 스크립트 파일을 만나실 수 있습니다. 그외에 https://github.com/huggingface/diffusers/tree/main/scripts를 참고하시면 여러가지 스크립트를 찾아 보실 수 있습니다. 만약 흥미가 있다면 여러 스크립트들을 탐색하는 것도 재미있을 것입니다. (Foo Fighters의 Dave를 보고 많이 반가웠습니다 특히 달을 배경으로 하는 사진은 "Next Year" M/V가 떠오르네요) [섹션5] image to image을 경험해 볼 수 있습니다. [섹션6] inpainting을 경험해 볼 수 있습니다. 만약 inpainting과 outpainting에 관심히 많다면 Fooocus를 실행해보세요.(https://github.com/lllyasviel/Fooocus) [섹션7] ControlNet를 경험해 볼 수 있습니다. 만약 수강중이라면 같은 강사의 opencv 강의가 유데미에 열려있습니다. 수강하시는 것을 추천드립니다.
Very good course.