First look at a few generated works
Realistic wind
two-dimensional
2.5d
Mecha wind
What is AI painting?
AI painting, also known as "ai painting", "artificial intelligence painting", is to obtain paintings or pictures through AI generation technology. AI painting has a long history, and there are many ways to create AI paintings, including rule-based image generation algorithms and deep learning algorithms. The most popular thing on the Internet recently is the form of generating paintings through text descriptions. You may wonder, it seems that AI generated pictures have been heard of for a long time, why is it so popular this time?
You may not think that the earliest AI painting system AARON was developed by Harold Cohen in the late 1960s. AARON uses a method based on symbolic rules to generate images. Unlike the current AI painting that outputs digital images, AARON really uses brushes and paints to paint.
As for the effect, that's it. . .
Although AI painting appeared a long time ago, the effect at that time was not satisfactory.
After continuous iterative updates, until midjourney was opened for public beta in July 2022, someone won an art award through the works generated by midjourney, which aroused widespread discussion.
The picture below is the work "Space Opera House", which was made by Midjourney and caused a lot of controversy.
Now mainstream AI painting
Go to search, there are already many AI painting tools, dazzling, how should we choose?
This is the ranking of various mainstream AI paintings I found on the Internet, but AI is updated very quickly, this is just for reference.
Some people also said that SD is the first
What is Stable Diffusion
In August 2022, Stability.ai released and open-sourced the Stable Diffusion AI painting model, which completely exploded the AI painting boom!
In October 2022, NovelAI, a two-dimensional AI painting model based on Stable Diffusion, was born. NovelAI's powerful two-dimensional illustration drawing ability has made it the world's best animation illustration generation model.
If midjourney is the beginning of AI painting, then Stable Diffusion should be the explosion of AI painting
According to public information, Stable Diffusion was proposed by StabilityAI in 2022, and both papers and codes have been open sourced. StabilityAI completed a financing of US$101 million on October 28, 22, and its current valuation has exceeded US$1 billion.
You can go to Stable Diffusion Online to experience Stable Diffusion https://stablediffusionweb.com/
For example, if I input A sunset over a mountain range, vector image(sunset on the mountains), look at the effect, and the corresponding picture can be generated in a few seconds
shortcoming
Most of the AI painting tools, such as MJ, have some common problems, such as: 收费、外网、使用复杂、全英文、不开源, or the quality of the output is poor
. Even the open source Stable Diffusion still has many problems in use.
Stable Diffusion is not a software in the true sense. At first, it was released as a source code that laymen could not understand. However, thanks to the strong open source community, its operability It has evolved rapidly in half a year.
Here I have to mention a developer on GitHub called Automatic1111, who made these codes into a small program that runs based on the browser page, which is what we often call Stable Diffusion WebUI(SD Web UI).
Later, some people in China repackaged this webUi. This is the release of the Stable Diffusion integrated package that I mainly want to introduce today—Huisho Launcher. This is an integrated installation package made by a big guy at station b. , which can be downloaded and installed immediately.
Pre-installation instructions
1. Computer configuration requirements
Operating system: Windows.10 or later
CPU: No mandatory requirements
Memory: 8G or more recommended
Graphics card: must be an Nvidia discrete graphics card, with a minimum memory of 4G, A card after 20 series is recommended, and the core display can only
run with the CPU Integrated package is recommended In SSD, improve model loading speed
2. This integration package is suitable for the following groups of people
People who have zero foundation and have never used AI painting People
who have broken the original integration package by themselves People
who don’t want to use their brains
3. Other instructions
If you have already installed the original integration package, the content of this integration package upgrade can be upgraded by yourself using the launcher, please decide for yourself.
The integration package is made based on the open source project stable diffusion webui. There has never been any official version of AI painting. This integration package will not be any different from the so-called "self-deploying official version". Generally speaking, if something goes wrong, it must be your own problem.
The integration package only packs the python and git environment necessary for running, and presets the model and adds some commonly used plug-ins. All environments are deleted as needed in this package. The environment is independent and virtual, and there will be no conflicts. In theory, it is better than your own deployment 稳定.
how to install
1. Warm reminder
1.1 Try not to have Chinese in the installation path
The installation path, that is, the folder where you placed Stable Diffusion WebUl and the folders it belongs to. Any Chinese in any layer of folders will cause some path indexing processes to report errors. For example: D:/Program&Files/Al drawing
(in order to reduce the probability of errors, you should also reduce the input spaces as much as possible.)
1.2. The WebUl folder is placed on an empty disk
AI painting needs to download and use a large number of model files. After the number of models is large, it will take up a lot of space. Try not to accumulate them on the system disk such as the C disk.
(The folder can be copied freely without affecting the use.)
2. Download
Baidu disk: https://pan.baidu.com/s/1_ibEk2OpKHxmEg4AnFOpSA
Extraction code: b145
3. Installation
After downloading and decompressing the integrated package, if you have not installed the launcher before, click the launcher to install the dependencies
4. start
After the installation is complete, click the launcher
and then click Start to
pop up the webpage, and the startup is complete, and you can start AI painting
easy to use
The positive Tags (desired content) are these, Tags for improving image quality:
masterpiece, best quality,
Universal negative tags (unwanted content), guarantee that there will be no tags used by ancient gods:
lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry
Complete the filling, save the others as default, and click Generate
Introduction of various parameters
There is usually a plug-in that automatically completes Tags in the integration package. If you don’t know those Tags, I will talk about how to find them below.
In addition, you may see some symbols in the Tags sent by others? Such as parentheses and so on. This is an advanced usage, which is only briefly mentioned here. Take the Tag girl as an example.
(girl) 加权重，这里是1.1倍。括号是可以叠加的，如（(girl)) 加很多权重。1.1*1.1=1.21倍
[girl] 减权重，一般用的少。减权重也一般就用下面的指定倍数。
(girl:1.5) 指定倍数，这里是1.5倍的权重。还可以 (girl:0.9) 达到减权重的效果
The number of sampling steps does not need to be too large, generally within 50. Usually 28 is a good value.
There are no better or worse samplers, but they differ in speed. It's all about personal preference. The recommended ones are the ones circled in the picture, and the speed and effect are all good.
The relevance of the prompt word represents how much the Tag you input guides the screen, which can be understood as "the smaller the AI, the more free it is to play"
If it is too large, the effect of sharpening and thickening of lines will appear. Too small AI can play freely, don't look at the Tag
The random seed is the source of all the randomness in the generation process Each seed is a different picture. The default -1 means to change a random seed each time. From the random seed, a random noise map is generated, and then handed over to AI for drawing.
Model download and use
Here is an option to choose the StableDiffusion model. Different models can bring completely different screen spaces and painting styles to your works. The launcher has already brought a two-dimensional character generation model by default. If you need other types, you can go by yourself. download
For AI painting, only by satisfying the prompt words + model + parameters can the content you want be produced
Introduction to model installation and use:
https://www.bilibili.com/read/cv21362202
Commonly used model download address recommendation
https://huggingface.co/
https://civitai.com/ (requires magic)
Tag reference website
tag collection store
https://tags.novelai.dev/
https://ai.dawnmark.cn/
http://www.atoolbox.net/Tool.php?Id=1101
Refer to the TAG generated by others
https://www.liblibai.com/
https://arthub.ai/
https://civitai.com/ (the aforementioned civitai is also available, but requires magic)
Find Model Classification
Since the types of models are different and the positions of action are also different, if you want to use these model files, you must distinguish these model categories and use the corresponding method correctly to make the model effective.
How to distinguish these models is a very difficult thing for novices, because they can all have the same suffix. Usually people without professional knowledge should simply judge the model category by the file size.
For this, don’t worry, I saw that Mr. Akiba also specially developed this tool to quickly identify the type of model. Just drag the model in
Link: https://spell.novelai.dev/
What will be updated later
Finally, depending on your feedback, the general direction of future updates may be but not limited to the following content,
- Install webui on mac
- Enter Chinese automatic tag prompt
- back push tag
- Use of graphs
- Introduction to plug-ins, various useful plug-ins
- Model introduction, download and installation of different models
- Partial redrawing, character change clothes, change background
- Post-processing, repair high-definition pictures
- The use of lora, recommend some lora, train yourself lora
- Generate pictures according to specific poses and expressions
- AI hand painting problem is the solution
Share some of my creations
Well, this is the end of this brief introduction. This issue is just an introduction, and we will see the situation later in the advanced stage.
If you still want to know more about AI painting, you can go online to find other courses to learn, or you can wait for my update, remember to follow me to get the latest news.
As long as you are familiar with it, you can generate corresponding pictures according to your own wishes
reference
[Address of Autumn Leaf Station B] https://space.bilibili.com/12566101
complete
Well, I am Xiang Yu, https://xiangyu.blog.csdn.net/
A developer who worked silently in a small company, out of interest, began to study unity by himself. Recently, I created a new column [You Ask Me Answer], mainly to collect your questions. Sometimes a question may not be clear in a few words, and I will answer it in the form of an article. Although I may not necessarily know some problems, I will check the information of various parties and try to give the best advice. I hope to help more people who want to learn programming. Let's encourage each other~
I know that you are a master whoring, but if I say there is a reward button below, how should you respond?
(Do what you can, although the strength is small, the heart is warm, your support is the biggest motivation for my creation)