Stable Diffusion is a deep learning text-to-image generation model. Its main function is to convert text narratives into images. This series of articles will introduce how to use it.
Text-to-map AI assistant
I believe that after the storm of “Electrical Drawing Godmother”, everyone has a little understanding of AI image calculation. This kind of technology can convert the text input by the user into images, and can generate many interesting pictures in just a few minutes. Not only It can be used as a helpful assistant for art workers, and it is also quite entertaining for ordinary people.
Among various image generation models, Stable Diffusion, jointly developed by companies such as StabilityAI, CompVis, and Runway, is one of the most popular models.Diffusion model(Diffusion Model) as the technical basis, and adoptOpen source releasethe user can follow thelicense termsfor non-commercial and commercial use.
However, the author would like to remind here that although the authorized transfer of the Stable Diffusion program can be used commercially, considering that many AI models may not meet the scope of commercial use, and the current works produced by AI calculations are not yet clear. Therefore, it is recommended to only conduct technical research and experience at this stage, and not to use the produced pictures for commercial use, so as to show respect for the original image author and avoid legal disputes.
Having said that, another feature of Stable Diffusion is that it can be deployed on your own computer. You only need to prepare a set of compatible computers, and you can use it freely without restrictions on the method and times of use. Although the thresholds such as pre-work and cost are higher than online The service is high, but the “playability” that can mess around is more attractive.
System Preparation for Stable Diffusion
Since Stable Diffusion is released in an open source format, many developers have launched different derivative branches and front-end tools. In this series of articles, the author will use the Stable Diffusion WebUI developed by AUTOMATIC1111 as an example, and build Python with Miniconda virtual environment.
However, it should be noted that although according to the AUTOMATIC1111 provided by GitillustrateStable Diffusion WebUI can not only be used with NVIDIA graphics cards, but also throughDirectMLmatchAMD graphics carduse, or inApple SiliconHowever, considering the compatibility and performance of the current implementation, the teaching will only use the NVIDIA graphics card as an example for the time being, and the teaching will be updated in the future depending on the situation.
In the part of hardware requirements for Stable Diffusion, the minimum recommended storage device is a traditional hard disk with a usable capacity of more than 20GB, but it is strongly recommended to prepare a large-capacity solid-state hard disk to store more models and increase the reading speed to improve the rendering accuracy, versatility and speed. As for the display card, it is a model equipped with 6GB or more of display memory. According to the feedback from the community, NVIDIA GeForce RTX 2060 can execute it, but it is still recommended to use a model with 12GB or more of display memory (such as GeForce RTX 3060). If it is paired with GeForce RTX 40 series graphics cards can naturally increase the calculation speed. Other specifications such as processor and memory have no special requirements.
The software part is relatively simple, as long as you install Git, Python, Miniconda and other software on the Windows operating system, and install the optimized function library corresponding to the graphics card, you can download the Stable Diffusion WebUI and various models, and start to calculate the map.
The author will actually demonstrate the operation process of installing and optimizing GeForce RTX 40 series graphics cards in Chapter 2-1 of the teaching, please refer to the series of article catalogs.
(There is also a series of article catalogs on the next page)
