Feb 26, 2023 23:45:00

``SD Toolset'' that helps you acquire the knowledge you need to acquire with the image generation AI ``StableDiffusion''

The model data of the image generation AI '

Stable Diffusion ' is open to the public, so anyone can easily operate it on their home PC . However, using Stable Diffusion to generate the image you want requires a variety of knowledge, and beginners tend to not know what to learn. The ' SD Toolset ', which is open to the public free of charge, organizes the knowledge that should be obtained in Stable Diffusion by unit, so you can learn the necessary knowledge with pinpoint accuracy.

SD Toolset
https://sdtools.org/

SD Toolset looks something like this, the units are visually summarized in the colorful concentric circles on the left, and you can click on the category you are interested in from inside the concentric circles. For example, if you want to know the basic part of Stable Diffusion, click 'Core'.

Then, it changes to a concentric circle with only 'Core' as shown below. Also, the explanation about 'Core' was displayed in English on the right side. If you want to know about model data, click 'Models'.

Major versions of Stable Diffusion have been released up to 2 at the time of article creation. If you want to know version 2, click '2'.

The explanation for ``2'' is ``Stable diffusion 2.0 and 2.1 are not much different, as version 2.1 was released as an improved version. Both LAION-5B (2B consists of 2 billion images 5B means that it consists of about 5 billion images).However, the biggest change seen by users is the open source version of CLIP from OpenAI's

CLIP It's probably changed to OpenCLIP , which is great from an open source perspective, since CLIP's learnings aren't public, but what was easy in v1 is now hard in v2. It is said that there are many cases where it is difficult to migrate easily.”

Regarding '2.1', '(The default resolution of the generated image) is 512 x 512 pixels and 768 x 768 pixels. Because OpenCLIP is used instead of CLIP, the workflow up to version 1.5 cannot be reproduced with version 2.1. It seems that there were many dissatisfied voices, but fine-tuned models and new embeddings (embedding) are appearing every day, and the functions of version 2.1 have been expanded and are developing rapidly.'

'512depth' is a function that can hold the composition of the read image with higher accuracy than the conventional img2img. With this kind of feeling, it is possible to systematically learn the knowledge necessary for using Stable Diffusion by using SD Toolset.

In addition, there are cases where links to useful tools are introduced. For example, when opening '4x RealESRGAN' with 'ESRGAN' from 'Upscaling' in 'Finishing', the link destination of '4x RealESRGAN', one of the super-resolution libraries, was posted. In this way, you can also know useful tools that can be used with Stable Diffusion.