I will tell you about the AI ​​picture WeBui (Stable Diffness)

 

Today I will guide you the site that draws the picture for free with AI!
You can draw a picture really easily!

Webui (Stable Diffness)

The Stability AI, released on August 22, 2022, is an AI model that converts text into an image. This model is distributed as an open source license, allowing a variety of users to use it freely. When you enter the text, the stability AI creates a high quality image based on that text.

Website address :
https://stability.ai/


characteristic :
Stable Diffusion is a deep learning artificial intelligence model developed based on the "High Resolution Image Synthesis Research" by the Machine Vision & Learning Group (COMPVIS) lab in Munich University in Munich, Germany. It has been developed with the support of stability AI and Runway ML.

Stability AI is a British company called Emad Mostaque, providing computing resources for Stable Diffness, allowing you to learn a LAION-5B database. Unlike the text-to-IMAGE models such as DALL-E 2 or IMAGEN, this model can also be used on computers with VRAMs of 4GB or less by greatly reducing computer resources.

In addition, even if it is expensive, it is open to open source and can be used by the general public. This has opened the age of painting AI, and the AI ​​image service function based on the model continues to increase.

You can use the plug -in "Controlnet" to make a pose. In addition, by using various controlnet auxiliary models, such as the Canny model derived from OpenPose, it adjusts the crops of the body area to help the sketch of the line level.

Stable Diffness is mainly composed of three artificial neural networks: Clip, UNET, and VAE (VAE). When the user enters the text, the text encoder, CLIP, converts the text to the token that UNET can understand. UNET creates an image by removing randomly generated noise based on tokens. Repeating the dinoizing process creates an accurate image, and the VAE plays a role in converting these images to pixels.

Unlike the traditional diffusion probability image creation model, the Stable Diffness has introduced the Otto Incoder before and after to solve the problem that resource consumption increases as the resolution increases. This allows you to create a relatively large resolution image by manipulating noise in a small level of potential space, not the whole image, and not requiring many computing resources. Therefore, the stable difference can be used as a resource of graphics cards used in general assumptions.


License:
STABLE AI introduced an open source license [4] for new machine learning. This license has a different feature from the usual open source license. If you provide a service using Stable Diffness, the user must explicitly comply with that license. In addition, when pin tuning the model, it should be used only for the specific use specified in the license, and should not be used for other purposes.

How to use :
Various open source projects have been developed using Stable Diffness. Below is a description of each project:

1. Stable Diffness Web UI: A project that provides a web -based user interface to make it easy to use the Stable Diffusion model. Developers continue to update and add a variety of features such as GFPGAN correction, ESRGAN upscaling, and Textual inversion in addition to the front end function of the Stable Diffusion.

2. Original author: The original project published by Compvis. Usability is limited, so it can be used for reference purposes.

3. DIFFUSERS: This is a frame for the new Diffusion model provided by Herging Face, a famous machine learning framework provider. It provides a way to make the FINETUNENG of Stable Diffusion easily. It also includes frameworks such as Transformers or Datasets.

4. Diffnessbee: You can run the stable difference directly with the app for the Mac. It is possible to enter text and images, and also support in -painting and outposting functions. The Apple Silicon version uses a neural engine inside Apple Silicon, and the HQ version uses the GPU to increase quality, but the speed is slow. It also supports Intel Mac and will support Windows in the future.

5. Draw Things: You can run Stable Diffness with apps for iOS, iPados and MacOS. It supports three modes: CPU + GPU, CPU + Neural Engine, CPU + GPU + Neural Engine (All). You can use CheckPoint, Lora, Textual Inversion, etc., and offers a similar feature to Webui. The expansion function is not supported, and due to memory capacity constraints, the app may often end due to lack of memory if it runs more than a certain resolution in both old and new devices.

6. Riffusion: This is an example of applying it to composition AI using the spectrogram.

7. Dish Inside AI Image Gallery: We have opened a dedicated gallery that can create AI images using Civitai's service.

The above projects use the Stable Diffusion to create AI images in various functions and environments.