bye midjourney! SDXL 1.0 - How to install Stable Diffusion XL 1.0 (Automatic1111 & ComfyUI Tutorial)

CoderX
26 Jul 202308:20

TLDRThe tutorial introduces the release of Stable Diffusion XL 1.0, a tool for generating diverse and high-quality images for free. It outlines three methods for using SDXL: running it locally with Automatic1111 UI, with ComfyUI which supports the refiner model, and using it online via clipdrop.co. The video demonstrates the installation process, image generation with different prompts, and the use of the refiner model for enhanced detail. It promises future content on advanced features and optimization for low VRAM devices.

Takeaways

  • 🌟 Stable Diffusion XL 1.0 (SDXL 1.0) is a free tool for generating various styles and types of images.
  • 🖼️ The tool can produce realistic and anime-style images, as well as other interesting content.
  • 🛠️ There are three methods to use SDXL 1.0: locally with Automatic1111, locally with ComfyUI, and online for free.
  • 📥 To run locally, download the base model and optionally the refiner model from the Hugging Face repository by Stability AI.
  • 🔗 Use the pre-built file from the official GitHub repository of Automatic1111 for installation without needing to install Python or Git.
  • 🔄 Extract the downloaded files and place the SDXL base model in the specified folder within the Automatic1111 web UI directory.
  • 🔧 Run the 'update.bat' and 'run.bat' scripts to update the code and start the server for the Automatic1111 web UI.
  • 🖼️ Adjust image resolution to 1024x1024 for better results, as SDXL was primarily trained on this size.
  • 🔄 For ComfyUI, download and extract the stable build, then place the SDXL models in the correct folder and run the 'run_nvidia_gpu.bat'.
  • 🔍 The refiner model can be used in ComfyUI to add more details to generated images, improving face, background, and texture quality.
  • 🌐 If local running is not possible, use clipdrop.co for free online image generation with a limit of 400 free generations.

Q & A

  • What is Stable Diffusion XL 1.0 (SDXL 1.0) and what can it be used for?

    -Stable Diffusion XL 1.0 is an AI model that can generate a wide variety of images in different styles, types, subjects, and backgrounds. It is used for creating realistic and anime-style images, among other types, and is available for free.

  • How many methods are presented in the script for using SDXL 1.0?

    -The script presents three methods for using SDXL 1.0: running it locally with automatic1111, running it locally with ComfyUI, and using it online for free.

  • What is the first step to run SDXL 1.0 locally using automatic1111?

    -The first step is to download the SDXL 1.0 base model from the huggingface repository by Stability AI. Optionally, the refiner model can also be downloaded.

  • Where can I find the pre-built file for automatic1111 to run SDXL 1.0 without installing Python or Git?

    -The pre-built file for automatic1111 can be found on the official GitHub repository of automatic1111, under the 'Installing and running' section for Windows 10/11 with NVIDIA GPUs.

  • What is the purpose of the refiner model in SDXL 1.0?

    -The refiner model in SDXL 1.0 is used to add more details to the generated images, enhancing the quality of faces, backgrounds, and other minor details. It is not required for Auto1111.

  • How does the script suggest changing the image resolution for better generation results with SDXL 1.0?

    -The script suggests changing the width and height to 1024x1024, which is the primary training size for SDXL, to generate better images. Other native resolutions can also be used for different types of images like landscapes.

  • What is ComfyUI and how does it differ from automatic1111 in terms of SDXL 1.0 support?

    -ComfyUI is another UI for running SDXL 1.0 locally. It has the best support for SDXL since its launch and natively supports the SDXL refiner, which is an advantage over automatic1111.

  • How can I access SDXL 1.0 if I am unable to run it locally?

    -If running SDXL 1.0 locally is not possible, the script suggests using it online for free on clipdrop.co, which is a product by Stability AI, the company behind SDXL.

  • What is the process of using the refiner model with ComfyUI as described in the script?

    -To use the refiner model with ComfyUI, one needs to download and install ComfyUI, paste the SDXL 1.0 base and refiner models into the appropriate folders, and then run the UI. After loading the refiner model, one can generate images and compare the results with and without the refiner.

  • What are some of the additional features or topics the script promises to cover in future videos?

    -The script mentions that future videos will cover topics such as LoRA training, Dreambooth, ControlNet, and running SDXL on low VRAM devices.

Outlines

00:00

🖼️ Introducing SDXL 1.0: Free Image Generation Tool

The script introduces the release of SDXL 1.0, a tool for generating a wide variety of images in different styles, types, subjects, and backgrounds at no cost. It showcases examples of realistic and anime-style images, as well as interesting random creations. The tutorial then outlines three methods for using SDXL: running it locally with automatic1111 UI, comfy UI, and online for free if local running is not feasible. The process includes downloading the base model and optionally the refiner model from the huggingface repository, using pre-built files for automatic1111, and setting up the environment on Windows with NVIDIA GPUs. The script also demonstrates how to generate images with specific prompts and resolutions, noting that SDXL performs best with 1024x1024 images.

05:01

🔧 Setting Up and Using ComfyUI with SDXL 1.0

This paragraph explains how to set up and use ComfyUI as an alternative to automatic1111 for running SDXL 1.0, especially when using the refiner model. It details the process of downloading and extracting ComfyUI, placing the SDXL 1.0 base and refiner models in the correct folders, and launching ComfyUI with a batch file. The script then guides through generating images with positive and negative prompts, adjusting the sampler, and changing image resolutions. It highlights the benefits of using the refiner model for additional image details and provides a comparison of image quality with and without the refiner. The paragraph concludes with a mention of an online option, clipdrop.co, for those who cannot run SDXL locally, and a teaser for future videos on related topics.

Mindmap

Keywords

💡Stable Diffusion XL 1.0

Stable Diffusion XL 1.0, often abbreviated as SDXL 1.0, is a significant update to the Stable Diffusion model, an AI-based image synthesis tool that can generate a wide range of images in various styles. In the video, it is presented as a free tool that can create realistic and anime-style images, among others. The script details the process of installing and using SDXL 1.0 to generate images, emphasizing its capabilities and the improvements over previous versions.

💡Automatic1111

Automatic1111 is mentioned as the most popular user interface for the Stable Diffusion model. It is used to run SDXL 1.0 locally on a user's computer. The script provides a step-by-step guide on how to download and set up Automatic1111 to work with the new SDXL 1.0 model, highlighting its ease of use and popularity among users.

💡ComfyUI

ComfyUI is another user interface option for running SDXL 1.0 locally, which is noted for its support for the refiner model and its user-friendly interface. The script explains how to install and use ComfyUI, including the process of adding the SDXL 1.0 base and refiner models to the application, and how it can enhance the image generation process with additional features.

💡Refiner Model

The refiner model is an optional component that can be used with SDXL 1.0 to enhance the quality and detail of generated images. The script describes how to download and integrate the refiner model into both Automatic1111 and ComfyUI, and it demonstrates the improved results when using the refiner model in conjunction with the base model.

💡Huggingface Repository

The Huggingface repository by Stability AI is the source from which users can download the SDXL 1.0 base model and the optional refiner model. The script instructs viewers to visit this repository and download the necessary .safetensor files to get started with image generation using SDXL 1.0.

💡GitHub Repository

The GitHub repository is where the source code and installation files for Automatic1111 and ComfyUI are located. The script directs users to the respective GitHub repositories to download the pre-built files for these UIs, emphasizing the ease of installation without the need for additional software like Python or Git.

💡Image Generation

Image generation is the core function of SDXL 1.0, allowing users to create images based on textual prompts. The script provides examples of image generation, including changing parameters such as width, height, and prompts to achieve desired results. It also discusses the use of different models and UIs to enhance the quality and style of generated images.

💡Positive Prompt

A positive prompt is a text description that guides the AI in generating an image with specific characteristics. In the script, examples of positive prompts like 'God of flowers, fantasy setting' are given to illustrate how to instruct SDXL 1.0 to create particular types of images.

💡Negative Prompt

A negative prompt is used to exclude certain elements or qualities from the generated image. The script mentions using a negative prompt like 'worst quality' to ensure that the generated image meets a certain standard of quality, avoiding undesirable traits.

💡Resolution

Resolution refers to the dimensions of the generated image, such as 1024x1024 or other sizes. The script explains that changing the resolution can affect the quality and aspect ratio of the image, with certain resolutions like 1024x1024 being natively supported and recommended for optimal results with SDXL 1.0.

💡Sampler

The sampler is an algorithm used in the image generation process to determine how the AI interprets the prompts and creates the image. The script mentions 'eulerA' as an example of a sampler that can be selected in ComfyUI to influence the style and quality of the generated images.

💡Clipdrop.co

Clipdrop.co is an online platform provided by Stability AI, the company behind SDXL, where users can try out the image generation capabilities of the AI for free, albeit with a limit on the number of generations. The script suggests this as an alternative for users who cannot run SDXL 1.0 locally due to hardware or software limitations.

Highlights

Stable Diffusion XL 1.0 (SDXL 1.0) is released and offers free image generation with various styles, types, subjects, and backgrounds.

The model can generate realistic and anime-style images, as well as interesting random content.

Three methods for using SDXL 1.0 are presented: local installation with Automatic1111, ComfyUI, and online use.

For local use, the base model of SDXL 1.0 is required, with an optional refiner model available for enhanced capabilities.

The huggingface repository by Stability AI is the source for downloading the .safetensor files of the base and refiner models.

Automatic1111's GitHub repository provides a pre-built file for easy installation without the need for Python or Git.

Instructions are given for downloading, extracting, and setting up the Automatic1111 UI for SDXL 1.0.

The update.bat script updates the code to the latest version supporting SDXL, while run.bat installs dependencies and starts the server.

Users can generate images by entering prompts, with recommendations to use 1024x1024 resolution for optimal results.

ComfyUI is highlighted as an alternative with support for the SDXL refiner and better performance for some users.

ComfyUI's GitHub repository offers direct download links for the latest stable and unstable builds.

After extracting ComfyUI, users place the SDXL 1.0 models in the specified folder and run the application using batch files.

ComfyUI allows users to select models and change settings such as sampler and image resolution for customized image generation.

The refiner model in ComfyUI can be loaded to enhance image details, demonstrated with a comparison of generated images.

For users unable to run SDXL 1.0 locally, clipdrop.co is suggested as an online platform with a limited number of free generations.

A future tutorial on running SDXL on Colab is teased, addressing the need for optimized resources for this model.

The video promises further exploration of SDXL 1.0 with topics like LoRA training, Dreambooth, ControlNet, and running on low VRAM devices.