SD3 - Local Install Guide! FASTEST Way to run the new Model - Stable Diffusion 3

Olivio Sarikas
12 Jun 202406:15

TLDRThis video tutorial guides viewers on how to download, install, and run Stable Diffusion 3 Medium for creating high-quality images on their computers. It covers signing a free license for non-commercial use, selecting the appropriate model files, and using specific workflows in comi, a popular tool for AI image generation. The host also shares a link to example prompts and discusses troubleshooting steps for comi updates, showcasing the creative capabilities of the model with a 'cat holding a sign' prompt example.

Takeaways

  • 😀 Stable Diffusion 3 Medium is released and the video will guide you through downloading and running it on your computer.
  • 📷 The images shown are first-roll renders with Stable Diffusion 3, and the prompts used are not optimized for better quality images yet.
  • 📝 To use Stable Diffusion 3, you need to visit Hugging Face and sign a free license for non-commercial use, or contact Stability AI for a commercial license.
  • 📚 There are multiple versions of the model available, but the video recommends downloading the 'sd3 medium including clip save tensor' file, which is around 6 GB.
  • 💾 The model file should be downloaded into the models folder for automatic 1111 or for Comfy UI, which the video will demonstrate.
  • 🔧 Comfy UI needs to be updated to use the new model, which can be done through the Comfy UI Manager or by manually updating Comfy UI and its dependencies.
  • 🛠️ If updating breaks the torch Cuda model, fix it by running the 'update Comfy UI and Python dependencies' file found in the Comfy UI Windows portable folder.
  • 🌐 The video provides links to workflows by Comfy UI Anonymous for both the medium model and the model with clip and T5 XXL fp8.
  • 📑 Workflows can be loaded into Comfy UI by dragging the downloaded images into the canvas.
  • 🎨 The video demonstrates a simple text-to-image workflow using the recommended settings by Comfy UI Anonymous, including the use of the SGM uniform scheduler and the ULER sampler.
  • 🐱 The model's creative decision-making is showcased with an example prompt, 'cat holding a sign with the text I love you,' resulting in an image with a heart, showing the model's understanding of the text.

Q & A

  • What is the purpose of the video?

    -The purpose of the video is to show how to download, update, and run Stable Diffusion 3 on a computer.

  • Where can you find the link to download Stable Diffusion 3?

    -The link to download Stable Diffusion 3 is provided below the video.

  • What must you do before downloading the Stable Diffusion 3 model?

    -Before downloading the Stable Diffusion 3 model, you must sign a free license with your name and email.

  • Which model version does the speaker recommend downloading?

    -The speaker recommends downloading the SD3 medium including clip safe tensor file, which is around 6 GB.

  • Where should you download the model file?

    -You should download the model file into your models folder for either Automatic 1111 or ComfyUI.

  • What should you do if ComfyUI does not start after updating?

    -If ComfyUI does not start after updating, go into the ComfyUI Windows portable folder, click on the update folder, and run the update ComfyUI and Python dependencies file.

  • How can you load the workflows in ComfyUI?

    -To load the workflows in ComfyUI, download the images of the workflows and drag them into the ComfyUI canvas.

  • What scheduler and settings are suggested for generating images?

    -The SGM uniform scheduler with 30 steps, a CFG value of 5.5, and the Euler sampler are suggested for generating images.

  • What is an example prompt used in the video?

    -An example prompt used in the video is 'cat holding a sign with the text I love you.'

  • What does the speaker suggest doing if the initial update method breaks the ComfyUI?

    -The speaker suggests running the update ComfyUI and Python dependencies file found in the update folder of the ComfyUI Windows portable folder.

  • What is included in the SD3 demo prompts file?

    -The SD3 demo prompts file includes multiple different prompts that you can try out to test the model on your computer.

Outlines

00:00

😀 Introduction to Stable Diffusion 3 Medium

The video script introduces the release of Stable Diffusion 3 Medium and guides viewers on how to download, update, and run it on their computers. It starts by directing viewers to Hugging Face to sign a free license for non-commercial use, with an option for a commercial license upon request. The script emphasizes choosing the 'sd3 medium including clip save tensor' file for download, which is around 6 GB, over the 'sd3 medium safe tensor' model that lacks a text encoder. It also mentions the availability of different workflows and demo prompts for testing the model in Comfy UI (comUI), a tool that receives updates first due to its popularity.

05:03

😲 Setting Up and Testing Stable Diffusion 3 Medium in Comfy UI

This paragraph details the setup process for using Stable Diffusion 3 Medium in Comfy UI. It advises updating Comfy UI using the Comfy UI Manager and addresses a potential issue with the torch Cuda model that can be resolved by updating Comfy UI's Python dependencies. The script then explains how to load different workflows provided by Comfy UI's developer, including one specifically for the 'sd3 medium including clip save tensor' model. It demonstrates how to use the workflow with a sample prompt, resulting in a creative image of a cat holding a sign that reads 'I love you' with a heart, showcasing the model's ability to understand and creatively interpret text prompts.

Mindmap

Keywords

💡Stable Diffusion 3

Stable Diffusion 3 is a new model in the realm of AI-generated images. It is a significant update to the previous versions, offering enhanced capabilities for rendering images from textual descriptions. In the video, the host demonstrates how to download and run this model on a computer, highlighting its ability to produce high-quality images from first-roll prompts without any improvements.

💡Hugging Face

Hugging Face is a platform that hosts machine learning models, including the Stable Diffusion 3 model mentioned in the video. The host instructs viewers to visit Hugging Face to sign a license agreement for non-commercial use of the model. This step is crucial for legal compliance and accessing the model files.

💡License

A license in this context refers to a legal agreement that allows users to use the Stable Diffusion 3 model for specific purposes. The video mentions that viewers need to sign a free license for non-commercial use. For commercial use, one must contact Stability AI to obtain the appropriate license.

💡Model Versions

The video script refers to different versions of the Stable Diffusion 3 model available for download. The host points out the importance of choosing the correct version, such as 'sd3 medium including clip save tensor' which is around 6 GB, or the larger 'clip and T5 XXL fp8' version, which is around 11 GB.

💡Comfy UI

Comfy UI, often abbreviated as com UI, is a user interface that simplifies the process of using AI models. In the video, the host uses Comfy UI to demonstrate how to load and run the Stable Diffusion 3 model, showcasing its ease of use for generating images.

💡Workflows

Workflows in the context of the video refer to the step-by-step processes or sequences of actions used to generate images with the Stable Diffusion 3 model. The host mentions different types of workflows available for download, such as basic, multi-prompt, and upscaling workflows, which can be used within Comfy UI.

💡Prompts

Prompts are textual descriptions or commands given to the AI model to generate specific images. The video script includes a 'sd3 demo prompts txt' file that contains various prompts for testing the model's capabilities. The host uses a prompt like 'cat holding a sign with the text I love you' to demonstrate the model's understanding and creative output.

💡Update

Updating, in the context of the video, refers to the process of ensuring that Comfy UI and its components are up-to-date to work with the new Stable Diffusion 3 model. The host describes how to update Comfy UI using the manager extension and dealing with potential issues that may arise during the update process.

💡Checkpoint

A checkpoint in AI model training is a saved state of the model at a certain point in time. In the video, the host mentions loading a checkpoint, specifically 'sd3 medium including clip save tensor', which is necessary for the Stable Diffusion 3 model to generate images based on the given prompts.

💡Scheduler

In the context of AI and machine learning, a scheduler is an algorithm that adjusts the learning rate or other parameters during the training process. The video mentions using an 'sgm uniform scheduler' with 30 steps, which is a specific type of scheduler that helps in the image generation process with the Stable Diffusion 3 model.

💡Uler Sampler

The Uler Sampler is a technique used in AI models to generate images from latent spaces. The host in the video uses the Uler Sampler with specific settings, such as a CFG value of 5.5, to produce images with the Stable Diffusion 3 model, demonstrating the model's ability to interpret and create images from textual prompts.

Highlights

Introduction to Stable Diffusion 3 medium and demonstration of rendered images.

Instructions on how to download Stable Diffusion 3 for non-commercial use and the process for commercial use licensing.

Explanation of the different versions of the model available for download.

Recommendation to use the 'sd3 medium including clip save tensor' file for better functionality.

Details on downloading the model into the automatic 1111 models folder or for com UI.

Introduction to com UI workflows and their download options.

Description of the 'sd3 demo prompts txt' for testing the model.

Steps to update com UI to ensure compatibility with the new model.

Troubleshooting tips for fixing the torch Cuda model if it fails to start.

Guidance on loading workflows in com UI post-update.

Introduction to additional workflows by comy Anonymous for different model versions.

Instructions on how to load and customize workflows in com UI.

Example of a Tex to image workflow setup with the 'sd3 medium including clip save tensor'.

Explanation of the settings recommended by comy Anonymous for the workflow.

Demonstration of a creative output from the model based on the prompt 'cat holding a sign with the text I love you'.

Conclusion and call to action for likes and subscriptions for more content.