SD3 - Local Install Guide! FASTEST Way to run the new Model - Stable Diffusion 3
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 emphasizes signing a free license for non-commercial use and choosing the right model file. The guide also covers setting up the software, updating COMU, and using different workflows for generating images based on text prompts. The host shares their experience with the model's creativity and understanding of text prompts, showcasing the potential of Stable Diffusion 3.
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 yet.
- 📝 To use Stable Diffusion 3, you need to sign a free license for non-commercial use on Hugging Face; for commercial use, contact Stability AI.
- 📚 There are multiple versions of the model available, but the one including the CLIP encoder is recommended for better functionality.
- 💾 The model file size ranges from 6 GB to 11 GB, depending on the version, and should be downloaded into the appropriate models folder.
- 🔧 COMfy UI (COMUI) is the recommended interface for running Stable Diffusion 3, and it offers various workflows to try out.
- 🔄 Before running the model, ensure COMUI is updated to the latest version, which may require updating through the COMUI manager.
- 🛠️ If there are issues with running COMUI after updating, manually update COMUI and Python dependencies from the portable folder.
- 📈 Workflows for different model versions are provided by the developer of COMUI, and they can be loaded by dragging into the COMUI canvas.
- 🎨 The video demonstrates a simple text-to-image workflow using the model, with customizable settings like the scheduler and sampler.
- 👍 The model's ability to understand and creatively interpret text prompts is highlighted, showcasing its advanced capabilities.
Q & A
What is the Stable Diffusion 3 medium model?
-The Stable Diffusion 3 medium model is a version of the AI model that does not include the text encoder. It is recommended to use the 'sd3 medium including clip save tensor' file, which is around 6 GB for better functionality.
Where can I find the Stable Diffusion 3 model for download?
-You can find the Stable Diffusion 3 model for download on Hugging Face, as mentioned in the transcript.
What is the purpose of signing the license on Hugging Face?
-The license on Hugging Face is for non-commercial use. It requires users to sign with their name and email to ensure proper usage terms. For commercial use, one must contact Stability AI for a commercial license.
What is the difference between the 'sd3 medium safe tensor' and 'sd3 medium including clip save tensor' models?
-The 'sd3 medium safe tensor' model does not include the text encoder, while the 'sd3 medium including clip save tensor' model does, making it more suitable for generating images based on text prompts.
How can I update COMfy to use the Stable Diffusion 3 model?
-To update COMfy, you need to use the COMfy manager to update all components and restart COMfy. If there are issues, you can manually update the COMfy and Python dependencies from the update folder.
What are the different workflows available for Stable Diffusion 3 in COMfy?
-There are multiple workflows available such as a basic workflow, a multi-prompt workflow, and an upscaling workflow. Users can download and try them out in COMfy.
How can I test the Stable Diffusion 3 model with different prompts?
-You can test the model using the 'sd3 demo prompts.txt' file provided, which contains multiple different prompts to try out.
What is the recommended workflow setting for the Stable Diffusion 3 model in COMfy?
-The recommended settings include using the SGM uniform scheduler with 30 steps and a CFG value of 5.5, along with the ULER sampler.
How can I fix the torch Cuda model if it breaks after updating COMfy?
-If the torch Cuda model breaks, you can fix it by going into the COMfy windows portable folder, accessing the update folder, and running the 'update COMfy and Python dependencies' file.
What is the role of the 'empty latent' in the workflow?
-The 'empty latent' in the workflow allows you to set the size of the empty latent space yourself, which can be adjusted according to your needs.
Can I use the Stable Diffusion 3 model for commercial purposes without a license?
-For commercial use of the Stable Diffusion 3 model, you must reach out to Stability AI to obtain a commercial use license.
Outlines
😀 Downloading and Setting Up Stable Diffusion 3 Medium
The video script introduces Stable Diffusion 3 Medium, a new AI model for image rendering. It guides viewers through the process of downloading the model from Hugging Face, signing a free license for non-commercial use, and choosing the appropriate model file (sd3 medium including clip save tensor, around 6 GB). The script also covers downloading workflows and demo prompts from the Hugging Face platform, and provides instructions for setting up the model in Comfy UI, including updating Comfy UI and dealing with potential issues like the broken torch Cuda model. The presenter mentions testing the model and sharing results in a follow-up video, with advice on obtaining higher quality images.
😺 Testing Stable Diffusion 3 Medium with a Creative Prompt
In this paragraph, the script describes a test of the Stable Diffusion 3 Medium model using Comfy UI. The presenter demonstrates how to load a checkpoint, set up a Tex to image workflow, and input a creative prompt ('cat holding a sign with the text I love you'). The model's output is a cat with a sign displaying 'I love you' and a heart, showcasing the model's ability to understand and creatively interpret text prompts. The video concludes with a call to action for viewers to like, subscribe, and look forward to more content.
Mindmap
Keywords
💡Stable Diffusion 3
💡Hugging Face
💡License
💡Model Versions
💡Comfy UI
💡Workflows
💡Prompts
💡Update
💡Checkpoint
💡Scheduler and Sampler
💡CFG Value
Highlights
Stable Diffusion 3 medium is released and the guide will show you how to download and run it on your computer.
Images rendered with Stable Diffusion 3 are showcased, with first roll prompts that will be improved in a follow-up video.
A free license is required for non-commercial use, available at Hugging Face, with a note on commercial use licensing.
Different versions of the model are available, with the recommendation to use the 'sd3 medium including clip save tensor' file for text encoding.
Instructions on downloading the model into the 'models' folder for automatic updates or for use with Comfy UI.
Comfy UI offers different workflows for testing Stable Diffusion 3, including basic, multi-prompt, and upscaling workflows.
A 'sd3 demo prompts' text file is available for testing various prompts with the model.
Comfy UI needs to be updated to use the new model, which can be done through the Comfy UI manager.
A potential issue with the torch Cuda model is mentioned, with a solution to fix Comfy UI after the update.
Workflows by Comfy Anonymous are introduced for different model versions, including one for the medium model and another for the model with clip and T5 XXL fp8.
A simple Tex to image workflow is demonstrated, showing how to load checkpoints and set parameters.
Settings recommended by Comfy Anonymous include using the SGM uniform scheduler with 30 steps and a CFG value of 5.5.
An example prompt 'cat holding a sign with the text I love you' is used to demonstrate the model's creative decision-making and text understanding.
The video concludes with a call to action for likes and subscriptions for more content like this.
A music outro is used to close the video.