Fooocus is the Stable Diffusion's Answer to Midjourney | Now with 13 Subtitles Languages

Matt Hallett Visual
25 Oct 202310:30

TLDRIn this AI for Architecture tutorial, Matt introduces Focus, an alternative to Automatic 11.11, and recommends a better version created by Moonride303. He guides viewers through installation, basic usage with simple prompts, and advanced features like prompt expansion and control nets for architectural rendering. Matt also discusses the importance of experimenting with different settings and shares his experiences with various parameters.

Takeaways

  • ๐Ÿš€ Introduction to Focus, an alternative to Automatic 11.11, as a tool for AI in architecture.
  • ๐Ÿ‘จโ€๐Ÿ’ป Matt's recommendation to check out Olivio Sarracus for AI-generated tips and tricks.
  • ๐Ÿ” Guidance on finding the main GitHub page for Focus and installing the preferred version by Moonride303.
  • ๐Ÿ“‚ Instructions for handling the 7Z compressed file and organizing the stable diffusion engines directory.
  • ๐Ÿ“ Details on editing the paths.json file for accessing different types of embeddings and control nets.
  • ๐Ÿ–ฅ๏ธ Explanation of the simple and intuitive interface of FocusMRE for quick results with minimal prompts.
  • ๐Ÿ  Demonstration of the prompt system, including the use of brackets for emphasis and the photo detail feature.
  • ๐ŸŽจ Discussion on the prompt expansion feature in Focus version 2 and its impact on rendering quality.
  • ๐Ÿ”„ Comparison of different speed and performance settings, highlighting the lack of noticeable differences.
  • ๐Ÿ”ง Insights on the refiner's limited impact with improved models and the suggestion to use realistic stock photo or Civit AI.
  • ๐ŸŽจ Exploration of image styles, LORAs, and sampling settings for achieving desired visual outcomes in architecture renderings.
  • ๐Ÿ–ผ๏ธ Importance of control nets when modifying existing images while preserving the original architectural design.

Q & A

  • What is the main topic of the tutorial?

    -The main topic of the tutorial is an introduction to Focus, an alternative to Automatic 11.11, for architectural design and rendering.

  • Who is the presenter of the tutorial?

    -The presenter of the tutorial is Matt.

  • What is the significance of the name 'Focus' in the context of this tutorial?

    -Focus is the name of a software or tool that serves as an alternative to Automatic 11.11, used for generating AI-based architectural designs and renderings.

  • What is the recommended way to download and install Focus?

    -The recommended way to download and install Focus is by visiting its main GitHub page, scrolling down to the download section, and downloading the Windows 7Z file. This file is a compressed zip file that requires a program like WinRAR to extract.

  • What is the purpose of the 'paths.json' file in Focus?

    -The 'paths.json' file in Focus is used to specify the directories where different types of model checkpoints, such as embeddings and control nets, are located. Users may need to edit this file to point to the correct locations on their system.

  • How does the prompt expansion feature in Focus work?

    -The prompt expansion feature in Focus allows users to add suffixes and additional details to their prompts for more nuanced and detailed image generation. This feature can be found and edited in the backend of the software.

  • What is the difference between the 'realistic stock photo' and 'custom' options in Focus?

    -The 'realistic stock photo' option is a default setting that aims to generate images with a high level of photorealism. The 'custom' option allows users to adjust various parameters and steps to achieve different visual effects and performance levels.

  • How does the 'refiner' function in Focus?

    -The 'refiner' function in Focus is used to further process and improve the generated images. However, according to the tutorial, the refiner does not significantly alter the images when using the base model.

  • What is the role of 'Control Net' in Focus?

    -Control Net in Focus is used when working with existing images to modify them through stable diffusion while maintaining certain aspects of the original image, such as the architectural design. It creates a noise matrix in the latent space that combines the prompts and the base image to generate a new image.

  • What is 'FreeU' and how does it affect the images generated by Focus?

    -FreeU is an add-on included in the version of Focus discussed in the tutorial. It appears to add drama and interesting color to the images, similar to passing the image through a filter in Photoshop, and can create an almost HDRI effect.

  • What resources does the presenter offer for learning more about stable diffusion in architecture?

    -The presenter offers tutorials on his website, HallettVisual.com, where he provides more in-depth information and guidance on using stable diffusion in the architectural workflow. He also shares his work and posts updates on Instagram and Facebook.

Outlines

00:00

๐Ÿ“š Introduction to Focus and Stable Diffusion Engines

The video begins with the host, Matt, welcoming viewers to an AI for Architecture tutorial focused on Focus, an alternative to Automatic 11.11. Matt introduces himself and recommends checking out Olivio Sarracus' channel for AI-generated tips and tricks. He proceeds to guide viewers to the main GitHub page for Focus and discusses a recommended fork by Moonride303, which offers improved features for architectural uses, such as control net. Matt explains the process of downloading and setting up the Focus engine, including the need to edit the paths.json file to access checkpoints and the importance of using the correct file structure and naming conventions. He emphasizes the simplicity of FocusMRE and the ability to generate images with minimal prompts, like 'a modern house in a forest next to a lake with fog.' Matt also discusses the prompt expansion feature and its impact on image generation, as well as the default settings for generating images.

05:02

๐Ÿ–ผ๏ธ Customizing Image Generation with Base Models and Styles

In this paragraph, Matt delves into the customization options available in Focus, starting with the base model. He mentions that the realistic stock photo may be the default setting, which is beneficial for image generation. Matt advises viewers on how to download and install additional models, such as the Civit AI, and provides guidance on maintaining the default settings for the demonstration. Matt introduces the concept of prompt expansion with image styles, demonstrating how adding 'photo long exposure' can affect the generated image. He also discusses the use of LORAs, which are add-ons to the base stable diffusion model, and the importance of the sampling.CFG setting in determining how closely the image generator adheres to the prompts. Matt touches on the sampler options, such as SD++, and the impact of different schedulers like Keras and Exponential. He introduces FreeU as an interesting add-on that adds drama and color to the images, comparing its effect to passing the image through a Photoshop filter. Matt then explains the significance of control nets when using existing images, particularly for architectural work, and how they can be used to modify images while preserving the original design. He also mentions his tutorial series on Control Net usage, available on his website, howlattvisual.com.

10:03

๐ŸŽจ Final Thoughts and Resources for Architecture and Stable Diffusion

Matt concludes the tutorial by reiterating his role as an architectural renderer and his focus on providing tutorials for architectural workflows with stable diffusion. He invites viewers to check out his website, HallettVisual.com, for more information on learning about stable diffusion in architecture. Matt also encourages viewers to follow him on Instagram and Facebook, where he shares his work and posts updates. He ends the video by thanking viewers for watching and expressing indifference towards likes and subscriptions, emphasizing his primary goal of sharing knowledge and resources for architectural rendering.

Mindmap

Keywords

๐Ÿ’กAI for Architecture

AI for Architecture refers to the application of artificial intelligence technologies in the field of architecture. In the context of the video, it involves using AI tools like Foocus to generate architectural visualizations and enhance the design process. The video aims to provide a tutorial on how to utilize these AI tools effectively for architectural renderings and design exploration.

๐Ÿ’กFoocus

Foocus is an AI tool mentioned in the video that serves as an alternative to another AI software called Automatic 11.11. It is used for generating images based on textual prompts, which can be particularly useful in creating architectural visualizations. The video provides guidance on how to install and use Foocus, including its features and settings.

๐Ÿ’กGitHub

GitHub is a web-based hosting service for version control and source code management, where developers can share and collaborate on projects. In the video, the speaker directs the audience to the main GitHub page for Foocus to download the software and mentions a fork created by a user named Moonride303, which he considers an improved version.

๐Ÿ’กControl Net

Control Net is a feature in AI image generation tools that allows users to influence the output by using an existing image as a reference. This helps in maintaining certain visual elements from the base image while introducing changes from the textual prompts. In the context of the video, Control Net is important for architectural visualizations where the integrity of the design needs to be preserved.

๐Ÿ’กPrompt Expansion

Prompt Expansion is a feature in AI image generation tools that automatically adds suffixes or additional descriptive words to the user's initial textual prompt. This aims to provide more detailed and nuanced outputs based on the prompt. The video explores the use of Prompt Expansion in Foocus and how it can affect the generation of images.

๐Ÿ’กStable Diffusion

Stable Diffusion is a type of AI model used for generating images from textual descriptions. It is the underlying technology in tools like Foocus. The video discusses the use of Stable Diffusion and its various settings, such as LORAs and sampling methods, to control the image generation process and achieve desired visual effects.

๐Ÿ’กPhoto Realism

Photo Realism is a visual quality in generated images that aims to closely resemble real-world photographs. In the context of the video, the speaker is interested in achieving photo realism with the AI-generated images for architectural visualizations, rather than having them look like renders or illustrations.

๐Ÿ’กCFG

CFG, or Context Free Grammar, is a parameter in AI image generation that determines how closely the generated image adheres to the textual prompts. A higher CFG value makes the image more rigidly follow the prompts, while a lower value allows for more flexibility and creative interpretation by the AI.

๐Ÿ’กSampler

In the context of AI image generation, a Sampler refers to the method used to transform the latent space representation back into an image. Different samplers can produce varying visual results and artistic styles. The video mentions different types of samplers like SD++ and their impact on the final image.

๐Ÿ’กScheduler

Scheduler refers to the algorithm used in AI image generation to determine the sampling process over the course of the image creation. Different schedulers, such as Keras and Exponential mentioned in the video, can lead to different results in terms of image quality and style.

๐Ÿ’กFreeU

FreeU is a feature included in the version of Foocus discussed in the video. It is based on a separate research paper and adds dramatic effects and interesting color variations to the generated images, akin to passing the image through a filter in Photoshop. It enhances the visual impact of the scenes by introducing subtle enhancements that contribute to a more dynamic and visually rich output.

Highlights

Introduction to Focus, an alternative to Automatic 11.11 for architectural design.

Recommendation to check out Olivio Sarracus's channel for AI-generated tips and tricks.

Instructions on navigating to the main GitHub page for Focus and installing the software.

Mention of a better version of Focus created by Moonride303 with improved features for architecture.

Explanation of how to handle the 7Z compressed file and its extraction using WinRAR.

Guidance on organizing stable diffusion engines and renaming the paths.example.json file.

Details on editing the paths.json file to adjust directories for various embeddings and control nets.

Running the MRE.bat file and the simplicity of FocusMRE's interface.

Demonstration of the mid-journey experience in FocusMRE with a simple prompt.

Explanation of the prompt expansion feature in Focus version 2 and its backend functionality.

Assessment of the value of prompt expansion in current stable diffusion practices.

Discussion on the performance and speed of image generation with different settings.

Evaluation of the refiner's effectiveness with improved models and the use of the base model.

Instructions on using the realistic stock photo default setting and the Civit AI model.

Explanation of the different styles and image modifications available in Focus.

Discussion on the use of LORAs and their impact on the base stable diffusion model.

Details on the sampling, CFG, and the difference between Keras and Exponential samplers.

Description of the FreeU feature and its effect on image drama and color.

Explanation of control nets and their importance when using existing images in stable diffusion.

Demonstration of how control nets can modify an existing image while preserving the architectural design.