NEW Stable diffusion 2.1 RELEASED!

Sebastian Kamph
7 Dec 202210:15

TLDRStable Fusion 2.1 has been released as an improvement over the poorly received 2.0 version. The new update focuses on better prompting styles, increased data training, and less restrictive data set filtering, addressing previous issues with image quality and variety. The update promises enhanced image generation, particularly in architecture, interiors, and landscapes, and introduces support for non-standard resolutions. Users are encouraged to explore the updated model and share their experiences.

Takeaways

  • πŸš€ Stable Fusion 2.1 has been released as an improvement over the poorly received 2.0 version.
  • πŸ€– The 2.0 models were not functioning as expected, leading to user dissatisfaction with the results.
  • 🌟 Users found ways to improve results with negative prompts after the 2.0 release.
  • πŸ“ˆ Stable Fusion 2.1 promises better performance and has been optimized based on user feedback.
  • 🎨 The new version 2.1 supports a new prompting style and brings back many prompts.
  • πŸ” There's more data and less restrictive filtering in the 2.1 model, addressing previous concerns.
  • πŸ™οΈ Improvements in image quality are noted for architecture, interior design, wildlife, and landscape scenes.
  • πŸ‘€ The issues with generating images of people have been addressed for better results.
  • πŸ–ΌοΈ The model now better handles a range of art styles and improved anatomy and hands.
  • πŸ“ Non-standard resolutions are supported for more creative flexibility in image generation.
  • πŸ’» The release is open-source and available on Hugging Face for those interested in using or testing it.

Q & A

  • What is the main issue addressed in the release of Stable Fusion 2.1?

    -The main issue addressed in Stable Fusion 2.1 is the improvement over the previous version, 2.0, which had a problematic release due to changes in the model's functioning that led to mostly poor results for users.

  • How did users respond to the release of Stable Fusion 2.0?

    -Users responded negatively to the release of Stable Fusion 2.0 because it resulted in terrible outcomes, primarily due to the significant changes in how the model worked and the restrictive filtering of the data set.

  • What was the primary complaint about the data set in Stable Fusion 2.0?

    -The primary complaint was the restrictive filtering of the data set, which led to the removal of many images and made it harder for users to generate desired results, particularly when it came to generating images of people.

  • What improvements have been made in Stable Fusion 2.1 regarding the data set?

    -Stable Fusion 2.1 has improved by having more data, more training, and less restrictive filtering of the data set. The new model aims to capture the best of both worlds, rendering beautiful architectural concepts, natural scenery, and producing fantastic images of people and pop culture.

  • How does Stable Fusion 2.1 address the issue of generating images of people?

    -Stable Fusion 2.1 addresses this issue by fine-tuning the model with an updated setting, which delivers improved anatomy, hands, and a better range of incredible art styles compared to version 2.0.

  • What new features are supported in Stable Fusion 2.1?

    -Stable Fusion 2.1 supports a new prompting style, brings back many prompts, and allows for non-standard resolutions that help create extreme aspect ratios and epic widescreen imaging.

  • How does Stable Fusion 2.1 handle adult content in the data set?

    -Stable Fusion 2.1 still filters out adult content but in a less aggressive manner, which reduces the number of false positives detected, allowing for a more diverse range of outputs.

  • What is the role of negative prompts in Stable Fusion 2.1?

    -Negative prompts continue to play a role in Stable Fusion 2.1 by helping to reinforce the visual fidelity and style of the generated images. However, the need for negative prompts has been reduced compared to version 2.0.

  • How can users access and use Stable Fusion 2.1?

    -Stable Fusion 2.1 is available as an open-source release on Hugging Face, and users can find the weights and checkpoints there. Additionally, users can utilize Dream Studio, a platform provided by Stability AI, to access and use the model.

  • What are some of the sample prompts showcased in the script for Stable Fusion 2.1?

    -Some of the sample prompts include 'a beautiful blonde woman fine art photography' and 'valley in the outfit sunset epic vista beautiful landscape 4K 8K'. These prompts demonstrate the model's capability to generate high-quality images with diverse styles and resolutions.

  • What is the advice given to users who are still using earlier versions of Stable Fusion?

    -Users who are still using earlier versions like 1.4 or 1.5 are encouraged to try out Stable Fusion 2.1 and share their experiences and results in the comments section of the related content.

Outlines

00:00

πŸš€ Introduction to Stable Fusion 2.1

This paragraph introduces the release of Stable Fusion 2.1, reflecting on the previous version 2.0 which was considered a fiasco due to the significant changes in the model's functionality that led to unsatisfactory results for most users. The new version 2.1 is presented as an improvement, with the developers learning from their mistakes and promising better performance. It is mentioned that the updated version supports a new prompting style, brings back various prompts, and includes more data and less restrictive filtering of the data set. The paragraph also discusses the improvements in image quality for architecture, interior design, wildlife, and landscape scenes, while addressing the previous issues with generating images of people.

05:01

🎨 Enhanced Features and User Feedback in Stable Fusion 2.1

The second paragraph delves into the specific enhancements made in Stable Fusion 2.1, such as improved anatomy, better handling of art styles, and the ability to render non-standard resolutions for more creative flexibility. It highlights how the developers have listened to user feedback and adjusted the filters to be less aggressive while still excluding adult content. The paragraph also discusses the model's ability to produce high-quality images of people and pop culture, and the various tools and methods available for using negative prompts. Additionally, it mentions the open-source nature of Stable Fusion and its availability on Hugging Face, as well as the anticipation for its integration with other platforms like Automatic 1.1.1.

10:03

πŸ“’ Closing Remarks on Stable Fusion 2.1

In the final paragraph, the speaker shares a closing remark on Stable Fusion 2.1, encouraging viewers to try out the new version and share their thoughts. The speaker also mentions a personal preference for earlier versions, particularly 1.4 or 1.5, and the versatility they offer when combined with other models. The paragraph concludes with an invitation for feedback in the comments section and a sign-off for the next video.

Mindmap

Keywords

πŸ’‘Stable Fusion 2.1

Stable Fusion 2.1 is the updated version of a machine learning model used for image generation. It is an improvement over the previous version, 2.0, which faced criticism for its output quality. The new version aims to address the issues by supporting a new prompting style, allowing for better image generation, particularly in areas such as architecture, interior design, wildlife, and landscape scenes.

πŸ’‘Negative Prompts

Negative prompts are a technique used in machine learning models to guide the output by specifying what should not be included in the generated images. They are used to refine the results and avoid unwanted features, such as 'blurry poorly drawn face' in the context of image generation.

πŸ’‘Data Set

The data set refers to the collection of data used to train machine learning models. In the context of Stable Fusion 2.1, the data set has been expanded and refiltered to include more diverse and wide-ranging content, while still excluding adult content, which was a point of contention in the previous version.

πŸ’‘Image Quality

Image quality refers to the resolution, clarity, and overall visual appeal of the images produced by the machine learning model. The improvements in Stable Fusion 2.1 are aimed at enhancing image quality, especially in generating architecture, interior design, and landscape scenes.

πŸ’‘Anatomy and Hands

Anatomy and hands refer to the accuracy and detail in the depiction of human anatomy and hands in the generated images. Stable Fusion 2.1 claims to have made significant improvements in this area, addressing one of the main criticisms of the previous version.

πŸ’‘Art Styles

Art styles encompass the various visual aesthetics and techniques used in creating images. The script suggests that Stable Fusion 2.1 has been enhanced to better handle a range of art styles, indicating an improvement in the model's versatility and adaptability.

πŸ’‘Aspect Ratio

Aspect ratio refers to the proportional relationship between the width and height of an image. The script indicates that Stable Fusion 2.1 now supports a wider range of aspect ratios, allowing for more varied and less cropped images.

πŸ’‘Open Source

Open source describes software or models that are publicly available for use and modification without restriction. The script mentions that Stable Fusion is an open-source release, meaning it can be accessed and used by the community for various purposes.

πŸ’‘Dream Studio

Dream Studio is a platform mentioned in the script that utilizes the Stable Fusion model for image generation. It represents an interface through which users can engage with the model and generate images without needing to download and work with the models directly.

πŸ’‘Non-Standard Resolution

Non-standard resolution refers to image dimensions that do not conform to common standards or formats. The script highlights that Stable Fusion 2.1 has the capability to render images at non-standard resolutions, which can be useful for creating images with unique aspect ratios or for specific display requirements.

Highlights

Stable Fusion 2.1 has been released as an improvement over the previous version 2.0.

The 2.0 release was considered a fiasco due to significant changes in the model's functionality and poor user results.

The new 2.1 version promises better performance and user experience, with faster releases as part of their improvement strategy.

The 2.1 update includes support for a new prompting style and brings back many prompts that were previously used.

Stable Fusion 2.1 addresses the issue of restrictive data filtering by easing up on the filters while still keeping NSFW content out.

The model now has a more diverse and wide-ranging data set, which has improved image quality in various categories.

One of the main issues with 2.0, generating poor images of people, is hoped to be resolved in the 2.1 version.

The new version is said to be better at rendering architecture, landscapes, and buildings.

Stable Fusion 2.1 allows for non-standard resolutions, opening up possibilities for creating unique images with various aspect ratios.

The model now delivers improved anatomy and hands, and is more adept at a range of art styles compared to version 2.0.

The release includes various examples, including superheroes like Superman and Wonder Woman, showcasing the model's capabilities.

Negative prompts are still used in 2.1, but the way they are implemented varies between different tools like Dream Studio, Automatic 11 11, and Invoke.

Stable Fusion 2.1 is an open-source release and available on Hugging Face for those interested in exploring it further.

Dream Studio, a platform by Stability AI, offers an easy way to use the new model without downloading weights or using other UIs.

The video encourages users to try out Stable Fusion 2.1 and share their experiences and creations in the comments.