Playground AI Tried to Break the Internet. NEW Stable Diffusion Base Model!

Sebastian Kamph
5 Dec 202310:04

TLDRThe video discusses the release of Playground Version 2, a new AI model developed by the team that claims to outperform the current leading open model, Stable Fusion XL. The model is available for free use on Playground and can be downloaded for local use. The comparison between Playground V2 and Stable Fusion XL is presented, with Playground V2 being preferred 2.5 times more in early benchmarks. The video explores the model's capabilities through various image examples and discusses the potential for commercial use. The model is also available on Hugging Face, and the video provides insights on how to download and utilize it, including some technical details on integrating it locally.


  • πŸš€ Playground has released a new AI model, version 2, claiming it outperforms the stable Fusion XL model.
  • πŸ†“ The new model is available for free use both on the Playground website and for local download.
  • πŸ”₯ The model was trained using SC batch and was reportedly preferred 2.5 times more than the leading open model, SdxL.
  • πŸ“Š A comparison between Playground V2 and SdxL was presented, though it was noted that the comparison was made with the base model of SdxL.
  • πŸ” The blog post suggests that the Playground model is also available on Hugging Face for commercial use.
  • 🎨 The new model offers different styles, including photorealistic and painterly, with varying levels of smoothness and detail.
  • πŸ“Έ Early benchmarks indicate that Playground is preferred over stable Fusion XL, but the comparison could have been made with a better SdxL model.
  • πŸ”— The video script provides a link to user data from the comparisons, which can be checked out for more detailed information.
  • πŸ› οΈ The video demonstrates how to download and use the Playground model locally, including some technical challenges and workarounds.
  • πŸ–ŒοΈ The Playground interface offers various settings for users to customize their AI-generated images, such as size, guidance, quality, and seed.
  • πŸ’¬ The video creator encourages viewers to test the new model and share their experiences and suggestions in the comments section.

Q & A

  • What is the main claim of the new image AI model introduced in the transcript?

    -The main claim is that the new image AI model, referred to as 'playground version 2', outperforms the stable Fusion XL model, which is a significant achievement in the field of AI and machine learning.

  • How can users access and use the new AI model?

    -The new AI model can be accessed and used both on the playground platform and can be downloaded locally for use. It is available for free, allowing users to experiment and utilize its capabilities.

  • What was the comparison basis for the new AI model's performance?

    -The performance of the new AI model was compared with the stable Fusion XL base model. However, the discussion suggests that it would be beneficial to compare it with custom sdx models to get a more accurate assessment of its capabilities.

  • What are the features of the playground version 2 AI model?

    -The playground version 2 AI model offers a range of features including different weight versions (256, 512, and 1024 pixel stages), the ability to generate increasingly powerful graphic models, and it is commercially usable, with permissions for commercial use.

  • Where can the open weights for playground V2 be found?

    -The open weights for playground V2 can be found on Hugging Face, a popular platform for sharing machine learning models.

  • What was the result of early benchmarks for the playground V2 model?

    -Early benchmarks indicated that the playground V2 model was preferred over the stable Fusion XL model by a significant margin, with a preference rate of 2.5 times more.

  • What are the limitations observed in the playground V2 model's image generation?

    -Some limitations observed include the generation of images that are a bit too smooth and airbrushed, with certain features like skin and eyes appearing unrealistic in some instances.

  • How can users provide feedback on the playground V2 model?

    -Users are encouraged to test the playground V2 model and provide their feedback and thoughts in the comments section of the blog post or platform where the model is introduced.

  • What was the process of trying to run the playground V2 model locally?

    -The process involved downloading the full 10 GB file of the model, attempting to integrate it with existing frameworks like comfy, and trying different settings to see how the model performs. However, the process was not straightforward and required some trial and error.

  • What were the results of comparing the playground V2 model with other models using different prompts?

    -The comparisons showed varying results. Some images generated by the playground V2 model were considered good, while others were not as impressive. The prompt and settings used had a significant impact on the quality and style of the generated images.

  • What was the overall conclusion about the playground V2 model?

    -The overall conclusion was that the playground V2 model shows promise and has potential, but it might not be fully optimized or easy to use, especially for non-experts. Further testing and user feedback will likely lead to improvements and a better understanding of its capabilities.



πŸš€ Introduction to Playground Version 2

This paragraph introduces Playground Version 2, a new image AI model claimed by the creators to outperform Stable Fusion XL. It is available for free use both online and for local download. The model was reportedly preferred 2.5 times more than the leading open model, SdxL. A comparison between Playground V2 and SdxL is presented, with Playground V2 showing promise but also highlighting the need for further comparisons with custom Sdx models. The paragraph also mentions the availability of the model on Hugging Face and the commercial use permissions. A blog post is referenced for more detailed comparisons and user data.


🎨 Exploring Playground V2's Image Generation Capabilities

This paragraph delves into the practical application of Playground V2, showcasing its image generation capabilities through various prompts and settings. The user experiments with different styles, including painterly and photorealistic, and compares the results with those produced by other models. The paragraph also discusses the process of running Playground V2 locally, including the challenges faced and the workarounds attempted. The user invites feedback and encourages others to test and share their experiences with the new model.



πŸ’‘AI model

An AI model refers to a system designed to process input data and produce output predictions or decisions based on patterns learned from training data. In the context of the video, the AI model is a new image-generating model that claims to outperform existing models like Stable Fusion XL.


Playground, as mentioned in the video, is a platform where users can access and utilize the new AI model for image generation. It provides an environment for both online use and local download, allowing users to experiment with the model's capabilities.

πŸ’‘Stable Fusion XL

Stable Fusion XL is a leading image-generating AI model that serves as a benchmark for comparison in the video. It is used to demonstrate the superior performance of the new Playground Version 2 AI model.

πŸ’‘Commercial use

Commercial use refers to the application of a product, service, or technology for monetary gain or profit-making purposes. In the context of the video, it is mentioned that the new AI model from Playground allows for commercial use, meaning users can utilize it for business purposes.

πŸ’‘Hugging Face

Hugging Face is a platform that provides a wide range of AI models, including those for natural language processing and computer vision tasks. In the video, it is mentioned as a place where users can find and download different versions of the Playground AI model.


A benchmark is a standard or point of reference against which performance or quality may be compared. In the context of the video, benchmarks are used to evaluate and compare the performance of different AI models, particularly the new Playground Version 2 against Stable Fusion XL.

πŸ’‘Open model

An open model refers to an AI model that is publicly accessible and can be used, modified, and distributed without strict restrictions. The video presents the Playground Version 2 as an open model, suggesting that it is freely available for a wide range of applications.


Photorealism is a style of art or image generation that closely resembles real-life photographs in terms of detail and visual fidelity. In the video, the term is used to express a desire for the AI model to produce images that are highly realistic and indistinguishable from actual photographs.

πŸ’‘User interface

A user interface (UI) is the space where interactions between users and a system, such as a software application or a website, occur. It includes the design and layout of elements that allow users to navigate and use the system effectively. In the video, the user interface of Playground is described, highlighting its features and how it facilitates the use of the AI model.


Oiler is a term used in the context of AI image generation models, referring to a sampling method or algorithm that helps in generating the final output image. It is part of the technical process that determines how the AI model interprets and visualizes the input prompts.

πŸ’‘Custom models

Custom models refer to AI models that have been modified or tailored to specific needs or preferences of the user. In the video, the speaker expresses a desire to compare the new Playground Version 2 model with custom versions of Stable Diffusion XL to evaluate its performance accurately.


Playground claims to have developed an AI model that outperforms Stable Fusion XL.

The new AI model is available for free use both on Playground and for local download.

The model was trained using SC batch and was reportedly preferred 2.5 times more than the current leading open model, SdxL.

Comparison between Playground V2 and SdxL shows that Playground V2 has a visually appealing output.

The blog post provides examples of images generated by Playground V2, showcasing its capabilities.

Playground V2's weights are available for commercial use and can be downloaded from Hugging Face.

The model comes in different pixel stages, including 256, 512, and 1024, with the latter being found through clever searching.

Early benchmarks indicate Playground V2 is preferred over Stable Fusion XL, with user data supporting these claims.

The user interface of Playground is described, including settings for size, guidance, quality, details, and seed.

The AI model can generate images with a painterly or cinematic style, as demonstrated by the examples provided.

The video includes a live demonstration of generating images using Playground V2 and its settings.

The video creator attempts to run the model locally, providing insights into the process and potential challenges.

Comparisons are made between Playground V2 and other models, such as custom SdxL models.

The video encourages viewers to test out the model and share their thoughts in the comments.

The video concludes with the creator's willingness to learn and improve their understanding of the AI model.