Playground v2 is a new text-to-image model that competes with Stable Diffusion XL



summary
Summary

Playground, a team of developers specializing in graphical AI, has released Playground v2, a new diffusion-based generative text-image model.

According to a user study conducted by Playground, images generated with Playground v2 are preferred 2.5 times more than images generated with Stable Diffusion XL. The team ran a comparative study in which thousands of users were asked to choose their favorite image between the two models.

Image: Playground

Playground offers its new model as a basic version with resolutions of 512px and 256px at Huggingface. An “aesthetic” version of the model with a resolution of 1024px is also available for download, as is the v1 version.

Although the resolution of the open models is relatively low, the AI-generated images can be easily scaled up using AI upscalers without significant loss of quality.

Ad

Ad

The AI image model is also available on the Playground website. Anyone who creates an account here can generate 500 images per day for free. The images can be used commercially.

For twelve US dollars per month, users have access to numerous image editing and prompting functions via a graphical user interface. Starting at one million images per month, Playground also offers API access.

Image: Playground AI

Playground AI was founded in San Francisco in 2022 and has raised about $40 million from investors to date. Its stated goal is to create visual AI systems that can make subtle changes to images based on prompts, including the creation of entire 3D environments or complex analysis of video scenes. The image generator is just the first step.

New benchmark and license conditions

In addition to the new model, Playground has also introduced a benchmark, MJHQ-30K, to automatically evaluate the aesthetic quality of a model. The benchmark calculates the Frechet Inception Distance (FID) based on a high-quality dataset. The dataset, curated with Midjourney, contains ten general categories with 3,000 examples each. It also shows how well the image AI can follow prompts.

Image: Playground AI

Playground v2 is available for download from Huggingface under a “community license”. You are granted a non-exclusive, worldwide, non-transferable, royalty-free, limited license under Playground’s intellectual property rights to use, reproduce, distribute, copy, create derivative works from, and modify the Playground material. This license is free for research and commercial use.

Recommendation

Leave a Comment

Your email address will not be published. Required fields are marked *

Scroll to Top