On December 7, 2022 Stability.AI has released Stable Diffusion v2.1, a new version of its open-source AI image generator Stable Diffusion. This version has more data and training, and fewer restrictions than Stable Diffusion v2.0. The company has also launched a new version of Dream Studio with “negative prompts” working better than with v2.0.
What’s new in Stable Diffusion v2.1 and DreamStudio
- a brand new text encoder – OpenCLIP from LAIO – that offers deeper range of expression
- a new prompting style
- less aggressive filtering of the dataset
- higher image quality in terms of architecture, interior design, wildlife and landscape
- better “negative prompts”
In the previous version 2.0 humans were less represented due to the NSFW filter used to remove the adult content. In v2.1 the NSFW filter is less restrictive so it can better generate humans.
Stable Diffusion v2.1 can render wonderful architectural concepts and deliver phenomenal pictures of individuals and pop culture items as well.
Learn more:
- Stability AI GitHub for the model and the release notes
- beta.dreamstudio.ai to create a DreamStudio account
- Hugging Face for weight and model cards associated with the Stable Diffusion v2-1 model
