Just days after the release of version 2.0, there is a new minor update with the Stable Diffusion v2.1 released December 7 2022.
According to the official announcement, Stable Diffusion v2.1 will bring improvements in training and dataset that should address some of the communities complaints around version 2.0, and is touted to improve the overall quality.
What’s new in Stable Diffusion version 2.1
Negative Prompts
Negative prompts are now supported in DreamStudio by appending “|<negative prompt>:-1” to the prompt.
For instance, appending: |disfigured, ugly:-1.0, too many fingers:-1.0 occasionally fixes the issue of generating too many fingers. 
Extreme aspect ratios
The model also has the power to render non-standard resolutions to give you beautiful vistas and epic widescreen imagery.
Filters
The development team say they’ve listened to their community, and adjusted the filters.
Whilst filters in version 2.0 still stripped out adult content, it was less aggressive, and this reduced the number of false positives it detected.
The Stable Diffusion version 2.1 model has been released with this updated setting, providing a model which captures the best of both worlds.
It can render beautiful architectural concepts and natural scenery with ease, and yet still produce fantastic images of people and pop culture too.
The new release delivers improved anatomy and hands and is much better at a range of incredible art styles than SD 2.0.
Read the full blog post about the release from Stability AI
Try Stable Diffusion version 2.1 now:
You can try v2.1 right now on DreamStudio at https://beta.dreamstudio.ai/
First look at Stable Diffusion v2.1
I tried it out using the new Dreambot (what a seriously cool feature in the official Discord Server), and the results were awesome. Not only did the Dreambot add in a whole heap of amazing negative prompts just by adding the term “negative_prompt”, but it also created the images at a massive upscale, and the results were pretty impressive.
I used this prompt for our first test:
a brown bear wearing a bowler hat and dressed in a tuxedo playing a double bass, artwork trending, hyper realistic, pencil drawing, steps:50 number:4 negative_prompt:ugly, tiling, poorly drawn hands, poorly drawn feet, poorly drawn face, out of frame, extra limbs, disfigured, deformed, body out of frame, blurry, bad anatomy, blurred, watermark, grainy, signature, cut off, draft seed:3805660482
and this is what I got:
Yeah, there’s still a few extra fingers on a couple of the images, but the raw quality was excellent, although a few elements of the prompt got excluded (particularly the double bass!).
For the second test, I tried this prompt out:
concept art of a greek goddess gathering plants and herbs in a luscious garden as nymphs watch her, hyper realistic, artwork trending, steps:50 number:4 seed:3960881049
As you can see, there’s still a lot of work to do on fingers and faces, so I’m not sure how well the negative prompts which were automatically included using Dreambot were working. It will be interesting to see how the community reacts to the updates.
Content Keywords: stable diffusion v2, stable diffusion version 2.1, stable diffusion 2.1, dreambot, stable diffusion update, stable diffusion minor update, stable diffusion negative prompt, stable diffusion seed, stable diffusion negative keyword, version 2.1 ai art, stability.ai version 2.1, is stable diffusion better than dall-e, is stable diffusion better than midjourney, is stable diffusion v2 better than v1, is stable diffusion version 1 better than version 2, when will stable diffusion v2.1 be released, stable diffusion improvements in version 2, stable diffusion version 2 updates, stable diffusion open source ai version 2, stable diffusion ai v2.1, stable diffusion artificial intelligence v2.1
