Stability AI Profile picture
Dec 7, 2022 7 tweets 4 min read Read on X
We’re happy to release Stable Diffusion, Version 2.1!

With new ways of prompting, 2.0 provided fantastic results, and 2.1 supports the new prompting style, but also brings back many of the old prompts too!

Link → stability.ai/blog/stabledif… Stable Diffusion v2.1-768 Credit: KaliYuga_ai
The differences are more data, more training, and a less restrictive filtering of the dataset. The dataset for v2.0 was filtered aggressively LAION’s NSFW filter, making it a bit harder to get similar results generating people.
We listened to our users & adjusted the filters. Adult content is still stripped out but less aggressively so it cuts down on too many false positives. We fine-tuned v2.0 with this updated setting, giving us a model which hopefully captures the best of both worlds! Prompt: a portrait of a beautiful blonde woman, fine - art p
The model also has the power to render non-standard resolutions. That helps you do all kinds of awesome new things, like works with extreme aspect ratios that give you beautiful vistas and epic widescreen imagery. #StableDiffusion2
The community noticed that “negative prompts” worked wonders with 2.0 & they work even better in 2.1! Negative prompts allows a user to tell the model what not to generate.

Negative prompts are now supported in @DreamStudioAI by appending “|<negative prompt>:-1” to the prompt. A Hyperrealistic photograph of German architectural modern h
For more details about accessing and downloading v2.1, please check out the release notes on our GitHub: github.com/Stability-AI/s…

Additionally, you can try v2.1 right now on @DreamStudioAI at beta.dreamstudio.ai
We know open is the future of AI and we’re committed to developing current & future versions of @StableDiffusion in the open. Expect more models and releases to come fast and furious and some amazing new capabilities as generative AI gets more and more powerful in the new year. Image Prompt: A hyperrealistic painting of an astronaut insi

• • •

Missing some Tweet in this thread? You can try to force a refresh
 

Keep Current with Stability AI

Stability AI Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!

PDF

Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @StabilityAI

Dec 6, 2022
The leading architecture magazine @Dezeen interviewed Stability AI’s Creative Director Bill Cusick @williamcusick about the uses of AI image generation in architecture, here are the highlights:

“AI is the foundation for the future of creativity."

Link → dezeen.com/2022/11/16/ai-…
AI is enabling new forms of creativity for architectural design. Cusick likens designing with AI to the experience of playing chess, in that it takes a short amount of time to learn but far longer to master.
AI has great potential as a tool for early stages of a project. Cusick created architectural sketches using DreamStudio to illustrate the approach best articulated by Andrew Kudless, @MatsysDesign

"It's meant to capture a vision of a project quickly."
Read 5 tweets
Nov 24, 2022
We are excited to announce the release of Stable Diffusion Version 2!

Stable Diffusion V1 changed the nature of open source AI & spawned hundreds of other innovations all over the world. We hope V2 also provides many new possibilities!

Link → stability.ai/blog/stable-di…
Text-to-Image

The V2.0 release includes robust text-to-image models trained using a new text encoder developed by LAION with support from Stability AI, which greatly improves the quality of the generated images compared to earlier V1 releases.
The text-to-image models can generate images with default resolutions of both 512x512 & 768x768.

The models are trained on a LAION-5B aesthetic subset, which is then further filtered with LAION’s NSFW filter.

Examples of images produced using V2.0, at 768x768 image resolution:
Read 9 tweets
Nov 10, 2022
Our very own @RiversHaveWings has trained a latent diffusion-based upscaler.

What does this mean and how does it work? (1/5)
The upscaler is itself also a diffusion model. It was trained on a high-resolution subset of the LAION-2B dataset. Being a 2x upscaler, it can take the usual 512x512 images obtained from Stable Diffusion and upscale it to 1024x1024. (2/5)
Like Stable Diffusion, the upscaler is itself a latent diffusion model, a diffusion model that operates in a compressed "latent" space, which is "decoded" into a full-resolution image. The upscaler uses the same encoder/decoder & therefore works in the same latent space. (3/5) Image
Read 5 tweets
Oct 29, 2022
We recently released two new fine-tuned decoders (trained by the amazing @rivershavewings) that improves the quality of images generated by @StableDiffusion.

Read on to see what this means and how you can try it out yourself! ↓
What does "fine-tuned decoders" even mean? Well, basically, Stable Diffusion actually is a diffusion model that operates in a compressed space, which is then "decoded" into a full-resolution image. This decoder itself is also a trained neural network.
So "fine-tuned decoder" simply means that this is a new set of decoder neural network parameters (i.e. changing the part circled in blue, diagram from @huggingface) after further training on a new dataset.
Read 8 tweets

Did Thread Reader help you today?

Support us! We are indie developers!


This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Don't want to be a Premium member but still want to support us?

Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal

Or Donate anonymously using crypto!

Ethereum

0xfe58350B80634f60Fa6Dc149a72b4DFbc17D341E copy

Bitcoin

3ATGMxNzCUFzxpMCHL5sWSt4DVtS8UqXpi copy

Thank you for your support!

Follow Us!

:(