Datawhale open source
Open Source: Image Generation Model, Source : Heart of the Machine
Open source is always nirvana.
Since the initial release of the text-to-image open source model Stable Diffusion, DreamStudio has been the primary interface for new models and features from Stability AI. To date, users have created millions of images with the help of DreamStudio.
Recently, DreamStudio has also been upgraded to SDXL, enabling more detailed images and compositions than its predecessor, Stable Diffusion 2.1, and the ability to create descriptive images with shorter prompts.
For now, DreamStudio can already generate such an image, the effect is not inferior to Midjourney 5.1:
Although DreamStudio provides an image generation method with almost no threshold, it also has a condition: payment. The pricing for the SDXL version is even higher.
Moreover, for some technically qualified developers, they also hope to carry out some application extensions on the basis of DreamStudio.
Maybe he heard the voices of netizens? Just now, Stability AI released the open source version of DreamStudio - StableStudio.
Open source address: https://github.com/Stability-AI/StableStudio
"We believe that the best way to expand technology implementation is through open, community-driven development, rather than privatized iterations of closed-source products." Stability AI said.
In the company's plans, StableStudio's goal is to collaborate with the wider community to build a world-class user interface for generative AI that gives users full control. While DreamStudio will remain an implementation of StableStudio hosted by Stability, the ultimate goal is to foster a project that can outperform any development by a single company.
Specifically, what is the difference between StableStudio and DreamStudio?
Stability AI made some tweaks to make the project more community-friendly. include:
Removed DreamStudio exclusive branding;
All online API calls have been replaced by the plugin system, users can easily replace the backend;
Account features specific to Stability, such as billing, API key management, etc., have been removed.
extensions
Originally conceived as an animation studio for Disco Diffusion, DreamStudio's focus shifted to image generation following the release of Stable Diffusion in 2022.
Inspired by the popular dialogue model this year, Stability AI released the open source LLM StableVicuna in April. StableStudio, the open source version of DreamStudio, will combine with StableVicuna to launch a chat interface.
StableVicuna dialog example.
"Stable Vicuna" generated by Stable Diffusion XL.
According to Stability AI, StableStudio may update the following features in the future:
Local inference via WebGPU
Local inference via stable-diffusion-webui
desktop installation
ControlNet Tools
Installation and use
The StableStudio project is live on GitHub, which includes a quick installation method.
Project address: https://github.com/Stability-AI/StableStudio
To use StableStudio, users need to install Node.js and Yarn first. Then clone the StableStudio repository as follows
git clone https://github.com/Stability-AI/StableStudio.git
cd StableStudio
yarn
yarn devBy default, StableStudio runs on localhost:3000. Users need to prepare an API key to use the default Stability API plugin; if not, they need to create an account on DreamStudio and get the key.
Image Generation Mainstream Models
Whether it is DreamStudio or the open source version of StableStudio, the basic model behind it is Stable Diffusion, an image generation model that is currently popular. Stable Diffusion is widely used for its ability to quickly generate high-quality images on consumer-grade GPUs.
The technology of the Stable Diffusion model was originally derived from a paper "High-Resolution Image Synthesis with Latent Diffusion Models" published in CVPR 2022. In order to train Stable Diffusion, the research team used 4,000 A100 Ezra-1 AI super clusters, which were provided by Stability AI.
In August 2022, Stability AI announced the release of Stable Diffusion and made it open source. In November, they released Stable Diffusion 2.0. Now, Stable Diffusion has replaced GAN as the mainstream model in the field of image generation.
As an open source version of the product platform DreamStudio, StableStudio replaces all online API calls with a plug-in system, allowing users to easily replace the backend, which will further promote the development of the field of image generation.
Reference link: https://stability.ai/blog/stablestudio-open-source-community-driven-future-dreamstudio-release
Dry goods learning, like three times ↓