Stability AI

Revision as of 22:08, 27 January 2023 by Alpha5 (talk | contribs)

Founded in 2020 by Emad Mostaque, Stability AI is a community-driven company on the field of artificial intelligence (AI). Its mission is "to build open-source AI projects through collaborative research efforts between leaders and experts in their fields; these include researchers, engineers, developers, PhD candidates, and AI artists across multiple disciplines." [1] The company is building AI tools through an open-source model attempting to democratize AI. [2] This attitude of openness is expressed in their motto, "AI by the people, for the people" (figure 1). [3]

Figure 1: Stability AI motto from their official website. Source: Stability AI

Stability AI trains its AI systems in a cluster of more than 4,000 Nvidia A100 GPUs running in AWS. According to Business Insider, maintaining this system is costly, with the company's operations and cloud costs exceeding $50 million. The company's CEO has stated that R&D will lead to a more efficient model training in the future. [4]

In 2022, Stable Diffusion was released. This AI-assisted image generator model had a great impact due to its open-source nature and low computing requirements. [5] [6]

The company has its main offices in Notting Hill, London, UK, with a team distributed around the globe. [2]

Founder

Emad Mostaque is a computer scientist with a background in tech and finance. [5] In 2005, he received a master's in mathematics and computer science from Oxford University [7]. He worked as an analyst at different hedge funds for 13 years. He founded Symmitree in 2019, a business enterprise with the goal of reducing the cost of technology for poor people. The project failed, but in 2020 he created Stability AI, aiming to build open-source AI projects. [7] [4]

Values and drive

The main values proposed by Stability AI is the support for open-source AI research and collaboration through communities of experts, leaders, and partners around the globe that are developing AI models for image, language, audio, video, 3D, and biology. [2] Their "primary drive is that creativity promotes the advancement and expansion of the human potential to develop breakthrough ideas and convert them into practical solutions to build an inclusive, more communicative, creative future for everyone." [1]

Monetization

Since the technology developed by Stability AI is open-source, it is free to use. However, the company monetizes its AI products and provides consulting services to businesses. [2] [7]

Stable Diffusion was monetized by a platform and API called DreamStudio, an application for text-to-image generation. [2] According to the company's CEO, DreamStudio had more than 1.5 million users. [4]

The company expertise in AI and private supercomputing resources allows it to provide consulting and contracting services to clients, assisting them into making their content more interactive, intelligent, and accessible in the era of artificial intelligence. [2]

Stability AI already has established some partnerships with governments and leading institutions. Its CEO mentioned that they had "negotiated massive deals," expecting "to be profitable at the door versus most money-losing big corporations." [7]

Development highlights

August 2022: Stability AI releases the Stable Diffusion model [8] with different versions—a public demo, DreamStudio, and a full-fat version of the model for download. [9]

October 2022: Announcement of a $101 million funding round led by Coatue, Lightspeed Ventures LLC, and O’Shaughnessy Ventures, with the company being valued around $1 billion. [10] Emad Mostaque mentioned that this new capital will support deploying custom versions of Stable Diffusion at a larger scale, investing in more supercomputing power, and hiring more people. [4]

November 2022: Stability AI selects Amazon Web Services (AWS) as its preferred cloud provider. According to Amazon, "Stability AI will build AI models on compute clusters with thousands of GPU or AWS Trainium chips, reducing training time and cost by 58%. Stability AI will also collaborate with AWS to enable students, researchers, startups, and enterprises around the world to use its open-source tools and models." [11]

Projects

 
Figure 2: Stability AI involvement in different communities and projects. Source: Stability AI.

The company is involved in the funding and development of open-source music- and image-generating models (figure 2). [4]

Stable Diffusion, the first model released, is a text-to-image model that allows people to quickly create stunning art through a text prompt. This system can run on consumer GPUs making it accessible to a greater number of people. [6]

Besides the text-to-image model, Stability AI is working on other projects for future commercialization like audio, language, 3D, and video generation. Dance Diffusion is a model trained on hundreds of hours of existing songs that generates clips of music. [4] This is being developed with Harmonai, a community-driven organization that releases open-source generative audio tools with Stability AI providing financial backing. This system is intended to make music production more approachable and enjoyable to the general public. [8]

Impact

The company already generated a buzz among the AI community and the broader public. Stable diffusion had an immediate impact after release, being downloaded more than 200,000 times. [4] DreamStudio quickly surpassed the mark of over a million registered users, with people from more than 50 countries creating over 170 million images. [12]

This popularity brought also controversy with it due to the open-source and free to use nature of the software and the hands-off approach of the company regarding the content created by the users. While other AI image generators have strict rules to prevent the creation of violent, pornographic or copyright-infringing images, Stability AI's model has only a basic safety filter that can be easily disable by users creating their own version of the software. [13]

Emad Mostaque has mentioned that he is not keen on the idea of content restriction, advocating radical freedom in order to achieve is vision of a democratizes AI that is independent from corporate influence. [13]

Criticisms to the platform can be grouped in three main categories:

1. Offensive content generation.

2. Misuse against personal rights.

3. Copyright infringement.

In December 2022,, Stability AI announced that artists are able to opt out of the training data for Stable Diffusion 3, [14] something that the artists' advocacy group Spawning was pushing for. [15]

External links

Stability AI API platform

The API allows developers to easily integrate the image generation system into their projects. It can be accessed here. [16]

DreamStudio

AI Image generator based on the open-source model, Stable Diffusion available here.

Researcher Access Form

Access early version of the models supported by Stability AI or other grants here.

References

  1. 1.0 1.1 Stability AI. Press. Stability AI. https://stability.ai/press
  2. 2.0 2.1 2.2 2.3 2.4 2.5 Stability AI. FAQ. Stability AI. https://stability.ai/faq
  3. Dilmegani, C (2022). Stability AI: Does open-sourcing democratize generative AI? AIMultiple. https://research.aimultiple.com/stability-ai/
  4. 4.0 4.1 4.2 4.3 4.4 4.5 4.6 Wiggers, K (2022). Stability AI, the startup behind Stable Diffusion, raises $101M. TechCrunch. https://techcrunch.com/2022/10/17/stability-ai-the-startup-behind-stable-diffusion-raises-101m/?guccounter=1
  5. 5.0 5.1 Vallance, C (2022). "Art is dead Dude" - the rise of the AI artists stirs debate. BBC. https://www.bbc.com/news/technology-62788725
  6. 6.0 6.1 Stability AI. Stable Diffusion launch announcement. Stability AI. https://stability.ai/blog/stable-diffusion-announcement
  7. 7.0 7.1 7.2 7.3 Cai, K (2022). Startup behind AI image generator Stable Diffusion is in talks to raise at a valuation up to $1 billion. Forbes. https://www.forbes.com/sites/kenrickcai/2022/09/07/stability-ai-funding-round-1-billion-valuation-stable-diffusion-text-to-image/
  8. 8.0 8.1 Yalalov, D (2022). StabilityAI announced AI music generator Harmonai based on Dance Diffusion model. Metaverse Post. https://mpost.io/stabilityai-announced-ai-music-generator-harmonai-based-on-dance-diffusion-model/
  9. Vincent, J (2022). Anyone can use this AI art generator — that’s the risk. The Verge. https://www.theverge.com/2022/9/15/23340673/ai-image-generation-stable-diffusion-explained-ethics-copyright-data
  10. Metinko, C (2022). Stability AI newest addition to unicorn stable. Crunchbase News. https://news.crunchbase.com/ai-robotics/stability-ai-unicorn/
  11. Bindal, A (2022). Stability AI builds foundation models on Amazon SageMaker. Amazon Web Services. https://aws.amazon.com/blogs/machine-learning/stability-ai-builds-foundation-models-on-amazon-sagemaker/
  12. Stability AI (2022). Stability AI announces $101 Million in funding for open-source artificial intelligence. PR Newswire. https://www.prnewswire.com/news-releases/stability-ai-announces-101-million-in-funding-for-open-source-artificial-intelligence-301650932.html
  13. 13.0 13.1 Roose, K (2022). A coming-out party for generative A.I., Silicon Valley’s new craze. New York Times. https://www.nytimes.com/2022/10/21/technology/generative-ai.html
  14. Goldman, S (2022). Artists can opt-out of next Stable Diffusion training. VentureBeat. https://venturebeat.com/ai/stability-ai-to-honor-artist-opt-out-requests-for-stable-diffusion-3/
  15. Edwards, B (2022). Stability AI plans to let artists opt out of Stable Diffusion 3 image training. Ars Technica. https://arstechnica.com/information-technology/2022/12/stability-ai-plans-to-let-artists-opt-out-of-stable-diffusion-3-image-training/
  16. Stability AI. Stability’s API platform. Stability AI. https://stability.ai/blog/api-platform-for-stability-ai