Stability AI

Stability AI is a British artificial intelligence company founded in 2019 and headquartered in London. The company became internationally prominent in August 2022 with the public release of Stable Diffusion, an open-weight text-to-image diffusion model trained in collaboration with the CompVis research group at the Ludwig Maximilian University of Munich and Runway ML. Unlike comparable systems from OpenAI and Google that were released only as hosted services, Stable Diffusion's weights were published with a permissive license, allowing anyone with a consumer GPU to run, modify, and fine-tune the model locally. This decision reshaped the generative image ecosystem and produced a large open-source community around image generation.

Products

Stability AI's product portfolio centres on open-weight or commercially licensed generative models across multiple modalities. Stable Diffusion is the company's flagship image generator. The 1.x and 2.x series used a latent diffusion architecture conditioned on text embeddings from a frozen CLIP text encoder. Stable Diffusion XL (SDXL), released in 2023, increased model capacity and image fidelity. Stable Diffusion 3 and Stable Diffusion 3.5, released through 2024, switched to a rectified-flow training objective and a multimodal diffusion transformer (MMDiT) architecture, with variants ranging from 2.5 billion to 8 billion parameters. SD 3.5 Large is positioned as the highest-quality model in the family, while SD 3.5 Large Turbo is optimised for four-step generation suitable for interactive use. Stable Video Diffusion generates short video clips from a still image or text prompt. Stable Video 4D, with its 2.0 release in 2025, extends this to novel-view video synthesis and 4D asset generation for use in 3D graphics pipelines. Stable Audio generates music and sound effects from text prompts. Stable LM and Stable Code were earlier ventures into text and code generation but did not achieve the prominence of the company's image work.

Business model and licensing

Stability AI's commercial strategy combines open releases with a tiered licensing scheme. Hobbyists and small businesses below a revenue threshold use the models for free under a community license; larger businesses purchase an enterprise license. The company also operates a hosted API and a partnership with cloud providers for managed deployment. As of 2025, more than 500,000 model files derived from Stable Diffusion have been published on Hugging Face, including fine-tunes, LoRAs, and merged variants, making it one of the most widely customised generative models in existence.

Corporate history

The company raised significant venture funding in 2022 and 2023, reportedly reaching a valuation above one billion US dollars. It experienced turbulence in 2024 following the departure of founder Emad Mostaque, a restructuring of the leadership team, and the announcement of a strategic refocus toward enterprise customers and specific industry verticals such as film, game development, and advertising. New investment in 2024 from a group led by Sean Parker and James Cameron stabilised the company, and former Weta Digital leader Prem Akkaraju was appointed CEO. Through 2025 the company continued to release new models including SD 3.5 and SV4D 2.0 while shifting commercial emphasis toward enterprise licensing. Stability AI has been involved in several high-profile legal proceedings. Getty Images filed lawsuits in the United States and United Kingdom alleging copyright infringement in the training data; a UK High Court ruling in November 2025 largely went against Getty on the central training-data claims, though some narrower claims survived. The company is also a defendant in class actions filed by groups of artists.

Open release debate

Stability AI's open-weight strategy has been both praised and criticised. Supporters argue that open releases democratise access to generative AI, enable research and downstream innovation, and prevent capability concentration in a small number of cloud providers. Critics, including some safety researchers and copyright holders, argue that open weights make it harder to enforce content policies, attribution, and licensing, and that responsibility for misuse becomes diffuse. The debate has shaped policy discussions including the EU AI Act and national AI strategies in many countries. Stable Diffusion has become widely used in Malaysian creative and marketing industries because it can be run on local hardware without sending content or prompts to a foreign provider — a useful property under the Personal Data Protection Act 2010 (PDPA) and for clients in sectors such as financial services where data residency matters. Advertising agencies, design studios, and in-house creative teams at companies including Maxis, CelcomDigi, Grab Malaysia, AirAsia, and several large GLCs use Stable Diffusion variants for concept art, mood boards, and rapid storyboard iteration. The Malaysian gaming and animation studios clustered in Cyberjaya, Kuala Lumpur, and Penang — including Lemon Sky Studios and Streamline Studios — have integrated Stable Diffusion and ComfyUI-based pipelines into pre-production workflows. Universities including Multimedia University (MMU), The One Academy, and Dasein Academy of Art have introduced Stable Diffusion to their digital art and design curricula, while HRD Corp-funded courses cover prompt engineering and image generation for SMEs. Malaysian policy bodies are engaging with the implications. The National AI Office Malaysia, MDEC, and the Communications Ministry have all raised concerns about deepfakes, election-period synthetic media, and copyright. The Malaysian Copyright Act 1987 was amended in recent years to clarify some aspects of digital reproduction, but the question of whether training generative models on copyrighted Malaysian works constitutes infringement remains unresolved domestically. The Personal Data Protection Commissioner has noted in public guidance that synthetic generation of identifiable individuals may raise PDPA issues even when no real personal data is used as the prompt source. Local AITG SDN BHD partners and other AWS Partner Network members frequently deploy Stable Diffusion models on Amazon Bedrock and SageMaker for Malaysian customers, with inference profiles tagged for Partner Revenue Measurement.

Significance

Stable Diffusion's release accelerated mainstream attention to generative AI in 2022 and 2023, predating but contributing to the conditions in which ChatGPT and Midjourney captured global attention. Whether Stability AI continues to drive frontier image research or settles into a sustainable enterprise role remains an open question into 2026, but the company's influence on the open-weight model ecosystem is already significant.

1. Rombach, R. et al. (2022). High-Resolution Image Synthesis with Latent Diffusion Models. CVPR.
2. Stability AI. (2024). Introducing Stable Diffusion 3.5. stability.ai/news.
3. Stability AI. (2025). Stable Video 4D 2.0 Release Notes. stability.ai.
4. High Court of Justice (England and Wales). (2025). Getty Images v. Stability AI Ltd, Judgment. judiciary.uk.
5. Wikipedia contributors. (2025). Stable Diffusion. en.wikipedia.org.