The rise and pivot of Stability AI

Open-source growth and technical foundations

Stability AI captured the generative image market by releasing open-source models that run on consumer-grade hardware. Emad Mostaque and Cyrus Hodes founded the company in late 2020. Their flagship product, Stable Diffusion, arrived in August 2022. This model allows users to generate detailed images from text descriptions. Most notably, Stable Diffusion based tools created 80% of the 15 billion images produced by text-to-image algorithms since 2022. The model runs on consumer GPUs with as little as 2.4 GB VRAM. This technology changed how artists use AI. Users generate 2 million images daily through official channels. Mostaque stated the model has more than 10 million users across all channels. The development involved researchers from the CompVis Group at LMU Munich and Runway. Stable Diffusion uses latent diffusion, a kind of deep generative artificial neural network. The VAE encoder compresses the image from pixel space to a smaller dimensional latent space. The U-Net block denoises the output from forward diffusion backwards to obtain a latent representation. Training the model required 256 Nvidia A100 GPUs on Amazon Web Services for 150,000 GPU-hours at a cost of $600,000. The model was trained on three subsets of LAION-5B, including laion2B-en, laion-high-resolution, and laion-aesthetics v2 5+. The original model produced images at 512×512 resolution, though the SD XL version introduced 1024×1024 resolution.

Leadership shifts and financial pressure

Internal instability and massive cloud debts forced a fundamental change in the company’s management. Emad Mostaque, a British-Bangladeshi former hedge fund manager with an Oxford degree, resigned as CEO in March 2024 to pursue decentralized AI. His departure followed reports that three of the five original researchers left the company. Stability AI also faced lawsuits regarding training data, including a case from Getty Images. The company owed nearly $100 million to cloud providers. The company raised $101 million in a seed round led by Coatue Management and Lightspeed Venture Partners at a $1 billion valuation in 2022, but it has since struggled to raise additional funds due to the lack of a business model. In Q1, the company generated less than $5 million in revenue while losing over $30 million. The company currently employs approximately 170 people. Stability AI also released AI models to generate audio and video using text prompts. During the leadership transition, the board appointed Shan Shan Wong and Christian Laforte as interim co-CEOs. Mostaque stated he would not beat centralized AI with more centralized AI. Does this financial pressure prevent further open-source development?

The enterprise pivot under Akkaraju

The company now prioritizes commercial viability through a leadership overhaul and enterprise-focused products. Prem Akkaraju, former head of Weta Digital, joined as CEO. The board added Sean Parker as chairman and included new members like Dana Settle and Colin Bryant. Akkaraju also invested in the company. The new funding includes $80 million from investors such as Eric Schmidt. Stability AI is moving away from Mostaque’s decentralized vision toward commercialized generative AI products. The company provides Stable Diffusion Reimagine, which uses an image encoder instead of a text encoder to create variations of an existing image. This Clipdrop tool allows users to upload an image to create new designs without complex prompts. The model adds noise to generate variation after the encoder processes the image. This approach produces similar-looking images with different details and compositions. The generator does not use a single pixel sourced from the original image. Stable Diffusion 3.0 uses a Rectified Flow Transformer. You should watch if this shift satisfies the original community of open-source developers. The company released Stable Cascade as an option for researchers on GitHub.