OpenAI (DALL-E 2) — DALL-E 2 and Stable Diffusion — consumer AI image generation

2022

Artists: OpenAI (DALL-E 2), Stability AI / CompVis (Stable Diffusion), Emad Mostaque (Stability CEO) · Year: 2022 · Medium: Text-to-image AI systems — user types text prompt, system generates corresponding image · Substrate: Diffusion models trained on billions of image-text pairs scraped from the public web; consumer-accessible via web interfaces and API · Place: San Francisco, USA / Munich, Germany / London, UK · Institution: OpenAI; Stability AI; CompVis LMU Munich OpenAI released DALL-E 2 in April 2022 as a commercial generative-AI image-generation service, and Stability AI released Stable Diffusion as a fully open-source model in August 2022. The two releases, four months apart, collectively constituted the specific inflection point at which generative-AI image synthesis became genuinely accessible to mass audiences. Where prior AI-image systems (the 2015-19 GAN research era, DALL-E 1 in early 2021, Midjourney's initial releases in mid-2022) had required either specialist technical access or substantial compute resources, DALL-E 2 made high-quality AI-image generation available through a simple web interface, and Stable Diffusion made the underlying weights downloadable and runnable on consumer hardware. The specific significance of this archive's entry is less the commercial platforms themselves than the specific open-source release of Stable Diffusion. The August 2022 release was politically consequential: Emad Mostaque's Stability AI had funded the model's training and then released it publicly without commercial restriction, in explicit contrast to OpenAI's commercial-access-only model. The immediate consequence was that thousands of artists, researchers, and tinkerers began building on Stable Diffusion immediately — fine-tuning models on specific artists' work, building downstream tools (DreamBooth, ControlNet, the vast subsequent ecosystem), and producing specifically-AI-generated artworks that entered galleries, NFT marketplaces, and commercial contexts. The 2022 inflection has had complex consequences that continue to play out. Questions of training-data authorship (models trained on copyrighted imagery without consent), labor-displacement of human illustrators, and the specific cultural-aesthetic homogenization that AI-image tools produce have all become central discussions in the subsequent years. The 2022 moment is the archive's marker for when these questions moved from theoretical to operational. Subsequent AI-image development has continued rapidly through 2023-25 with OpenAI's GPT-4 Vision, Google's Imagen, and numerous other releases. The 2022 specific inflection remains the dominant reference point. Lore The Stable Diffusion release was preceded by specific controversy around the model's training data. The underlying LAION-5B dataset contained images that had been scraped from public web sources without explicit artist consent, and several specific cases (the Getty Images lawsuit, the concerns raised by individual artists whose work had been absorbed into the training distribution) have produced ongoing legal and cultural disputes. The specific political framing of "open-source" AI-image generation has subsequently been contested — open-sourcing the weights does not resolve the training-data-authorship questions — and the discourse has continued to evolve through 2023-25. Sources • OpenAI — DALL-E 2 system card and research paper (April 2022). • Stability AI — Stable Diffusion v1.5 release notes (August 2022). • Gershgorn, Dave — "The Release of Stable Diffusion Marks a Turning Point" (The Verge, August 2022).