About a dozen companies train foundational AI image models from scratch; the rest build products on top of them

The AI image generation ecosystem is built on a four-layer stack: model-first companies (frontier builders with mass-market products), model-only companies (building and distributing foundational models via open source or APIs), product-first builders (companies that evolved from using third-party models to building their own), and orchestrators (platforms routing access to others' models).

Only about a dozen companies worldwide train their own models from scratch and produce mass-market products; the rest, hundreds of apps and platforms, are built on top of those. Owning a foundational model controls cost structure, latency, deployment, differentiation, and independence—without it, companies compete on UX, curation, or price but remain tenants on someone else's infrastructure.

Adobe operates as a hybrid: both a model builder (Firefly Image Model 4 and Ultra with 18B+ assets) and a major orchestrator, allowing users to pick Google Imagen 3, OpenAI GPT Image, and others alongside its own models.