Stable Diffusion in brand workflows: generating isn't the problem

Stable Diffusion had been public for two weeks. Generating a thousand variations of an image was trivial and easy. The legal team reviewed the pilots and rejected every one. Image rights and brand compliance were the bottleneck, not visual quality.

The early pilots generated images of products and people. That created liability nobody wanted to own legally. What was the training data source? Who did we need to credit? Whose likeness did we just generate with the model.

We scoped narrowly and carefully. The model could generate still life: textures, backgrounds, environments, abstract compositions. Products and people stayed outside the generative scope. That is where human work remained essential.

That clear boundary on scope largely eliminated the legal risk. A background texture or procedurally-generated landscape felt like tool output rather than a derived work requiring copyright clearance or carrying image rights liability.

With that limit clearly drawn, the content team could move fast and produce variants. They quadrupled the number of campaign variants they could produce without waiting for photography or opening a single legal case.

The pattern held for other image generation work in production. There is always a boundary between what you can generate safely and what requires human judgment or rights clearance and legal review.