By incorporating differentiable constraint evaluation functions directly into stable diffusion’s iterative sampling, this approach enforces physical and functional requirements in generative tasks.