The paper presents a method called "latent transparency" that enables large-scale pretrained latent diffusion models to generate transparent images and multiple transparent layers. The method encodes alpha channel transparency into the latent manifold of a pretrained latent diffusion model, preserving the production-ready quality of the model by regulating the added transparency as a latent offset with minimal changes to the original latent distribution. The approach is trained using a human-in-the-loop scheme, collecting 1M pairs of transparent image layers. The method can be applied to different open-source image generators and various conditional control systems, achieving applications like foreground/background-conditioned layer generation, joint layer generation, and structural control of layer contents. User studies show that 97% of users prefer the natively generated transparent content over ad-hoc solutions like generating and then matting, and the quality of the generated transparent images is comparable to real commercial transparent assets like Adobe Stock. The paper also discusses related work, including image matting and image harmonization, and provides a detailed methodological explanation, experimental results, and a conclusion.The paper presents a method called "latent transparency" that enables large-scale pretrained latent diffusion models to generate transparent images and multiple transparent layers. The method encodes alpha channel transparency into the latent manifold of a pretrained latent diffusion model, preserving the production-ready quality of the model by regulating the added transparency as a latent offset with minimal changes to the original latent distribution. The approach is trained using a human-in-the-loop scheme, collecting 1M pairs of transparent image layers. The method can be applied to different open-source image generators and various conditional control systems, achieving applications like foreground/background-conditioned layer generation, joint layer generation, and structural control of layer contents. User studies show that 97% of users prefer the natively generated transparent content over ad-hoc solutions like generating and then matting, and the quality of the generated transparent images is comparable to real commercial transparent assets like Adobe Stock. The paper also discusses related work, including image matting and image harmonization, and provides a detailed methodological explanation, experimental results, and a conclusion.