Diffusion Model Denoising Vulnerabilities to Exploit?

Question

For the past year I have been seeing perturbation tools that target generative models, aiming to corrupt the images these models output (text-to-image). I am basing my study on them.

I have decided to specifically target diffusion models because of their popularity. Right now, I am trying to learn about how diffusion models work and how they can generate images, because I am trying to look for weaknesses and vulnerabilities to exploit for my custom adversarial tool. I'm reading up on the original paper of DDPM but it's honestly overwhelming me with how technical it is. I honestly don't know where to look for help and I don't trust asking AI to help me.