I'd like something that explains _how_ diffusion models work, their probabilistic nature, how they're trained on images/text and maybe even a high level description of what the different parts of a setup in, say, ComfyUI are doing (LoRA, VAE, CLIP, UNet models).

These things hardly matter for artists. Most artists don't even know how a Gaussian blur filter works and they can still produce astonishing artworks.