Again; nope. Vague crap definition by whom? Somebody on TicTok? Wasn't me so don't try and put words in my mouth.

"But most people would not be able understanding the signification of this text ! same like placing a mathematical Formula to explain the algorithms"

Huh? the signification of this text. Really? Reminds me of the Homer Simpson quote. "smrt"

Diffusion models are built by repeatedly adding noise to an image until the training image is beyond recognition. The reverse generates an image.

Example is a photo of a cat. The cat image has a predetermined amount of random noise added and the label ( the text part of the model) says it a cat. There is more noise added and the label is cat. More noise is added and the image is labeled cat. Soon there is only static noise in the training image and it is labeled cat. That is your "model". A set of rules for denoising an image. To get an image you type c a t and a random noise is generated and the denoising starts. The operator (you) determines the strength of the denoising logarithm and the number of iterations it makes in an attempt to create the request. The result, for better or worse; is "cat"

That's it in a simplified nutshell as diffusion models now stand. There is a massive legal mess happening around this method as we have all heard. Well the method is changing. GANs are being used to train now. General Adversarial Networks. This is closer to autonomous driving training than current diffusion models, but that is beside the point and a different conversation yet again.

I'm not responding any more to this. The subject is drifting off topic into personal attack which is against the TOS.