AFAIK, The newer models for image gen like this OpenAI one, don’t actually use the normal diffusion process (image generates all at once from blurry to finished

AFAIK, The newer models for image gen like this OpenAI one, don’t actually use the normal diffusion process (image generates all at once from blurry to finished), but use transformer architecture where the full final image is generated from top to bottom, as a stream of ‘tokens’. That’s why when you generate an image in chatgpt nowadays, it will start displaying in full resolution from the top pixel row and start loading towards the bottom.