The series of images above started with a badly taken selfie. I fed this image to an LLM (Gemini 3.5) and asked it to "describe the person in the image. The description will be used to create a portrait of the person".
I then copied and pasted the description given by the LLM into a new chat (but same model), with the instruction: "create a potrait of the person described below".
I repeated the steps above, in a chain, feeding the LLM each time the new image generated based on the description of the image created in the previous step.
The loop could continue forever, but I find it more interesting to leave the rest of the possible images to the imagination.