Raw LLM Responses

3:30 Friendly reminder that AI image generators also have a problem because of how they are trained, that they cannot conceive of ideas it has never seen. It can combine things it has seen, but for example, if you ask for it to show an image of a glass of wine that is completely full to the point of over spilling, it'll do everything in its power to keep the wine at the halfway point and even make the wine clip through the glass to spill it, instead of just filling the glass, because images of a glass of wine that is completely full are VERY rare on the internet and most likely have never been introduced to their training data. The same applies if you want to make creature designs that are more than just lazy combinations of animals, the AI just is unable to comprehend it. It is also unable to comprehend the lack of a body part most of the time, barely able to understand asymmetry, and is very inconsistent with object consistency if its shape is partially concealed behind another object, like for example, AI image generators sometimes begin with a tail curving towards the thighs, only for the next part to come out on the ankle, or the tail to fuse the tip into itself. The AI doesn't learn in the conventional sense. It is shaped, forged into learning. Artists learn, adapt and shape themselves with their experiences. My own art style was heavily influenced by PopCross Studio here on YouTube, but it is still very distinct because I draw on paper, not on a computer. Trying to adapt his colouring style to paper was how I managed to learn and adapt my own art style.