Im fascinated with the history of image AI as just some algorithms to help clean up noisy photos. Eventually they advanced enough that you just feed it complete noise, tell it what its an image of, and it just “gets rid of the noise”
Yeah thats the mechanism its generating with, and why every generation is different in some way. Theres a lot more cool components to the whole process and history tho. Tying language models into image recognition models, and then the image generation algorithm has an arms race with the image recognition, where the image generation is learning by trying to deceive image recognition, and image recognition is trying to distinguish between generated images and real images its also being fed.
Im fascinated with the history of image AI as just some algorithms to help clean up noisy photos. Eventually they advanced enough that you just feed it complete noise, tell it what its an image of, and it just “gets rid of the noise”
Wait is that how image generation works? Denoising a random noise seed?
Yes! It’s called “stable diffusion”, and it’s how you get a unique image everytime.
Yeah thats the mechanism its generating with, and why every generation is different in some way. Theres a lot more cool components to the whole process and history tho. Tying language models into image recognition models, and then the image generation algorithm has an arms race with the image recognition, where the image generation is learning by trying to deceive image recognition, and image recognition is trying to distinguish between generated images and real images its also being fed.