How an image is made
Diffusion models learn to remove noise from images. When generating, they start with pure noise and turn it step by step into an image that matches the text. A language model “explains” the text to them.
Problems
- Copyright: the models learned from images on the internet without the authors' consent, and lawsuits are under way.
- Deepfakes: fake videos and voices of well-known people are misused for scams. If you get an unusual request for money, even in the voice of someone close to you, verify it through another channel.
- Labelling: the European AI Act requires artificially created content to be recognisable.
A creative tool
Designers and game developers use generative AI for quick drafts and ideas. For photographers and illustrators it is both competition and a new tool. You will find the earlier history of images in the article Photography.