AI Image Generators in Art Class

Many widely used image generators are diffusion models or close relatives of them. They are trained on very large collections of images paired with text descriptions, much of it gathered from the web. Given a prompt, a diffusion model starts from random noise and refines it step by step into an image that matches the text. Well-known tools include OpenAI's DALL-E and the image generation in ChatGPT, Midjourney, Stable Diffusion from Stability AI, and Adobe Firefly. Adobe says Firefly is trained on licensed content such as Adobe Stock, plus public domain material. In art class, these tools raise questions that make good lessons. On creativity: is writing a prompt and choosing among the outputs a form of authorship? Students can compare the decisions they make when painting with the ones they make when prompting. On critique: generated images often fall back on clichés, glossy styles and stereotypes about people, and students can learn to spot these. On copyright: the US Copyright Office has said that material generated by AI without enough human authorship cannot be registered. In its 2023 decision on the comic 'Zarya of the Dawn', it protected the comic's text and arrangement, but not the individual images made with Midjourney. Artists have also sued AI companies over the use of their work in training data, and several cases are ongoing. Attribution is a practical classroom rule: label AI-assisted work, name the tool, keep the prompts, and describe what the student changed. Safety depends on age. Many consumer generators require users to be 13 or older, some 18, and filters are imperfect. Schools should use approved tools with education accounts, avoid uploading students' faces or personal photos, and review the privacy terms. A common misconception is that the AI stores existing images and pastes them together. It learns statistical patterns, although it can sometimes produce close copies of images that appeared many times in its training data.