If you want to add new stuff to Stable Diffusion - you don't have to retrain the main model from scratch. You can train it only on a few hundred images at a ti

If you want to add new stuff to Stable Diffusion - you don't have to retrain the main model from scratch. You can train it only on a few hundred images at a time and then add the resulting model as an extension or merge it into the main model. People train their models on a single celebrity or a single artist that way. Other types of AI could be trained in a similar way.