DreamBooth

A fine-tuningfine-tuning, n.: Taking an AI model that has already learned a lot and training it a little more on a smaller, focused set of examples so it gets better at one job or behaves a certain way. method that teaches a text-to-image modeltext-to-image model, n.: An AI system that creates a picture from a written description, such as 'a fox reading a newspaper, in watercolor', usually within seconds. one particular subject — a certain dog, face or product — from typically three to five photos, after which prompts such as 'a [V] dog on the moon' render that subject faithfully in new settings, poses and styles. The photos are paired with prompts in which a rarely used token, here [V], serves as the subject's name beside its class word ('a [V] dog'), and the whole model is updated. So that the model neither forgets what ordinary members of the class look like nor turns every dog into the subject (language drift), a class-specific prior-preservation loss also trains it on images it generated beforehand from the plain class prompt. Full DreamBooth produces a multi-gigabyte model per subject; community practice soon combined it with LoRAlow-rank adaptation, n.: A cheap way to customize a big AI model: instead of changing all its billions of numbers, you train a small add-on of a few million numbers that slightly adjusts how it behaves. to produce small add-on files. Because it can depict real people realistically from a few photos, it raised immediate concerns about non-consensual imagery and deepfakesdeepfake, n.: A fake but realistic video, photo or voice recording made with AI that puts words in a real person's mouth or shows them in a scene that never happened..