textual inversion

A personalization method that teaches a frozen 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. a new concept — a particular object, person or artistic style — by adding one invented word (a pseudo-word such as S*) to its vocabulary and learning nothing but that word's embeddingembedding, n.: A list of numbers that stands for a word, picture or other item, arranged so that similar items get similar lists. It lets a computer measure similarity of meaning as distance., after which any prompt containing the word evokes the concept. Starting from typically three to five example images, it optimizes the single vector in the text encoder's input space against the model's ordinary denoising loss while every network weight stays fixed; the authors reported that one such vector was enough for a wide range of concepts. The resulting files are a few kilobytes and can be shared and combined, and the community also trains 'negative embeddings' that encode unwanted features for use in negative promptsnegative prompt, n.: A second prompt listing what you do not want in a generated image — 'blurry, extra fingers, watermark' — which the generator then steers away from.. Fidelity is lower than with weight-updating methods such as DreamBoothDreamBooth, prop. n.: A way to teach an image generator one specific subject — your dog, your face, a product — from a handful of photos, so it can draw that exact subject in new scenes. or 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., because a single token cannot inject what the model cannot already express.