Text-to-image personalization

<p>Text-to-Image personalization is a task in <a href="/facts/Deep_learning/JLuwD3ea">deep learning</a> for <a href="/facts/Computer_graphics/lapBE8wv">computer graphics</a> that augments pre-trained <a href="/facts/Text-to-image_model/ez6APLzp">text-to-image</a> <a href="/facts/Generative_model/JUgExNIP">generative models</a>. In this task, a generative model that was trained on large-scale data (usually a <a href="/facts/Foundation_models/67pefJcT">foundation model</a>), is adapted such that it can generate images of novel, user-provided concepts. These concepts are typically unseen during training, and may represent specific objects (such as the user's pet) or more abstract categories (new artistic style or object relations).
</p><p>Text-to-Image personalization methods typically bind the novel (personal) concept to new words in the vocabulary of the model. These words can then be used in future <a href="/facts/Prompt_engineering/SVE53ytb">prompts</a> to invoke the concept for subject-driven generation, <a href="/facts/Inpainting/5fHD96N8">inpainting</a>, <a href="/facts/Neural_style_transfer/kDD1jdfC">style transfer</a> and even to correct <a href="/facts/Biases_in_AI/vHaHsIdV">biases</a> in the model. To do so, models either optimize <a href="/facts/Word_embedding/7uRcBPqo">word-embeddings</a>, <a href="/facts/Fine-tuning_(deep_learning)/9eRVyFNy">fine-tune</a> the generative model itself, or employ a mixture of both approaches.
</p>

Text-to-image personalization open-in-new

Text-to-image personalization