(A single style reference image is shown in white inset box)
(A single style reference image is shown in white inset box)
We present StyleDrop that enables the synthesis of images that faithfully follow a specific style, powered by Muse, a text-to-image generative vision transformer. StyleDrop is extremely versatile and captures nuances and details of a user-provided style, such as color schemes, shading, design patterns, and local and global effects. StyleDrop works by efficiently learning a new style by fine-tuning very few trainable parameters (less than 1% of total model parameters), and improving the quality via iterative training with either human or automated feedback. Better yet, StyleDrop is able to deliver impressive results even when the user supplies only a single image specifying the desired style. An extensive study shows that, for the task of style tuning text-to-image models, Styledrop on Muse convincingly outperforms other methods, including DreamBooth and Textual Inversion on Imagen or Stable Diffusion.
StyleDrop generates high-quality images from text prompts of any style described by a single reference image. A style descriptor in natural language (e.g., "in melting golden 3d rendering style") is appended to the content descriptors both at training and synthesis.
StyleDrop generates images of alphabets with a consistent style described by a single reference image. A style descriptor in natural language (e.g., "in abstract rainbow colored flowing smoke wave design") is appended to the content descriptors both at training and synthesis.
StyleDrop is easy to train with your own brand assets and helps you to quickly prototype ideas in your own style. A style descriptor in natural language is appended to the content descriptors both at training and synthesis.
StyleDrop on Muse, a discrete-token based vision transformer, convincingly outperforms in style-tuning over existing methods based on diffusion (Imagen, Stable Diffusion) models.
