Search for: All records

The popularization of Text-to-Image (T2I) diffusion models enables the genera- tion of high-quality images from text descriptions. However, generating diverse customized images with reference visual attributes remains challenging. This work focuses on personalizing T2I diffusion models at a more abstract concept or category level, adapting commonalities from a set of reference images while creating new instances with sufficient variations. We introduce a solution that al- lows a pretrained T2I diffusion model to learn a set of soft prompts, enabling the generation of novel images by sampling prompts from the learned distribution. These prompts offer text-guided editing capabilities and additional flexibility in controlling variation and mixing between multiple distributions. We also show the adaptability of the learned prompt distribution to other tasks, such as text- to-3D. Finally we demonstrate effectiveness of our approach through quantitative analysis including automatic evaluation and human assessment. 

Recent research has made significant progress in text-to-image editing, yet numerous areas remain under explored. In this work, we propose a novel application in the culinary arts, leveraging diffusion models to adjust a range of dishes into a variety of cuisines. Our approach infuses each dish with unique twists representative of diverse culinary traditions and ingredient profiles. We introduce the Cuisine Transfer task and a comprehensive framework for its execution, along with a curated dataset comprising over 1600 unique food samples at the ingredient level. Additionally, we propose three Cuisine Transfer task specific metrics to accurately evaluate our method and address common failure scenarios in existing image editing techniques. Our evaluations demonstrate that our method significantly outperforms baseline models on the Cuisine Transfer task 

The popularization of Text-to-Image (T2I) diffusion mod- els enables the generation of high-quality images from text descriptions. However, generating diverse customized im- ages with reference visual attributes remains challenging. This work focuses on personalizing T2I diffusion models at a more abstract concept or category level, adapting com- monalities from a set of reference images while creating new instances with sufficient variations. We introduce a solution that allows a pretrained T2I diffusion model to learn a set of soft prompts, enabling the generation of novel images by sampling prompts from the learned distri- bution. These prompts offer text-guided editing capabilities and additional flexibility in controlling variation and mix- ing between multiple distributions. We also show the adapt- ability of the learned prompt distribution to other tasks, such as text-to-3D. Finally we demonstrate effectiveness of our approach through quantitative analysis including auto- matic evaluation and human assessment.