Personalizing Text-to-Image Generation via Aesthetic Gradients
NeurIPS, Workshop on ML for Creativity and Design, 2022
Abstract
This work proposes Aesthetic Gradients, a method to personalize diffusion models by guiding the generative process towards custom aesthetics defined by the user from a set of images. The approach is validated with qualitative and quantitative experiments, using the recent stable diffusion model and several aesthetically-filtered datasets.
The study introduces Aesthetic Gradients for personalizing diffusion models, advancing user-interactive technology and impacting fields like digital art, design, and marketing.
Where does it apply?
It can be used in digital art and design to create unique, AI-generated visuals guided by custom aesthetics. In marketing, it can produce customized ad content in line with brand aesthetics.
The entertainment industry can use it to generate personalized visual effects or graphics. It can also be used in personal software applications for user-customizable aesthetic outcomes.
Why does it matters?
The proposed aesthetic gradients matter as they allow for a new level of personalization in the generative systems.
With this method, a user can guide the generative model towards custom aesthetics defined by a chosen set of images. This enables the creation of unique, personalized AI-generated content, enhancing the usability and applications of such models in fields like digital art, design, marketing, and entertainment.
Additionally, by providing both qualitative and quantitative validation, this work advances our understanding and effectiveness of user-guided and aesthetically-informed machine learning models. Thus, it holds significant implications for the future of creative AI and user-interactive technology.
Personalizing Text-to-Image Generation via Aesthetic Gradients
NeurIPS, Workshop on ML for Creativity and Design, 2022
Other publications
Adversarial Machine Learning: Bayesian Perspectives
AI in drug development: a multidisciplinary perspective
How do tuna schools associate to dFADs? A study using echo-sounder buoys to identify global patterns
A Bayesian approach to enhance defense against security threats in Machine Learning systems, applicable across cybersecurity, unsupervised learning and natural language processing.
The review explores the use of AI, particularly Bayesian approaches, in speeding up and reducing the cost of drug development across all stages.
Machine Learning to analyze tuna behavior around Fish Aggregating Devices, providing insights to enhance sustainable fishing, marine biology research and conservation policies.