Text-to-Image Diffusion Models for Product Image Generation

Ecomtent focuses on developing vertical-specific generative AI models for e-commerce brands, offering a self-service tool to allow customers to generate an unlimited number of high-quality images in any scenario. To this end, we leverage a textto- image model which will be trained to recontextualize any image via a simple text prompt. In particular, we seek to explore additional data-type, beyond just text-prompts and images, that the model can be trained on in order to improve the fidelity of its output. The successful completion of this project will enable Ecomtent to offer their customers a model whose output is realistically recontextualized, yet also very faithful to the original image’s details. This will provide Ecomtent a competitive edge in the content-generation industry.

Faculty Supervisor:

Kirill Serkh;Sushant Sachdeva

Student:

Partner:

Ecomtent

Discipline:

Computer science

Sector:

Professional, scientific and technical services

University:

University of Toronto

Program: