Name My Lipstick

This my 1st project for the KaggleX 2023 cohort 3 mentorship program.

The goal is to style transfer lipstick naming of a famous cosmetic brand based on the selected color using LLMs.

The dataset was built from the image below using image and ocr processing

More information can be found on the blog post 💄 Name-My-Lipstick: Where AI Crafts Stunning Names and Fabulous Ad Images

Try on https://namemylipstick.streamlit.app/

Features

• generate a lipstick name using zero-shot, few-shot and many-shot prompt approaches based on the select color using GPT-3.5 and other models;
• generate an ad image for the generated lipstick name using DALLE2 and DALLE3.

Architecture

Work

• extracted data from an image with the 2018 lipstick name and color of a famous cosmetic brand;
• preprocessed the collected data to feed few-shot and many-shot prompts;
• implemented a similirity color search based on CIELab to feed the few-shot prompt;
• created a UI using Streamlit.
• prompts
• api calls
• generate image from name + rgb color
• add dalle3
• remove langchain dependency to better follow openai updates