Reproduce “Embedding an Ethical Mind: Aligning Text-to-Image Synthesis via Lightweight Value Optimization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Xingqi Wang, Xiaoyuan Yi, Xing Xie, Jia Jia
LiVO adds a learned value encoder to steer text-to-image generation according to selected principles. It combines an encoded principle with the prompt representation and trains the encoder using paired preferred and dispreferred images, while keeping Stable Diffusion frozen. On horror prompts, its toxicity ratio is 1.03, versus 77.38 for Stable Diffusion; this classifier-based result covers constructed categories, not all harms.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.