Reproduce “Contrastive Prompts Improve Disentanglement in Text-to-Image Diffusion Models” 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
Chen Wu, Fernando De la Torre
Contrastive Guidance extends classifier-free diffusion guidance by comparing a desired prompt with a baseline prompt, steering generation by subtracting their text-conditioned signals. The method is applied to domain-specific generation, continuous controls, and zero-shot editing. With sampling randomness fixed, an AFHQ Cat analysis reports pixel-wise L2 distances of 0.271, versus 0.341 for CFG. This measures image change, not guaranteed factor preservation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.