Reproduce “Generative Visual Prompt: Unifying Distributional Control of Pre-Trained Generative 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 Henry Wu, Saman Motamed, Shaunak Srivastava, Fernando De la Torre
PromptGen learns a sampling distribution over a pretrained generator’s latent codes, using an invertible network guided by an energy-based control such as CLIP. This replaces per-image latent optimization with feed-forward inference without changing generator weights. On StyleGAN2’s baby prompt, it finds lower CLIP energy and 0.2-second sampling versus PPGM’s 21.5 seconds; this excludes training cost, and outputs remain limited to generator coverage.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.