Reproduce “A Task is Worth One Word: Learning with Task Prompts for High-Quality Versatile Image Inpainting” 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
Junhao Zhuang, Yanhong Zeng, Wenran Liu, Chun Yuan, Kai Chen
PowerPaint introduces a versatile inpainting model that achieves state-of-the-art performance across text-guided object synthesis, context-aware image filling, and object removal by integrating learnable task prompts into a pre-trained text-to-image diffusion model. This unified approach provides robust control for diverse inpainting tasks and significantly improves object removal capabilities compared to prior methods.
Key insights
- A single unified inpainting model can achieve state-of-the-art results across diverse tasks without compromising individual task performance.
- Learnable task-specific prompts provide an effective and flexible mechanism to guide complex generative models for distinct inpainting intents.
- Combining positive and negative task prompts via classifier-free guidance offers a robust strategy for object removal, even in challenging contexts.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.