New ASUKA framework improves AI image inpainting by reducing artifacts

Researchers have developed a new framework called ASUKA (Aligned Stable Inpainting with UnKnown Areas prior) to address common issues in generative image inpainting. This post-hoc method aims to reduce unwanted object insertion by using reconstruction-based priors and improve color consistency through a specialized VAE decoder that treats decoding as a local harmonization task. ASUKA has demonstrated effectiveness on U-Net and Diffusion Transformer models, showing significant improvements on the Places2 and MISATO benchmarks. AI

IMPACT This research offers a novel approach to enhance the quality and consistency of AI-generated images, potentially improving applications in digital art and media.

RANK_REASON The cluster contains an academic paper detailing a new method for image inpainting. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• Aligned Stable Inpainting
• ASUKA
• Diffusion Transformer
• GitHub
• Misato
• Places2
• U-Net
• Yikai Wang

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →