Researchers propose InverFill, a one-step inversion method for image inpainting that injects semantic information from the input masked image into the initial noise. This approach enables high-fidelity few-step inpainting and improves upon existing methods by leveraging few-step text-to-image models in a blended sampling pipeline. Experiments show that InverFill boosts baseline few-step models, improving image quality and text coherence without costly retraining or heavy iterative optimization. --- Why it matters: This matters to AI engineers because it addresses the trade-off between photorealism and practical use in image inpainting tasks. By providing a more efficient method for high-fidelity inpainting, InverFill has significant implications for applications such as image editing and restoration. Source: https://arxiv.org/abs/2603.23463