InstantRetouch: Efficient and High-Fidelity Instruction-Guided Image Retouching with Bilateral Space
Researchers have developed InstantRetouch, a novel method for instruction-guided image retouching that prioritizes both fidelity and efficiency. This approach utilizes a low-resolution bilateral grid of affine transforms, guided by a learned map and distilled from a diffusion model, to edit images. InstantRetouch aims to overcome the content drift and slow sampling times associated with diffusion-based methods, offering faster processing while maintaining high visual quality and accurate instruction following. AI
IMPACT Introduces a new technique for efficient and high-fidelity image editing, potentially improving user experience in creative applications.