Blending-NeRF: Text-Driven Localized Editing in Neural Radiance Fields

TL;DR

Blending-NeRF introduces a novel approach for localized, text-driven editing of 3D objects using dual NeRF networks and CLIP guidance, enabling natural modifications without distorting object form.

Contribution

The paper presents a new NeRF-based model with blending operations and dual networks for precise, text-guided local editing of 3D objects, addressing a key challenge in 3D content manipulation.

Findings

Produces natural, localized 3D edits from text prompts

Effectively modifies colors, textures, and parts of objects

Maintains object integrity during editing

Abstract

Text-driven localized editing of 3D objects is particularly difficult as locally mixing the original 3D object with the intended new object and style effects without distorting the object's form is not a straightforward process. To address this issue, we propose a novel NeRF-based model, Blending-NeRF, which consists of two NeRF networks: pretrained NeRF and editable NeRF. Additionally, we introduce new blending operations that allow Blending-NeRF to properly edit target regions which are localized by text. By using a pretrained vision-language aligned model, CLIP, we guide Blending-NeRF to add new objects with varying colors and densities, modify textures, and remove parts of the original object. Our extensive experiments demonstrate that Blending-NeRF produces naturally and locally edited 3D objects from various text prompts. Our project page is available at https://seokhunchoi.github.io/Blending-NeRF/