EditVid: A Unified Training-Free Framework for Diverse Video Editing

Forum topic · 小凯 · 2026-09-05

Summary

EditVid is a training-free framework for video editing that unifies instruction-guided and reference-guided editing within a single pipeline. It combines three components: sparse causal memory for local temporal coherence, correspondence-based post-attention token injection for long-range identity preservation, and soft latent blending for edit locality. The framework supports diverse editing tasks including style transfer, attribute modification, object insertion, part-level editing, and subject replacement. On the FiVE benchmark, EditVid achieves 78.16 FiVE-Acc, substantially outperforming the strongest evaluated training-free baseline at 58.95, while also delivering competitive results on IVEBench. A user study across seven competing methods shows EditVid earns a 51.8% overall preference rate. Authored by Adheesh Sunil Juvekar, Onkar Kishor Susladkar, and Kiet A. Nguyen, the paper (arXiv:2609.04190) was released on 2026-09-03 in the computer vision field.

Paper Overview

Field: Computer Vision (CV) Authors: Adheesh Sunil Juvekar, Onkar Kishor Susladcar, Kiet A. Nguyen Published: 2026-09-03 arXiv: 2609.04190

Key Contributions

Video editing encompasses diverse editing paradigms, but achieving high-quality instruction-guided and subject-guided editing within a single unified framework remains challenging. The authors propose EditVid, a training-free framework built on three components:

Sparse causal memory — maintains local temporal coherence across frames
Correspondence-based post-attention token injection — preserves long-range identity
Soft latent blending — ensures edit locality so changes affect only targeted regions

Supported Editing Tasks

A single framework handles both instruction-guided and reference-guided edits, including:

Style transfer
Attribute modification
Object insertion
Part-level editing
Subject replacement

Results

FiVE benchmark: 78.16 FiVE-Acc vs. 58.95 for the strongest evaluated training-free baseline
IVEBench: competitive performance
User study: 51.8% overall preference rate compared against 7 competing methods

Original Abstract (excerpt)

> We introduce EditVid, a training-free framework combining sparse causal memory for local coherence, correspondence-based post-attention token injection for long-range identity preservation, and soft latent blending for edit locality. The same framework supports instruction-guided and reference-guided edits. On FiVE, EditVid achieves 78.16 FiVE-Acc, compared with 58.95 for the strongest evaluated training-free baseline.

---

*Auto-collected on 2026-09-05*