Paper page - Goku: A Million-Scale Universal Dataset and Benchmark for Instruction-Based Video Editing

Source: https://huggingface.co/papers/2606.30599 Authors:

,

,

,

,

,

,

,

,

,

Abstract

A large-scale video editing dataset and model are introduced that support multi-task and structural manipulations through advanced data synthesis and network architectures.

Existing instruction-based video editing datasets commonly focus on single-task appearance editing, failing to meet the complex creative demands of real-world scenarios. To bridge this gap, we present Goku, a large-scale dataset featuring 2 million high-quality,instruction-aligned video editing pairs, which is the first to extend task boundaries from basic appearance editing to multi-task and structural manipulations(e.g., precise control of subject movement). To tackle the data synthesis challenges inherent in these complex tasks, we design an efficientdata synthesis pipelinethat decomposes complex edits into controllable sub-problems and introduce aprogressive filtering systemfor data reliability throughout the whole process. Furthermore, we explore the optimal network structures on Goku, and propose Goku-Edit. To deeply comprehend complex editing instructions, Goku-Edit leverages anMLLMas its text encoder and adopts adecoupled dual-branch design: a dedicatedmask branchhandles structural control, freeing the main branch forappearance rendering. A comprehensivevideo editing benchmark, Goku-Bench, is also proposed with 1,000 human-verified test cases and 7novel editing-specific metrics. Evaluated on Goku-Bench, Goku-Edit obtains up to +8% improvement on other open-source models in terms of instruction following.

View arXiv pageView PDFProject pageAdd to collection

Get this paper in your agent:

hf papers read 2606\.30599

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2606.30599 in a model README.md to link it from this page.

Datasets citing this paper2

#### Goku-2M/GOKU-2M Preview• Updatedabout 2 hours ago • 24 • 3 #### bigfacing/GOKU-2M Preview• Updatedabout 2 hours ago • 24 • 6

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2606.30599 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.