FlowAnchor: Stabilizing the Editing Signal for Inversion-Free Video Editing
Ze Chen, Lan Chen, Yuanhang Li, Qi Mao
2026-04-27
Summary
This paper introduces FlowAnchor, a new method for editing videos using artificial intelligence without needing to retrain the AI model. It allows for changes to videos based on instructions, like text prompts, in a stable and efficient way.
What's the problem?
Existing methods for editing videos with AI, inspired by successful image editing techniques, struggle when dealing with complex scenes containing multiple objects or videos with many frames. The core issue is that the instructions given to the AI become unstable and lose their effect as they're applied to the high-dimensional data representing video, meaning the edits don't happen accurately or consistently throughout the video. Specifically, the AI has trouble pinpointing *where* to edit and *how much* to edit in each frame.
What's the solution?
FlowAnchor tackles this instability by making the editing process more focused. It uses two key ideas: Spatial-aware Attention Refinement, which makes sure the AI understands which parts of the video correspond to the editing instructions, and Adaptive Magnitude Modulation, which ensures the AI applies enough of an edit to actually see a change. These techniques work together to create a stronger, more reliable editing signal that guides the AI to make the desired changes to the video.
Why it matters?
This research is important because it makes video editing with AI more practical and effective. It allows for more realistic and coherent edits, even in challenging videos with lots going on, and it does so without the huge computational cost of retraining the AI model every time you want to make a different edit. This opens the door to easier and more accessible video manipulation for a variety of applications.
Abstract
We propose FlowAnchor, a training-free framework for stable and efficient inversion-free, flow-based video editing. Inversion-free editing methods have recently shown impressive efficiency and structure preservation in images by directly steering the sampling trajectory with an editing signal. However, extending this paradigm to videos remains challenging, often failing in multi-object scenes or with increased frame counts. We identify the root cause as the instability of the editing signal in high-dimensional video latent spaces, which arises from imprecise spatial localization and length-induced magnitude attenuation. To overcome this challenge, FlowAnchor explicitly anchors both where to edit and how strongly to edit. It introduces Spatial-aware Attention Refinement, which enforces consistent alignment between textual guidance and spatial regions, and Adaptive Magnitude Modulation, which adaptively preserves sufficient editing strength. Together, these mechanisms stabilize the editing signal and guide the flow-based evolution toward the desired target distribution. Extensive experiments demonstrate that FlowAnchor achieves more faithful, temporally coherent, and computationally efficient video editing across challenging multi-object and fast-motion scenarios. The project page is available at https://cuc-mipg.github.io/FlowAnchor.github.io/.