Image-driven video editing aims to propagate edit contents from the modified first frame to the remaining frames. Existing methods usually invert the source video to noise using a pre-trained image-to-video (I2V) model and then guide the sampling process using the edited first frame. Generally, a popular choice for maintaining motion and layout from the source video is intervening in the denoising process by injecting attention during reconstruction. However, such injection often leads to unsatisfactory results, where excessive injection leads to conflicting semantics with the source video while insufficient injection brings limited source representation. Recognizing this, we propose an Editing-awaRE (REE) injection method to modulate the injection intensity of each token. Specifically, we first compute the pixel difference between the source and edited first frame to form a corresponding editing mask. Next, we track the editing area throughout the entire video by using optical flow to warp the first-frame mask. Then, editing-aware feature injection intensity for each token is generated accordingly, where injection is not conducted in editing areas. Building upon REE injection, we further propose a zero-shot image-driven video editing framework with recent-emerging rectified-Flow models, dubbed FREE-Edit. Without fine-tuning or training, our FREE-Edit demonstrates effectiveness in various image-driven video editing scenarios, showing its capability to produce higher-quality outputs compared with existing techniques. Project page: https://free-edit.github.io/page/.
Figures & tables
Figure 2: The pipeline illustration of our FREE-Edit. Top: It obeys an “inversion-then-editing” pipeline. Starting from inverted noisy latent z1 , the reconstructed video z0 and edited video z~0 take the source X1 and edited first frame X^1 as conditions, respectively. Bottom: We design an E diting-awa RE (REE) injection method, which designs a modulation weight λ to adaptively replace the intermediate model representations ( Q~ and K~ ) in the editing process with those ( Q and K ) in the reconstruction process through self-attention blocks. Here, we first use optical flow to warp the automatically calculated first-frame editing mask, which yields tracked editing masks for subsequent frames. Based on them, we compute the modulation weight λ for each token, where no injection is performed in editing regions.
Figure 3: Qualitative ablation of different threshold ( thr ) in Eq. ( 9 ) for mask M1 generation, which are used to indicate editing regions.
Settings
CLIP score ↑
Warp Error ↓
thr=5
0.920
2.106
thr=35
0.937
0.595
thr=65
0.877
0.764
Table 1: Quantitative ablation of different threshold thr .
Figure 4: Qualitative comparison between standard FREE-Edit ( w/ REE injection), and that of w/o injection, and w/ vanilla injection.
Figure 5: Qualitative comparison to existing image-guided video editing methods Go-with-the-Flow [ 4 ] , AnyV2V [ 14 ] , I2VEdit [ 21 ] , and VideoShop [ 6 ] on various editing scenarios: adding/removing objects, style transfer, and background editing .
Settings
Whole-Image
Non-editing Area
Speed (min)
CLIP score ↑
Warp Error ↓
CLIP score + ↑
Warp Error + ↓
SSIM + ↑
PSNR + ↑
LPIPS + ↓
AnyV2V [ 14 ]
0.908
2.190
0.911
1.872
0.684
22.08
0.218
3.61
I2VEdit [ 21 ]
0.909
2.886
0.914
1.794
0.745
22.57
0.255
127
VideoShop [ 6 ]
0.822
7.415
0.854
5.298
0.568
18.08
0.417
2.45
Go-with-the-Flow [ 4 ]
0.930
2.152
0.935
1.765
0.737
22.66
0.306
3.42
FREE-Edit w/ REE injection
0.937
0.595
0.941
0.471
0.811
27.05
0.161
1.53
Table 2: Quantitative comparison with the existing first-frame-guided video editing methods. We also provide the results of FREE-Edit w/o injection, and w/ vanilla injection. The best and second-best results are marked with bold and underlines.
Ours vs
Editing Consistency ↑
Overall Quality ↑
AnyV2V
79.6%
81.9%
I2VEdit
78.5%
74.6%
VideoShop
85.2%
81.4%
Go-with-the-Flow
72.5%
68.4%
FREE-Edit w/o injection
82.2%
71.3%
FREE-Edit w/ vanilla injection
77.8%
69.8%
Table 3: User Study. Our method outperforms different editing techniques across both aspects.
Figure 6: Qualitative comparison to existing text-guided video editing methods: TokenFlow [ 7 ] , VidToMe [ 18 ] , and RF-Edit [ 33 ] .
Figure 7: FREE-Edit based on Wan2.1-14B [ 32 ] and LTX-Video-2B [ 9 ] . Although time-consuming, Wan2.1-14B has the potential to provide better frame quality than LTX-Video-2B.
Figure 8: Our limitation. When performing adding objects , our FREE-Edit cannot endow them with free motion. For example, a newly added turtle cannot have a new trajectory because there is no corresponding motion information in the source video.
Appendix figures & tables2 assets
Supplementary material from the paper’s appendix.
Appendix
Ap-Fig. 2 : More comparison between standard FREE-Edit ( w/ REE injection), and that of w/o injection, and w/ vanilla injection.
Ap-Fig. 3 : Left: Results of large changes in viewpoint. Right: A failure case with fast motion.
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/.
Ze Chen, Lan Chen, Yuanhang Li +1
MIPG, Communication University of China Beijing, China
Video editing spans diverse editing paradigms, yet achieving high-quality instruction-guided and subject-guided editing within a single unified framework remains challenging. 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, including style transfer, attribute modification, object insertion, part-level editing, and subject replacement. On FiVE, EditVid achieves 78.16 FiVE-Acc, compared with 58.95 for the strongest evaluated training-free baseline, while obtaining competitive results on IVEBench. A user study further shows a 51.8% overall preference for EditVid over 7 competing methods.
Adheesh Sunil Juvekar, Onkar Kishor Susladkar, Kiet A. Nguyen +6
Current video editing models often rely on expensive paired video data, which limits their practical scalability. In essence, most video editing tasks can be formulated as a decoupled spatiotemporal process, where the temporal dynamics of the pretrained model are preserved while spatial content is selectively and precisely modified. Based on this insight, we propose ImVideoEdit, an efficient framework that learns video editing capabilities entirely from image pairs. By freezing the pre-trained 3D attention modules and treating images as single-frame videos, we decouple the 2D spatial learning process to help preserve the original temporal dynamics. The core of our approach is a Predict-Update Spatial Difference Attention module that progressively extracts and injects spatial differences. Rather than relying on rigid external masks, we incorporate a Text-Guided Dynamic Semantic Gating mechanism for adaptive and implicit text-driven modifications. Despite training on only 13K image pairs for 5 epochs with exceptionally low computational overhead, ImVideoEdit achieves editing fidelity and temporal consistency comparable to larger models trained on extensive video datasets.