Recent video editing models have achieved impressive results, but most still require large-scale paired datasets. Collecting such naturally aligned pairs at scale remains highly challenging and constitutes a critical bottleneck, especially for local video editing data. Existing workarounds transfer image editing to video through global motion control for pair-free video editing, but such designs struggle with background and temporal consistency. In this paper, we propose NOVA: Sparse Control & Dense Synthesis, a new framework for unpaired video editing. Specifically, the sparse branch provides semantic guidance through user-edited keyframes distributed across the video, and the dense branch continuously incorporates motion and texture information from the original video to maintain high fidelity and coherence. Moreover, we introduce a degradation-simulation training strategy that enables the model to learn motion reconstruction and temporal consistency by training on artificially degraded videos, thus eliminating the need for paired data. Our extensive experiments demonstrate that NOVA outperforms existing approaches in edit fidelity, motion preservation, and temporal coherence.
Figures & tables
Figure 2: Chanllenges in Local Editing . Exisiting general video editing methods ( e.g . VACE [ 22 ] ) and datasets ( e.g . Senorita-2M [ 58 ] ) perform well on global editing tasks but struggle with local editing, often producing artifacts and inconsistent edits in the targeted areas.
Figure 3: The Limitations of Existing Schemes Previous methods are often limited by either requiring costly per-video finetuning [ 35 , 14 ] or pre-training on large-scale paired video data [ 58 , 22 ] , which is difficult to acquire. Our method decouples control and synthesis signals, enabling a self-supervised framework that learns from unpaired data while maintaining high fidelity to the source video.
Figure 4: Inconsistent backgrounds in naive multi-keyframe approach We designate frames 0, 20, 40, 60, and 80 as edited keyframes (anchors). In non-keyframes reconstructed background exhibits inconsistent textures (e.g., on the building wall) and implausible motion (e.g., in the trees), as the model hallucinates content without access to the original video.
Figure 5: Training Pipeline . (1) Center (Model Architecture) : The core model learns to denoise the source video by processing conditional inputs through a Sparse Branch S and a Dense Branch D , which interact via cross-attention; (2) Left (Anchored Control Pipe) : A degraded reference video is generated by linearly interpolating between sparsely selected keyframes, providing sparse temporal control; (3) Right (Source Fidelity Pipe) : A synthetic edited video is created using a cut-and-paste method to simulate realistic artifacts, serving as a dense synthesis target.
Figure 6: Inference Pipeline : We first edit the user-specified keyframes sequentially with reference to the first edited frame, producing temporally coherent anchors. We then construct a degraded reference video by interpolating between these keyframes, and feed it into the sparse branch S . The original unedited video is passed through the dense branch D .
Table 1: Quantitative comparison with state-of-the-art methods : We compare our method with recent frame-guided video editing approaches on multiple metrics. Our method outperforms all baselines in most metrics while requiring no per-video finetuning.
Figure 8: Ablation of the Dense Branch : The inclusion of the Dense Branch enables the model to capture background details within the input video, thereby significantly enhancing the per-frame CLIP Similarity with the ground truth.
Figure 9: Ablation on the Role and Robustness of the Dense Branch. (Left) Deactivating the dense branch entirely results in hallucinated details in the background. (Middle) Providing the clean source video enables a near-perfect reconstruction that preserves high-frequency textures. (Right) The model recovers a background that is visibly sharper than the blurred input.
Figure 10: Ablation of Consistency-Aware Keyframe Editing
Model
TC ↑
FC ↑
BG-SSIM ↑
FLUX.1-Kontext-dev
0.92
0.85
0.88
Qwen-Image-Edit-2509
0.88
0.87
0.89
Table 2: Ablation Study of Keyframe Editing Models.
Figure 11: Ablation study on the keyframe interval used during inference : The model was trained with a fixed interval of 10 . We evaluate its robustness by testing with different intervals.
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
Text-guided video editing modifies target content while preserving the appearance and temporal coherence of unedited regions. Training-based approaches provide strong control but demand substantial data and computation. Training-free methods fall into inversion-free and inversion-based paradigms. Inversion-free approaches avoid trajectory recovery, but their source-preserving guidance can limit editing strength and leave semantic changes incomplete. Inversion-based approaches recover a latent trajectory before regeneration, where approximation errors can accumulate and cause source-content drift and temporal inconsistency. We introduce Edit-VAR, the first training-free and inversion-free framework for text-guided video editing with a pretrained visual autoregressive video model. Edit-VAR directly encodes the source video into multi-scale discrete tokens and performs probability-guided conditional token replacement for source preservation. Attention-guided token-wise and scale-aware modulation selectively relaxes source constraints over edit-relevant positions and generation stages. Scale-Decoupled Generation, implemented as late-scale constraint release, regenerates motion-consistent details and reduces texture fragmentation. Residual-guided token pruning further exploits redundancy at the final two high-resolution scales to reduce inference cost. Extensive experiments and a blind user study demonstrate that Edit-VAR outperforms existing training-free video editing methods overall in editing fidelity, source preservation, temporal coherence, and inference efficiency.
Chongbo Zhao, Jiangming Wang, Xilai Wang +5
Sun Yat-sen University · South China University of Technology · Tsinghua University +3
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.