Research map / Video / Video generation
SA-V2V: Training-Free Subject-Aware Video-to-Video Personalization
Soobin Park, Seohyeon Yoo, Jiwon Kim, SeonHwa Kim, Kyong Hwan Jin, Eunju Cha
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add stochastic temporal injection (ablation) | SA-V2V · Video personalization (ablation set) | FVD: 103.82 -> 82.84 | Improved |
| add stochastic temporal injection (ablation) | SA-V2V · Video personalization (ablation set) | Masked CLIP background similarity: 0.8146 -> 0.8783 | Improved |
| add background target-aware feature guidance (B-TFG) (ablation) | SA-V2V · Video personalization (ablation set) | Masked CLIP background similarity: 0.8313 -> 0.8783 | Improved |
| add background target-aware feature guidance (B-TFG) (ablation) | SA-V2V · Video personalization (ablation set) | FVD: 97.27 -> 82.84 | Improved |
| add subject target-aware feature guidance (S-TFG) (ablation) | SA-V2V · Video personalization (ablation set) | Masked CLIP subject similarity: 0.8101 -> 0.8312 | Improved |
| add subject target-aware feature guidance (S-TFG) (ablation) | SA-V2V · Video personalization (ablation set) | Masked CLIP background similarity: 0.9041 -> 0.8783 | Hurt |
| add subject target-aware feature guidance (S-TFG) (ablation) | SA-V2V · Video personalization (ablation set) | FVD: 72.30 -> 82.84 | Hurt |
| add spatial masking to TFG (ablation) | SA-V2V TFG · qualitative (Fig. 7) | feature separation | Improved |
| apply TFG alternately timestep-wise instead of block-wise (ablation) | SA-V2V TFG · qualitative (Fig. 7) | appearance guidance | Improved |
| use linear interpolation instead of direct spatial feature injection (ablation) | SA-V2V TFG · qualitative (Fig. 7) | background preservation | Improved |
| use training-free design instead of training-based baselines | SA-V2V · Computational cost (CogVideoX-5B) | Peak GPU memory (GB): 21.72 -> 19.11 | Improved |
| SA-V2V vs MotionClone on subject fidelity | SA-V2V · Video personalization (main comparison) | Masked CLIP subject similarity: 0.8243 -> 0.8140 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- DualReal — DualReal: Adaptive Joint Training for Lossless Identity-Motion Fusion in Video Customization (ICCV 2025) · better on 5 of 5 shared results
- MotionClone — MotionClone: Training-Free Motion Cloning for Controllable Video Generation (ICLR 2025) · better on 3 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Multi-subject Open-set Personalization in Video Generation (CVPR 2025 · similarity 0.89)
- VideoMage: Multi-Subject and Motion Customization of Text-to-Video Diffusion Models (CVPR 2025 · similarity 0.88)
- VMC: Video Motion Customization using Temporal Attention Adaption for Text-to-Video Diffusion Models (CVPR 2024 · similarity 0.88)
- Video Diffusion Models are Training-free Motion Interpreter and Controller (NeurIPS 2024 · similarity 0.86)
- CustomCrafter: Customized Video Generation with Preserving Motion and Concept Composition Abilities (AAAI 2025 · similarity 0.86)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
