Research map / Video / Video generation
Memory-V2V: Memory-Augmented Video-to-Video Diffusion for Consistent Multi-Turn Editing
Dohun Lee, Chun-Hao Huang, Xuelin Chen, Jong Chul Ye, Duygu Ceylan, Hyeonho Jeong
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 video FOV retrieval to dynamic tokenization (ablation) | Memory-V2V · Component ablation (5th iteration) | MEt3R (1st vs 5th): 0.2234 -> 0.2169 | Improved |
| add adaptive token merging to dynamic tokenization (ablation) | Memory-V2V · Component ablation (5th iteration) | time (s): 980.80 -> 661.31 | Improved |
| add adaptive token merging without retrieval (ablation) | Memory-V2V · Component ablation (5th iteration) | MEt3R (1st vs 5th): 0.2234 -> 0.2344 | Hurt |
| combine retrieval and token merging (full model) (ablation) | Memory-V2V · Component ablation (5th iteration) | time (s): 980.80 -> 648.5 | Improved |
| use Memory-V2V instead of independent ReCamMaster | ReCamMaster · Multi-turn NVS | MEt3R avg consistency: 0.1892 -> 0.1357 | Improved |
| use memory instead of autoregressive chaining of ReCamMaster | ReCam (AR) · Multi-turn NVS | MEt3R 1st vs 3rd iter: 0.1985 -> 0.1525 | Improved |
| add memory to ReCamMaster for camera accuracy | ReCamMaster · Multi-turn NVS | TransErr: 24.23 -> 13.47 | Improved |
| retrieve top-3 instead of top-1 memory video (ablation) | Memory-V2V · MultiCamVideo GT comparison | PSNR: 17.15 -> 18.05 | Improved |
| add memory to LucyEdit for long video editing | LucyEdit · Long video editing | Subject Consistency: 0.8683 -> 0.9326 | Improved |
| Memory-V2V vs FlowEdit on cross-frame DINO/CLIP similarity | Memory-V2V · Long video editing | DINO-F: 0.8612 -> 0.8019 | Hurt |
| merge or discard highly responsive tokens (ablation) | Memory-V2V · qualitative token reduction | cross-turn consistency | Hurt |
| merge low-responsive tokens instead of discarding them (ablation) | Memory-V2V · qualitative token reduction | visual fidelity | Improved |
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.
- ReCamMaster — ReCamMaster: Camera-Controlled Generative Rendering from A Single Video (ICCV 2025) · better on 2 of 2 shared results
- RAVE — RAVE: Randomized Noise Shuffling for Fast and Consistent Video Editing with Diffusion Models (CVPR 2024) · better on 1 of 1 shared result
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
- Fairy: Fast Parallelized Instruction-Guided Video-to-Video Synthesis (CVPR 2024 · similarity 0.88)
- COVE: Unleashing the Diffusion Feature Correspondence for Consistent Video Editing (NeurIPS 2024 · similarity 0.87)
- VideoDirector: Precise Video Editing via Text-to-Video Models (CVPR 2025 · similarity 0.87)
- Transforming Image Editors into Video Editors (NeurIPS 2026 · similarity 0.87)
- BIVDiff: A Training-Free Framework for General-Purpose Video Synthesis via Bridging Image and Video Diffusion Models (CVPR 2024 · similarity 0.86)
- Video-P2P: Video Editing with Cross-attention Control (CVPR 2024 · 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.
