Research map / Video / Video-language understanding
LIVEditor-14B: Lightning Unified Video Editing via In-Context Sparse Attention
Shitong Shao, zikai zhou, Haopeng Li, Yingwei Song, Wenliang Zhong, Lichen Bai, Zeke Xie
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 |
|---|---|---|---|
| apply ISA training-free instead of full attention | LIVEditor-14B · EditVerseBench | speedup vs FA3: 1.00 -> 1.47 | Improved |
| train with ISA instead of full attention (ablation) | LIVEditor-14B · EditVerseBench | VLM evaluation score (first column): 7.62 -> 7.89 | Improved |
| use ISA instead of other sparse attention (Radial, Sparge, STA, SWA, VSA) training-free (ablation) | LIVEditor-14B (full-attn) · EditVerseBench | VLM evaluation score (first column): 7.78 vs 7.44 (Sparge Attention) | Improved |
| add Stage II fine-tuning on 0.089M curated high-quality samples (ablation) | LIVEditor (ISA) · EditVerseBench | VLM evaluation score (first column): 6.46 -> 7.89 | Improved |
| add Stage II fine-tuning on high-quality data (ablation) | LIVEditor (ISA) · VIE-Bench | Avg. score: 5.16 -> 6.81 | Improved |
| reduce Flat Ratio (more sparsity) (ablation) | ISA (training-free) · EditVerseBench | all metrics | Hurt |
| reduce No Sparsity Ratio and Select Ratio (ablation) | ISA (training-free) · EditVerseBench | all metrics | No clear effect |
| fine-tune with ISA (post-training) rather than training-free (ablation) | ISA · attention approximation error across blocks | mean absolute error vs full attention | Improved |
| use 1st- and 2nd-order Taylor expansion instead of 0-th order (ablation) | ISA sparse attention · kernel implementation | hardware efficiency | Hurt |
| use query-key deviation norm instead of sharpness as error proxy for grouping (ablation) | ISA grouping · Taylor error correlation | proxy quality | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- EditVerse — EditVerse: Unifying Image and Video Editing and Generation with In-Context Learning (ICLR 2026) · 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
- iCATS: Fast Video Generation via Interaction-Aware Sparse Attention and Timestep-Adaptive Sparsity (NeurIPS 2026 · similarity 0.86)
- Faster Video Diffusion with Trainable Sparse Attention (NeurIPS 2025 · similarity 0.86)
- FADE: Frequency-Aware Diffusion Model Factorization for Video Editing (CVPR 2025 · similarity 0.85)
- Bidirectional Sparse Attention for Faster Video Diffusion Training (NeurIPS 2026 · similarity 0.85)
- FLATTEN: optical FLow-guided ATTENtion for consistent text-to-video editing (ICLR 2024 · similarity 0.85)
- Enabling Instructional Image Editing with In-Context Generation in Large Scale Diffusion Transformer (NeurIPS 2025 · similarity 0.85)
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.
