AI Paper Atlas

Research map / Video / Video generation

VINCIE: Unlocking In-context Image Editing from Video

Leigang Qu, Feng Cheng, Ziyan Yang, Qi Zhao, Shanchuan Lin, Yichun Shi, Yicong Li, Wenjie Wang, Tat-Seng Chua, Lu Jiang

ICLR 2026 · Video › Video generation · approach: Compositional text-to-video diffusion · 6 citations (OpenAlex)

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 changedWhereMeasuredEffect
add segmentation prediction (CS -> I at inference) (ablation)VINCIE (intermediate ckpt) · MSE-Benchturn-5 success rate: 0.113 -> 0.173Improved
add segmentation tasks in training, image-only inference (ablation)VINCIE (intermediate ckpt) · MSE-Benchturn-5 success rate: 0.113 -> 0.103No clear effect
add segmentation (CS -> NS -> I) (ablation)VINCIE (intermediate ckpt) · MagicBrushturn-3 DINO: 0.592 -> 0.679Improved
train on video sequence data instead of pairwise data (ablation)VINCIE · MSE-Benchturn-5 success rate: 0.010 -> 0.220Improved
SFT on pairwise data after sequence pretraining (ablation)VINCIE · MSE-Benchturn-5 success rate: 0.220 -> 0.250Improved
add SFT to 7B modelVINCIE 7B · MSE-Benchturn-5 success rate: 0.350 -> 0.487Improved
scale model from 3B to 7B (no SFT)VINCIE · MSE-Benchturn-5 success rate: 0.210 -> 0.350Improved
scale model from 3B to 7B (no SFT)VINCIE · MSE-Benchturn-1 success rate: 0.913 -> 0.837Hurt
add dummy context before turn-1 (ablation)VINCIE · MagicBrushturn-1 L1: 0.155 -> 0.086Improved
add dummy context at turn 2-3 when history present (ablation)VINCIE · MagicBrushturn-2 L1: 0.099 -> 0.087No clear effect
interleaved RoPE instead of text-then-image (full attention) (ablation)VINCIE · MSE-Benchturn-5 success rate: 0.160 -> 0.183Mixed
block-causal instead of full attention (ablation)VINCIE · MSE-Benchturn-5 success rate: 0.183 -> 0.120Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy