AI Paper Atlas

Research map / Video / Video generation

VGEdit: Unlocking Video Generation Priors for Reasoning-Informed Image Editing

Haiquan Lu, Gongfan Fang, Xinyin Ma, Xinchao Wang

ECCV 2026 · Video › Video generation · approach: Trajectory-guided motion control · 0 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 MLLM instruction transfer to video-oriented descriptions (ablation)HunyuanVideo-1.5 · RISE-Bench Overallaccuracy (%): 6.2 -> 15.2Improved
add instruction transfer (ablation)HunyuanVideo-1.5 · KRIS-Bench Overalloverall score: 57.71 -> 67.96Improved
add instruction transfer (ablation)HunyuanVideo-1.5 · ImgEdit Overalloverall score: 2.87 -> 3.56Improved
add RL adaptation on top of instruction transfer (ablation)HunyuanVideo-1.5 + instruction transfer · RISE-Bench Overallaccuracy (%): 15.2 -> 22.3Improved
add RL adaptation (ablation)HunyuanVideo-1.5 + instruction transfer · KRIS-Bench Overalloverall score: 67.96 -> 72.48Improved
add RL adaptation (ablation)HunyuanVideo-1.5 + instruction transfer · ImgEdit Overalloverall score: 3.56 -> 4.33Improved
score a collage of the whole trajectory instead of the final frame (logit reward) (ablation)VGEdit · RISE-Bench Overallaccuracy (%): 20.4 -> 22.3Improved
score the whole trajectory instead of the final frame (sampling reward) (ablation)VGEdit · RISE-Bench Overallaccuracy (%): 15.7 -> 17.5Improved
use logit-based expectation reward instead of sampled token (trajectory-level) (ablation)VGEdit · RISE-Bench Overallaccuracy (%): 17.5 -> 22.3Improved
use logit-based expectation reward instead of sampled token (trajectory-level) (ablation)VGEdit · ImgEdit Overalloverall score: 4.01 -> 4.33Improved
use logit reward instead of sampling (final-frame) (ablation)VGEdit · KRIS-Bench Overalloverall score: 69.34 -> 70.43Improved
repurpose video model with VGEdit instead of image editor Qwen-ImageVGEdit vs Qwen-Image · RISE-Bench Overallaccuracy (%): 8.9 -> 22.3Improved

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