AI Paper Atlas

Research map / Video / Video generation

MoCha: End-to-End Video Character Replacement without Structural Guidance

Zhengbo Xu, Jie Ma, Ziheng Wang, Zhan Peng, Jun Liang, Jing Li

CVPR 2026 · Video › Video generation · approach: Diffusion-based human image animation

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 the real-human datasets (expression-driven and augmented video-mask data) to rendered data (ablation)MoCha · Qualitative, real-human casesfacial appearance and realismImproved
add RL-based identity-enhancing post-training with reward LoRA (ablation)MoCha · Qualitativefacial identity preservationImproved
use single-frame mask without structural guidance instead of reconstruction-based paradigmMoCha vs Wan-Animate · Synthetic benchmarkPSNR: 19.20 -> 23.09Improved
use MoCha instead of Wan-AnimateWan-Animate · Synthetic benchmarkLPIPS: 0.213 -> 0.152Improved
use MoCha instead of VACEVACE · Real-world VBenchSubject Consistency: 71.19 -> 92.25Improved
use MoCha instead of VACEVACE · Real-world VBenchImaging Quality: 60.88 -> 59.58Hurt
use MoCha instead of HunyuanCustomHunyuanCustom · Real-world VBenchTemporal Flickering: 97.98 -> 97.98No clear effect
assign same RoPE frame index to source and target tokens and fixed index -1 to references (condition-aware RoPE)MoCha · DesignflexibilityImproved

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