TencentARC/GenCompositor
[ICLR 2026] GenCompositor: Generative Video Compositing with Diffusion Transformer
Based on the README, here's a technical summary: Combines a Diffusion Transformer backbone (CogVideoX-5B) with specialized modules for video compositing: a masked-token background preservation branch maintains target video consistency, a DiT fusion block with full self-attention integrates foreground elements, and Extended Rotary Position Embedding (ERoPE) enables flexible spatial layout control. Includes SAM2-based foreground segmentation, trajectory-guided motion injection, and trains on a curated 61K-video dataset (VideoComp) with interactive user control over element placement and dynamics.
150 stars.
Stars
150
Forks
6
Language
Python
License
—
Category
Last pushed
Mar 16, 2026
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/diffusion/TencentARC/GenCompositor"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
hao-ai-lab/FastVideo
A unified inference and post-training framework for accelerated video generation.
thu-ml/TurboDiffusion
TurboDiffusion: 100–200× Acceleration for Video Diffusion Models
ModelTC/LightX2V
Light Image Video Generation Inference Framework
PKU-YuanGroup/Helios
Helios: Real Real-Time Long Video Generation Model
PKU-YuanGroup/MagicTime
[TPAMI 2025🔥] MagicTime: Time-lapse Video Generation Models as Metamorphic Simulators