zju3dv
Featured
INTACT-JEPA
INTACT is an end-to-end JEPA world model that learns an isomorphic intent-to-action interface, enabling search-free direct control for goal-conditioned robot tasks with high success rates after just one epoch of training.
lingbot-video
LingBot-Video is the first open-source large-scale Mixture-of-Experts (MoE) video generation model designed for embodied intelligence, offering efficient inference and state-of-the-art performance on robotics-related benchmarks.
Stars
878
7-day growth
No data
Forks
39
Open issues
9
License
Apache-2.0
Last updated
2026-07-10
AI repository intelligence
FR-AI / ANALYSISWhy it is worth attention
It is the first open-source MoE video generation model specifically targeting embodied intelligence, achieving top rankings on the RBench leaderboard while being fully open-sourced under Apache 2.0.
Who it is for
- Researchers in embodied AI and robotics
- Video generation model developers
- Practitioners in robot simulation and training
- Open-source AI enthusiasts
Use cases
- Generating physically plausible videos for robot task planning and simulation
- Creating text-to-video and text-to-image-to-video content with high motion and physical rationality
- Evaluating video synthesis models on embodied benchmarks (manipulation, navigation, etc.)
- Developing custom prompt rewriting and auto-negative pipelines for video generation
Strengths
- Efficient MoE architecture with ~3x faster inference compared to dense models of equivalent capacity
- Trained on 70,000+ hours of embodied data alongside massive web videos, enabling strong physical understanding
- Top performance on RBench leaderboard across multiple embodied categories (avg. 0.620) among open-source models
- Comprehensive inference workflow with prompt rewriting, auto-negative, and multi-GPU support (FSDP, CP8, SGLang)
Considerations
- Requires specific runtime versions (Python ≥3.10, custom torch builds) and careful environment setup
- The 30B-A3B MoE model demands significant GPU memory and system RAM for loading, even with FSDP sharding
- Inference expects structured JSON captions rather than natural language prompts, necessitating the rewriter pipeline
README quick start
LingBot-Video
🌐 Project Page | 🤗 Hugging Face | 🤖 ModelScope | 📄 Paper | ⚖️ License| 💬 WeChat 微信 Group
📘 English Usage: English Documentation
📕 中文使用文档: 中文文档
We are excited to introduce LingBot-Video, the first open-source large-scale MoE (Mixture-of-Experts) video generation model dedicated to embodied intelligence. As a top-tier video model, LingBot-Video is designed to bridge the gap between video synthesis and physical world understanding.
🔥 Key Highlights
- 🚀 Efficient MoE Architecture: Scaled from scratch; balanced between capacity and cost with ~3x faster inference.
- 📦 Data Engine: Trained on massive web videos integrated with 70,000+ hours of embodied data.
- ⚖️ Multi Reward System: Rewarded for high aesthetics, physical rationality, and task completion.
🎬 Video Demos
🔥 Latest News
- July 9, 2026: 🎉 We release the technical report, code, models, rewriters for LingBot-Video.
📦 Model Download
| Model Name | Components | Tasks | Download |
|---|---|---|---|
| ⚡ LingBot-Video-Dense | Dense (1.3B) | T2I, T2V, TI2V | 🤗 Huggingface 🤖 ModelScope |
| 💪 LingBot-Video-MoE | MoE (30B-A3B) + Refiner | T2I, T2V, TI2V, Refinement | 🤗 Huggingface 🤖 ModelScope |
| 📝 LingBot-Video-Rewriter-Base | Qwen3.6-27B official | Prompt rewriter (Expand) | 🤗 Huggingface 🤖 ModelScope |
| 📝 LingBot-Video-Rewriter-Adapter | Qwen3.6-27B LoRA | Prompt rewriter (Json) | 🤗 Huggingface 🤖 ModelScope |
🚀 Quick Start
🛠️ Installation
The root `requireme
Description
Scaling Mixture-of-Experts Video Pretraining for Embodied Intelligence
Related repositories
Similar projects matched by category, topics, and programming language.
lopopolo
Featured
harness-engineering
Harness Engineering is a methodology for improving coding agent outputs by carefully crafting the environment around them—providing curated context, tools, and executable constraints that encode an organization’s nonfunctional requirements and cumulative lessons.
slvDev
Featured
esp32-ai
A 28.9 million parameter language model runs on an $8 ESP32-S3 microcontroller entirely on-device, generating simple stories at about 9.5 tokens per second.