#3737th of 80 in Open modelsFor image, video and design, video and animation
lingbot-world
Image-to-video world model with camera or action control, built on Wan2.2
Live demo ↗
(opens in a new tab)Website ↗
(opens in a new tab)Repository on GitHub ↗
(opens in a new tab)
- Stars
- 4.6k
- License
- Apache-2.0
- Last commit
- Jul 2026
- Language
- Python
Missed the latest GitHub check; these facts may be older.
Overview
LingBot-World generates interactive video from an input image and text prompt, steered by camera poses or action inputs, with minute-level horizons. Weights are released for Base (Cam), Base (Act) and Fast variants at 480P and 720P, run through torchrun scripts on top of Wan2.2. The README states this repository is no longer maintained and points to lingbot-world-v2.
Who it is for: Researchers and engineers with multi-GPU rigs experimenting with world models
Strengths
- Three open-weight variants: Base (Cam), Base (Act) and Fast, at 480P and 720P
- Fast variant uses chunked causal inference with KV caching; README claims under 1 s latency at 16 FPS
- Camera control accepts poses.npy and intrinsics.npy, which can be estimated from video with ViPE
- Community 4-bit NF4 quantized checkpoint exists for limited GPU memory, inference only
Weaknesses
- Repository is marked no longer maintained; development moved to lingbot-world-v2
- Reference commands assume 8 GPUs with FSDP and Ulysses parallelism
- Setup defers to Wan2.2 docs and needs flash-attn built from source
- Quantized variant shows minor loss in visual fidelity and temporal consistency
What it needs
- GPU required
- Needs PyTorch >= 2.4.0, flash-attn, Wan2.2 codebase, ViPE (optional, for camera poses)
- Models: LingBot-World-Base (Cam), LingBot-World-Base (Act), LingBot-World-Fast, NF4 4-bit quantized Base (Cam), community
Also in Open models
See all 80| Rank | Project | Score |
|---|---|---|
| 1 | ruviewWiFi CSI sensing on ESP32 for presence, breathing, heart rate and pose | 94 out of 100 |
| 2 | layaSingle-pass multilingual classifier returning typed decisions with calibrated odds | 88 out of 100 |
| 3 | rf-detrReal-time transformer models for object detection, segmentation and keypoints | 81 out of 100 |
| #4 | timesfmPretrained time-series foundation model for zero-shot forecasting from Google Research | 78 out of 100 |
| #5 | open_clipPyTorch library for training and running CLIP-style image-text models | 78 out of 100 |