---
title: lingbot-world
canonical_url: "https://www.modelscope.ai/studios/Iceberg-Yang/lingbot-world"
md_url: "https://www.modelscope.ai/studios/Iceberg-Yang/lingbot-world.md"
repository: Iceberg-Yang/lingbot-world
chinese_name: LingBot-World-Infinity
last_updated: 2026-09-15
sdk_type: gradio
sdk_version: 6.17.3
downloads: 0
stars: 8
---

# lingbot-world

> lingbot-world - An interactive demo by Iceberg-Yang on ModelScope. Generate camera-controlled MP4 videos from an image and prompt. Independent English adaptation of LingBot-World V2 1.3B. CC BY-NC-SA 4.0; non-commercial use only.

Iceberg-Yang/lingbot-world is an interactive demo (Studio) on ModelScope. built with gradio 6.17.3. also known as 「LingBot-World-Infinity」.

- **Repository**: Iceberg-Yang/lingbot-world
- **SDK**: gradio
- **SDK version**: 6.17.3
- **Downloads**: 0
- **Stars**: 8
- **Last updated**: 2026-09-15

Source: https://www.modelscope.ai/studios/Iceberg-Yang/lingbot-world

---

# LingBot-World V2 · Image to Video

Upload an image, enter a prompt, and choose a preset camera motion to generate a complete MP4 for playback and download. Adapted from the [hugging-apps Space](https://huggingface.co/spaces/hugging-apps/lingbot-world-v2-1-3b-demo), pinned to source revision `792aa925d91127ba4d78e0752adfc038605369c9`. The model and core algorithms are provided by [Robbyant / LingBot-World](https://github.com/Robbyant/lingbot-world-v2). This is an independent ModelScope adaptation, not an official service from the authors.

Primary model: [ModelScope · LingBot 1.3B causal-fast](https://modelscope.ai/models/Robbyant/lingbot-world-v2-1.3b-causal-fast), pinned to revision `02c64a949b3336dccd4757ebaca4efb43f88d73d`. T5, VAE, and tokenizer components come from [Wan-AI](https://modelscope.ai/models/Wan-AI/Wan2.1-T2V-1.3B). Apache-2.0 applies only to the auxiliary components. The primary model, source Space, and derivative adaptation are governed by **CC BY-NC-SA 4.0 (Attribution, NonCommercial, ShareAlike)**. The original license is distributed as `LICENSE.txt` inside `space_bundle.tar`.

Downloads retain the verified `modelscope.cn` endpoints and pinned revisions with size and SHA256 checks; the international file API did not resolve these exact revisions during preparation. No Hugging Face download fallback is used.

Adaptations: removed ZeroGPU and dynamic Hugging Face downloads; reused persistent ModelScope caches, an isolated worker, and cuDNN library isolation; retained FP32 primary weights and BF16 autocast; applied a blue/purple Soft theme and an English interface. Examples only populate inputs. Parameter limits, GPU mutual exclusion, timeouts, MP4 readback verification, and private diagnostic isolation remain enabled. Platform Torch/CUDA is not replaced, and SGLang is not used.

Uses the source Space's shift=10, chunk=3, four-step sampling, and 81 frames by default (16 FPS playback, approximately 5.06 seconds). Output pixel area is capped at 832×480, with dimensions adapted to the input aspect ratio. 237-frame generation is not enabled pending separate L20 validation. Parts of the 1.3B configuration are reconstructed from weights and reference configurations; strict loading does not imply author-approved visual quality. The first run prepares the environment and verifies approximately 18.73 GB of cached components.

The Chinese MP4 version is frozen at tag `mp4-cn-fb19a55` (commit `fb19a558bf75cb8f35e58654b36fe1ba547573af`) in the Chinese Studio repository. This English edition is maintained separately and does not update that deployment. The earlier keyboard experiment remains at tag `keyboard-experiment-e9ac99e` (commit `e9ac99e5c93a6b926c33e34a1e338f9ccfa3493a`). This interface does not provide real-time keyboard controls. Generated videos and job records expire; download results promptly.

#### Clone with HTTP
```bash
 git clone https://modelscope.ai/studios/Iceberg-Yang/lingbot-world.git
```