ABot-World-0 is an action-conditioned video world model that can run continuous interactive world generation on a single NVIDIA RTX 5090 GPU at 720P resolu...
ABot-World-0 is an action-conditioned video world model that can run continuous interactive world
generation on a single NVIDIA RTX 5090 GPU at 720P resolution, up to 16 FPS, with 1.2-second action-to-first-frame latency, and approximately 19 GiB peak memory usage. The model uses a unified frame-synchronized keyboard action interface, addresses long-term autoregressive drift issues through a LongForcing distribution matching phase, and integrates system-level optimizations such as a lightweight VAE decoder, low-bit DiT inference, and local context KV caching.
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs ：https://info.arxiv.org/labs/index.html.
ABot-World-0是一种动作条件视频世界模型。据摘要，其可在单张NVIDIA RTX 5090上以720P、最高16 FPS运行，并支持无限交互式世界生成。
该模型采用统一的帧同步键盘动作接口，并通过LongForcing分布匹配阶段处理长程自回归漂移，同时结合轻量级VAE解码器等系统优化。
Aioga判断，这项研究值得关注之处在于将交互生成与桌面级单卡推理结合，但现有材料未提供画质、稳定性、任务覆盖范围或对比实验细节。
若摘要所述性能在更多场景中得到验证，该方案可能降低交互式世界模型的运行门槛，并为低比特DiT推理和局部上下文KV缓存的组合应用提供参考。 值得关注论文后续披露的完整实验设置、长时间交互结果、画质评估及硬件测试方法，并核验最高16 FPS、1.2秒延迟和约19 GiB显存数据的适用条件。
The readable text on this page was extracted from the public source and organized with attribution, publication time and the original link. Copyright remains with the original author and publisher.
Ingestion channel: Summary aggregation · Source domain: arxiv.org
Source: HuggingFace Daily Papers（社区热门论文）
Original link: Open original source
Aioga archive: Open intelligence page
Content record: source-page · Updated: 2026-07-21T15:26:50.000Z
统一接入主流 AI 模型 API，为开发、测试与生产环境提供稳定调用入口。
立即访问 api.w173.com
Scan to open this articleAioga aggregates global AI updates and preserves source information for verification and citation.