Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) has unveiled WorldGuide, a video world model designed to plan and execute procedural tasks such as folding origami, assembling furniture and cooking.
Starting with an image and a task goal, WorldGuide selects an action, generates a short video clip and uses the result to decide what to do next or when to stop. The system combines a ContextPlanner, an Executor and hierarchical visual memory for managing longer sequences.
MBZUAI also released WorldGuide Bench, a benchmark containing about 59,000 procedural videos across 245 tasks and 27 categories. On the benchmark, WorldGuide achieved a 33.33% task success rate, compared with 29.90% for MiniMax-H3. On Video-CraftBench, it scored 47.69%, versus 32.73% for MiniMax-H3.
The university said closed-loop execution improved task success by 21.62% over an open-loop version, while visual feedback added 18.61%. Visual memory also improved results by 5.77% on WorldGuide Bench.
MBZUAI noted that execution and stopping errors remain a challenge. The research team includes Ankan Deria, Komal Kumar, Hisham Cholakkal, Fahad Shahbaz Khan and Salman Khan.
Source: Middle East AI News
