AI

WorldGuide: Goal-Directed Video World Model for Procedural Task Execution

Read full article at source ↗

Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 09/10/2026, 09:18:24 UTC.

Briefings

WorldGuide is a video generation model that solves the gap between planning and executing procedural tasks by coupling planning with learned execution. It predicts atomic actions, generates video clips, and selects subsequent actions based on generated results. The system uses hierarchical visual memory to maintain state across long-horizon tasks with bounded token costs. It achieved a 33.33% task success rate on the WorldGuide-Bench compared to 29.90% for MiniMax-H3, despite MiniMax-H3 receiving reference action plans.

Key points

• WorldGuide predicts atomic actions, generates video clips, and selects subsequent actions based on generated results to solve procedural tasks.
• WorldGuide-Bench contains approximately 59K step-annotated videos across 245 tasks and 27 procedural categories.

Context

Video generators can synthesize plausible visual trajectories but fail to adapt to execution outcomes. Existing closed-loop systems rely on pretrained executors or indirect verification, leaving a gap between deciding an action and realizing it. WorldGuide addresses this by coupling planning with learned execution in visual world space.

Limitations

The study lacks step-level action-video supervision for joint planner-executor training. Results are based on a single snapshot from October 2026, and the benchmark includes 245 tasks and 27 categories, limiting generalizability to other procedural domains.

Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 15/11/2026, 13:43:33 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.

View archived briefing on IPFS
bafybeifgooys5kl2nhmqcqfh72aw7rqr2kvdbvzigkdgly5efcewgosr7m