Sora Extend: Generate Sora 2 videos of infinite length (github.com)

Sora Extend is a workflow that lets you generate videos longer than OpenAI Sora 2’s 12‑second cap by breaking a longer prompt into coherent segments, feeding each segment into Sora 2 sequentially, and using the final frame of each clip as contextual input for the next. The system deconstructs an overarching prompt into bite‑sized instructions that preserve narrative and visual continuity, runs Sora 2 on each segment independently, captures the last frame as a seed for the following generation, and automatically concatenates the resulting clips into a continuous video without obvious transitions. For AI/ML practitioners and creators this unlocks long‑form Sora videos without manually stitching disparate outputs. Key technical elements are prompt segmentation, explicit inter‑segment context (final‑frame seeding), and automated concatenation; these minimize visual and thematic drift while still operating within Sora 2’s native generation limits. The approach reduces manual post‑production but increases reliance on careful prompt engineering, more API calls and compute, and faces potential cumulative drift or artifact accumulation over many segments. Use cases include extended storytelling, tutorials, and ads, while researchers should note tradeoffs in consistency, latency, and cost when scaling to very long durations.