Sora 2 API Capabilities Update

OpenAI has released an updated prompting guide for Sora 2, documenting the latest features available in the video generation API. The guide provides practical guidance for developers and creators on how to craft effective video prompts and leverage the model's capabilities.

Key New Features

The latest Sora 2 release includes several significant improvements:

• Character references – Upload character reference clips once and reuse them across multiple video generations with consistent appearance. You can reference up to two uploaded characters per generation.
• Extended video length – Maximum clip duration has increased from 12 seconds to 20 seconds, with support for "4", "8", "12", "16", and "20" second durations.
• Higher resolution exports – Sora-2-pro now supports resolutions up to 1920×1080 and 1080×1920, in addition to existing 720x1280 and 1280x720 options.
• Video extension – Continue existing videos using the full original clip as context, not just the final frame.
• Batch API support – Run asynchronous video generation jobs for larger production workflows.

API Parameters and Endpoints

The guide clarifies that video attributes like resolution, duration, and character references must be set explicitly as API parameters and cannot be requested in prose. The Sora API supports four main endpoints:

• POST /v1/videos – Create a new video from a prompt
• POST /v1/videos/characters – Upload a character reference clip
• POST /v1/videos/extensions – Extend an existing video
• POST /v1/videos/{video_id}/edits – Edit an existing video with a new prompt

Prompting Best Practices

The guide emphasizes balancing specificity with creative freedom. Detailed prompts provide more control and consistency, while lighter prompts allow the model more creative freedom. Key recommendations include:

• Describe shots as if sketching a storyboard, including camera framing, depth of field, action, lighting, and color palette
• Shorter clips (4-8 seconds) generally follow instructions more reliably than longer ones
• Use the same prompt multiple times to generate different variations – this enables iteration and discovery
• Structure multi-shot prompts by keeping each shot block distinct with one camera setup, subject action, and lighting recipe

The guide provides concrete examples and emphasizes that prompting is a collaborative, iterative process rather than an exact science.