We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control
deepmind.google·6d ago
AI Engineering
OpenAI Blog·Sep 30·2 min read
Sora 2 is here
TL;DR
Prior video generation models struggled with physical accuracy, realism, and fine-grained control over output. OpenAI built Sora 2, a video generation model that improves physical plausibility, visual quality, and user control while adding synchronized dialogue and sound effects.
✦ Why It Matters
Engineers can now build video generation into applications with better control, realism, and multimodal output (video plus synchronized audio).
Key Takeaways
- Sora 2 improves physical accuracy and realism in generated video content.
- New synchronized dialogue and sound effects enable multimodal video creation.
- Sora app provides direct access for experimentation and production use.
- Enhanced controllability allows more precise specification of video output.
Full Summary
Video generation—the AI task of creating moving images from text or images—has historically produced unrealistic physics and limited user control over specific details. OpenAI released Sora 2, an upgraded video generation model designed to address these gaps.
The model improves three core capabilities: physical accuracy (objects and scenes behave realistically), visual realism (higher quality and detail), and controllability (users can specify what they want more precisely). A new feature is synchronized dialogue and sound effects, meaning audio matches video timing and content.
The model is accessible through a new Sora app, enabling engineers and creators to experiment with video generation at scale. Specific performance metrics or benchmarks are not provided in this announcement.