The players

Sora (OpenAI): the most coherent in long sequences. Closed beta, gradual access. Runway Gen-3: the most consumer mature. Subscription. Kling (Kuaishou, China): surprising in physical realism. Veo (Google): strong specs, limited access.

Comparison by dimension

Length: Sora leads (60s). Runway/Kling typically 5-10s.

Physical coherence: Sora and Kling tied. Runway slightly behind.

Style control: Runway leads (good UI, predefined styles).

Price: Runway most accessible ($15-95/month). Sora API expensive when available.

Latency: 2-5 min per clip in all of them.

Use cases by tool

Marketing/advertising: Runway is the choice for creative teams.

Pre-visualization (storyboarding): any of them works.

Short content for social: Runway or Kling.

Coherent long sequences: Sora when you have access.

Common limitations

Inconsistent humans: faces change between frames in long shots. Complex physics: objects breaking, water in detail, complex interactions fail. Cost at scale: generating hundreds of clips is expensive. Detailed control: "this character has this hair, this clothing, this color" is hard to control consistently.

Where this is going

The next leap will be controllable video: keyframes, character consistency across multiple clips, manipulable physics. Companies that get this win the next phase: not just creative video, but reliable production.

Conclusion

Generative video in 2026 is functional for advertising, marketing and storyboarding — not yet for professional production replacement. The next 18 months will see massive adoption in middle-budget projects. Top professional production remains in director + crew domain for now.