Interesting and early. Worth a spike or exploration session.
Video Generation
Video generation jumped from novelty to production tool in one quarter — real ad campaigns and 4K/60fps output mean it's worth serious evaluation.
Multimodal·Emerging
openai.comOur Take
What It Is
AI video generation creates video content from text descriptions, reference images, or existing footage. The category saw an unprecedented cluster of launches in early 2026: Kling 3.0 (Kuaishou) offers 4K/60fps with cinematic camera controls, Sora 2 Pro (OpenAI) powers real advertising campaigns, Google's Veo 3.1 produces footage described as indistinguishable from shot video, and Seedance 2.0 (ByteDance) brings character-consistent dance and motion generation. Four of six major models now generate synchronised audio natively.
Why It Matters
We moved video generation to Emerging because the gap between demo and production closed faster than anyone expected. Sora 2 Pro powering actual ad campaigns is a different signal than Sora 1 generating impressive clips that nobody used commercially. Kling 3.0's camera controls (dolly, crane, handheld) give directors real creative tools, not just a "generate random video" button.
Native audio synchronisation across most models removes what was the most jarring limitation. When generated video has matching sound effects and ambient audio, the uncanny valley narrows significantly. The market projection ($32B to $133B by 2030 at 33% CAGR) reflects commercial demand beyond the creator economy — enterprise video for training, marketing, and product visualisation.
Key Developments
- Mar 2026: Kling 3.0 launches with 4K/60fps output and cinematic camera movement controls (dolly, crane, handheld).
- Feb 2026: Sora 2 Pro released, actively powering commercial advertising campaigns.
- Feb 2026: Seedance 2.0 from ByteDance brings character-consistent motion generation.
- Jan 2026: Google's Veo 3.1 produces footage described as indistinguishable from real shot video.
What to Watch
The consistency problem is the barrier to longer-form content. Current models excel at 5-30 second clips but struggle with character consistency, scene continuity, and narrative coherence across longer videos. Watch for models that solve multi-shot consistency — that's the breakthrough that moves this from Emerging to Promising. Also track enterprise adoption patterns: are companies using this for internal content (training videos, product demos) or customer-facing marketing? The risk tolerance differs significantly.
Strengths
- Quality threshold crossed: 4K/60fps output with cinematic camera controls and native audio puts generated video in the range of professional use.
- Production validation: Sora 2 Pro powering real ad campaigns proves commercial viability beyond demos and experiments.
- Competitive pressure: Multiple strong entrants (Kling, Sora, Veo, Seedance) are driving rapid quality improvements through competition.
- Cost reduction: Generated video for training, demos, and marketing at a fraction of traditional video production costs.
Considerations
- Consistency limitations: Character consistency and scene continuity across shots remain unsolved for longer-form content.
- Copyright and IP: Training data sourcing for video models faces ongoing legal scrutiny. Commercial use carries IP risk.
- Detection challenges: AI-generated video that's indistinguishable from real footage creates deepfake and misinformation risks.
- Compute costs: High-quality video generation is compute-intensive and expensive at scale. Pricing models are still settling.