The technical story of AI in this industry is almost boring now: generating and editing media is cheap and getting cheaper. The genuinely hard, unsolved problem is not capability. It’s consent.

When anyone’s likeness can be synthesised convincingly, three questions move from theoretical to urgent:

• Whose face is it? Non-consensual deepfakes are the obvious abuse, and the law is scrambling to define liability.
• Who gets paid? If a performer’s likeness or style trains a model, what do they own, and what are they owed?
• How do you prove provenance? Watermarking, verified consent records, and identity checks become core infrastructure, not nice-to-haves.

The platforms and tools that win here long-term will likely be the ones that treat consent and provenance as the product, not an afterthought. The capability is commoditised; trust isn’t.