The Visual Turing Test: Victor Riparbelli (Synthesia) and Matt Rouif (Photoroom) at CVAI London

Despite the hype and impressive demos, AI generated video is still wildly overpriced for any broad applications — around $250 a minute, estimated Victor Riparbelli, the cofounder of AI video generator startup Synthesia. His hope is the open source video models will bring the cost down. But we’re still at least six months away from that category to catch up with what top players like Google’s Veo3 can do. Still, we’re at a time where the capabilities of AI video models are improving by huge leaps. “With video it feels like we’re just now on the ramp up,” he said, contrasting that with text models which he argued have leveled off in improvements. On the AI image-generation side, we’re already nearing a point where the technology can radically alter some jobs, according to Matt Rouif, the CEO and co-founder of Photoroom. “An interior designer can be replaced by an AI at some level. That wasn’t the case before,” Rouif said, pointing to the breakthroughs made by OpenAI’s GPT4 image generators. Riparbelli also envisioned a future where video becomes so easy to create that future generations will nearly give up reading text altogether. It’s a post-literate society that he’s discussed widely, where about people replace text with more “intuitive” forms of communication. “I think at some point text will feel a bit like vinyls or something,” he said.