The hardest part of real-time interaction has never been about how realistic the image looks—it’s about keeping the conversation going. Characters here just takes a photo, and within seconds you get a character you can actually talk to, no extra fine-tuning needed. It handles both photorealistic and illustrated styles, and the frames stream out one by one, so it can sustain continuous dialogue for over half an hour—which, for people making interactive content, matters way more than resolution.
It launched back in March, and now making it into the Real-Time Live! segment at least proves it runs live on the spot, not some pre-edited clip. The team mentioned future directions like navigable environments, multi-reference images, and memory.
The first two, if they actually ship, basically mean a virtual actor studio you can walk into. Memory is the toughest one, and it’s also what decides whether this is a toy or a real tool. I’m more curious about the latency curve though—does it stay stable after twenty minutes of chatting?