There's that other idiot who has all those shorts and people keep sending them to me. We have no idea what the default model is on their eyeProduct app. It has to be one of the mini models with a small context window. We also don't know if it's doing speech-to-text and using that in the model, or if the model is getting voice samples directly. The image stuff could be multi-modal, or could just be processed by another model and injected as text. A lot of variables. I bet if you made a harness with a web camera and pumped it through OpenRouter, it would probably get it right.