RetinAI addresses the challenge of early retinoblastoma (RB) detection in settings where timely, accurate eye exams are limited and late-stage disease often leads to blindness or enucleation. It is the first low-cost wearable headset and retinal camera with integrated AI for early RB screening. The system combines an extraocular leukocoria detector powered by a YOLOv11 model, and an intraocular retinal imaging system using ResNet-50 or YOLOv11 to identify retinal tumors. Its hardware includes a 3D-printed headset, Pi Camera 3, dual infrared/white-LED illumination, LCD screen, 20D lens, and Raspberry Pi 5. Color analysis confirmed that leukocoria differs from normal pupils in Value, Hue, and Saturation, and the YOLOv11 detector achieved 98% mAP, identifying leukocoria as small as 1 mm. For tumor detection, ResNet-50 reached 97% accuracy and YOLOv11 reached 96% mAP. Clinical testing on both normal and RB patients showed that RetinAI accurately detects normal retina and retinoblastoma tumors. As the first low-cost system capable of identifying tumors from both external pupil images and intraocular retinal images—without pupil-dilation medication—RetinAI enables earlier, accessible RB screening in homes, routine check-ups, and small clinics without eye specialists.