Use the female character in the reference image as the sole character reference, maintaining high consistency in facial features, hairstyle, skin tone, clothing, and overall temperament.
9:16 vertical screen, 10 seconds, ultra-realistic live-action knowledge science broadcast.
The character is sitting in a real home study, medium-close up from the chest up, fixed camera, essentially no movement. Natural window light shines from the side front, background bookshelves are slightly blurred, looking like a real knowledge blogger filmed with a camera.
Character naturally broadcasts to the camera: 'Many people think that AI answering questions is finding answers in a database. Actually it's not; it's more like continuously predicting the most likely next word based on the previous content.'
Performance requirements: 0-2s: Naturally look directly at the camera to start, conversational tone, slight blinking. 2-5s: When saying 'Actually it's not,' gently shake head once, eyebrows have natural changes. 5-8s: When explaining 'based on previous content,' one hand naturally enters the frame to make a very small explanatory gesture. 8-10s: After saying 'most likely next word,' finish with a slight nod. The whole process should not feel like reading a script or have a host's broadcasting tone. Mouth movements strictly match Chinese speech. Retain natural blinking, breathing, micro-expressions, slight head movements, and real skin texture. Eyes can occasionally briefly deviate from the camera and then naturally look back. Restrained gestures, do not wave hands frequently. Real mobile phone/mirrorless camera shooting texture, slight sensor noise, real skin quality, no airbrushing, no internet celebrity filters, no plastic skin, no exaggerated movie lighting, no camera movement, no subtitles, no watermarks, no screen text.