To make a realistic AI person, generate a photorealistic reference image with GPT Image 2, then animate it with Seedance 2.0 image-to-video. Starting every video from the same image is what makes the AI person look like one consistent human rather than a different face in each clip.
Realistic AI people are built in two stages. First you create a still image of the person. Then you bring that still to life as video.
GPT Image 2 handles the still: face, lighting, wardrobe, and setting. Seedance 2.0 handles the motion, expression, and audio while preserving that identity.
Add specific, candid details to the prompt: natural phone-camera realism, real skin texture, ordinary lighting, and a concrete setting like a kitchen, car, or dorm room.
Avoid generic 'beautiful model' prompts. The most believable AI people look like ordinary people filming themselves, not polished studio portraits.
The workflow is the same on every project: generate a person with GPT Image 2, animate it with Seedance 2.0, extract the last frame, and use that frame as the next clip's start frame so the same face carries across the whole ad.
Real AI UGC Studio uses GPT Image 2 for the person image and Seedance 2.0 for video. This narrow stack is intentional: cheaper models produced lower-quality, less believable results.
Use your AI person image as the Seedance 2.0 start frame and generate a vertical clip. Seedance 2.0 adds realistic motion and native audio.
Yes. Keep one reference image and reuse it as the start frame, then chain each clip's last frame into the next to maintain the same identity.