GPT Image 2.6 vs Seedream 5.5: A Photorealistic Image Generation Test

Create this image with maximum photographic accuracy.
Scene: A casual full-body mirror selfie inside a modern, dimly lit elevator. Dark navy geometric patterned wall panels on the left, warm brown wood-paneled walls on the right, black reflective elevator doors in the center, illuminated elevator buttons, horizontal silver handrails, and a light gray tiled floor.
Subject: A young woman with straight, shoulder-length dark brown hair parted in the middle, natural facial features, and minimal makeup. She is standing slightly angled toward the mirror, holding a black iPhone with a transparent protective case high above her head with her right arm fully extended. Her phone is positioned in the upper-left corner of the composition. She looks slightly downward at the mirror with puffed cheeks and a playful, serious expression. Her left arm hangs naturally beside her body.
Outfit: A fitted dark charcoal-gray short-sleeve athletic jersey with white side stripes, white sports logos and lettering across the chest, loose oversized gray pinstriped pajama-style pants, and simple black flip-flops. A thin black hair tie around her raised wrist.
Composition: Vertical 9:16 smartphone photo, full-body framing from a slightly elevated mirror-selfie perspective. Subject positioned slightly left of center, raised arm creating a strong diagonal, legs relaxed and slightly apart. Preserve the exact pose, body proportions, camera angle, perspective, outfit, background layout, and framing of the reference.
Photography: Authentic unedited iPhone mirror selfie. Warm overhead elevator lighting mixed with cool ambient reflections, slightly underexposed shadows, realistic skin texture, subtle smartphone grain, natural reflections, imperfect sharpness, and ordinary indoor white balance.
CRITICAL: The result must look like a real spontaneous smartphone snapshot, NOT an AI-generated photograph. No cinematic lighting, no professional photography, no beauty filters, no skin smoothing, no HDR, no artificial sharpness, no color grading, no stylization. Preserve the original reference's realistic colors, exposure, shadows, visual imperfections, and casual everyday aesthetic.