gpt-image-2 Image-to-Image Portrait Style Demo: Dual Paths of Realism and Artistry

Using a single portrait reference image as input, this demo showcases gpt-image-2's image-to-image capability by comparing the results of the same subject under two different prompt directions: one takes a photorealistic route (natural skin texture, soft window light, medium-format quality), and the other takes an artistic route (painterly color blocks, strong light contrast, editorial fashion atmosphere). The work focuses on prompt-writing methods: the realism direction emphasizes light source, lens focal length, skin texture, and color fidelity; the artistic direction emphasizes brushwork, color palette, compositional tension, and stylistic references. Both prompts retain constraints on the person's identity and pose, making it easy for developers to compare the degree of output drift under "the same reference image + different constraint words." Input composition: 1 portrait reference image (generated by the production line, bound to image_urls[0]), quality set to high to observe detail fidelity, resolution set to 2K, and aspect_ratio fixed at 4:5 for portrait vertical framing. Use cases: style exploration for commercial portrait photography, accumulation of image-to-image prompt templates, and multi-version iteration of editorial content. Note that some claims in the research regarding 4K output and text rendering accuracy come partly from third-party platforms and lack independent benchmark validation; high resolution and high quality tiers significantly affect call cost and latency.