Portrait Style Redrawing: Two Paths of Realism and Artistic Treatment

Using the same portrait reference image as a base, this demonstrates two paths for gpt-image-2 image-to-image generation under prompt control: one pursues photo-level realism (retaining skin texture, natural light, and shallow depth of field), while the other moves toward artistic treatment (painterly brushstrokes, color remapping, and graphic backgrounds). The work only uses the capabilities declared in the author's pipeline: passing the reference image via input_urls, and controlling the output with prompt, aspect_ratio, and resolution, without involving parameters beyond the schema. It is suitable for scenarios requiring multi-style extension of a fixed character image, such as character visual design, social media avatars, and consistency exploration for layout illustrations.