用 JSON 提炼图像风格并复用创作
通过提取、复用和对比迭代视觉风格，让 ChatGPT 更稳定地生成接近参考图的作品。
How to get ChatGPT to reverse-engineer the style of an image you find online.. the process is stupidly simple: step 1. upload your reference image to ChatGPT. step 2. paste this prompt: “extract this image’s visual style as reusable JSON: colors, lighting, typography, composition, textures and effects. identify the 3 defining traits to preserve. leave out the subject, logos and wording.” step 3. save the JSON as your style profile. step 4. attach it whenever you want that look and ask: “create an image of [your idea] using this style profile.” step 5. upload the result alongside your reference and ask: “compare the styles of these two images. identify the 3 biggest differences and rewrite my generation prompt to bring the style closer to the reference while keeping my new subject.” the hardest part is finding an image you like.