GPT Image 2 リバースプロンプト：どんな画像も再現する方法

[Upload your reference image]

Analyze this image and write a detailed image generation prompt
that would reproduce it. Include: subject, scene, style, lighting,
camera (if photographic), color palette, and composition. Format
the output as a single comma-separated prompt I can paste directly
into GPT Image 2.

from openai import OpenAI
import base64

client = OpenAI()

# Step 1: encode the reference image
with open("reference.jpg", "rb") as f:
    image_data = base64.b64encode(f.read()).decode()

# Step 2: ask GPT to write a reverse prompt
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[
        {
            "role": "user",
            "content": [
                {"type": "text", "text": "Write a detailed GPT Image 2 prompt that would reproduce this image. Format as a single comma-separated string."},
                {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{image_data}"}},
            ],
        }
    ],
)

reverse_prompt = response.choices[0].message.content

# Step 3: use the reverse prompt to generate variants
new_image = client.images.generate(
    model="gpt-image-2",
    prompt=reverse_prompt,
    quality="medium",
)

result = client.images.edit(
    model="gpt-image-2",
    image=open("reference.jpg", "rb"),
    prompt="Generate a similar image in the same style and composition, but with a different subject: [your new subject].",
    quality="medium",
)

A young woman with long brown hair, sitting in a sunlit cafe by
the window, editorial film photography style with slight grain,
warm afternoon natural light from the left, 35mm lens with shallow
depth of field, warm beige and muted green palette, mid-shot
off-center composition following rule of thirds.

Analyze this image and write a detailed GPT Image 2 prompt that
would reproduce its style. Include: 1) subject and scene, 2)
specific photographic style reference (e.g., Kodak Portra 400,
Wes Anderson, National Geographic), 3) lens / aperture / lighting,
4) color palette in 2-3 hex values or descriptive colors, 5)
composition and aspect ratio. Format as a single comma-separated
prompt.

Analyze this illustration and write a GPT Image 2 prompt to
reproduce its style with a different subject. Identify: 1) the
art style (Studio Ghibli, Pixar, ukiyo-e, watercolor, etc.), 2)
line work and texture, 3) color palette, 4) lighting/mood, 5)
composition. Format as a single comma-separated prompt with
[SUBJECT] as a placeholder I'll replace.

Analyze this commercial image and identify the brand visual
language. Write a reproducible GPT Image 2 prompt covering: 1)
subject placement, 2) background and props, 3) lighting setup
(e.g., softbox, hard light, natural), 4) color palette of the
brand, 5) typography placement (where text appears, even if I'll
add the actual text later), 6) overall mood. Format as a
comma-separated prompt.

import os
from openai import OpenAI

client = OpenAI()

def reverse_prompt(image_path: str) -> str:
    with open(image_path, "rb") as f:
        image_data = base64.b64encode(f.read()).decode()
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[{
            "role": "user",
            "content": [
                {"type": "text", "text": "Write a detailed comma-separated GPT Image 2 prompt to reproduce this image's style."},
                {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{image_data}"}},
            ],
        }],
    )
    return response.choices[0].message.content

# Build a prompt library from a folder of reference images
prompts = {}
for filename in os.listdir("references"):
    if filename.endswith((".jpg", ".png", ".jpeg")):
        prompts[filename] = reverse_prompt(f"references/{filename}")

# Save as a JSON prompt library
import json
with open("prompt_library.json", "w") as f:
    json.dump(prompts, f, indent=2, ensure_ascii=False)