Reverse Engineer an Image Prompt

[Attach the reference image, then use this prompt:]

Analyze this image and write me a prompt that would reproduce its look with a different subject.

First, describe what you actually see, in these terms:
- **Subject and composition**: what is in frame, where it sits, how much of the frame it fills, the crop and aspect ratio, the eye level and angle.
- **Lighting**: the direction and height of the key light, how hard or soft the shadows are, whether there is fill, rim, or backlight, the color temperature, and where the brightest and darkest points fall.
- **Lens and depth**: the apparent focal length, how compressed or wide the perspective looks, how shallow the focus is and where the focal plane sits.
- **Color**: the palette in 3-5 named colors, the overall grade (warm, cool, muted, high contrast, lifted blacks), and any color cast.
- **Texture and medium**: photographic, illustrated, rendered, painted; grain, noise, halation, brushwork, edge quality.
- **The signature detail**: the single element most responsible for why this image looks the way it does. Usually one lighting or color choice, not the subject.

Then give me:

1. **The reusable prompt**, written as a scene description in natural sentences, with the subject as a [SUBJECT] placeholder and every style element stated explicitly. Keep the style sentences in a fixed order so I can swap subjects and hold the look.

2. **A version for [Midjourney / Nano Banana / Flux / Ideogram]**, in that tool's idiom: keyword-and-parameter style with aspect ratio and version flags for Midjourney, or narrative scene description for the others.

3. **The three phrases doing the most work**, so I know what to keep when I shorten it.

4. **What cannot be prompted**: anything in this image that comes from a specific reference, a real location, a post-process, or a lucky generation. Say so rather than writing a phrase that will not reproduce it.

Describe only what is visible. Do not guess at the tool, the model, or the original prompt.