photo-realism

Installation
SKILL.md

Photo Realism

A typical "photorealistic, 8k" prompt next to the same request rewritten by this skill

Left: the prompt most people write, "photorealistic photo of a beautiful girl on the street, 8k…". Right: what the same image model returns once this skill has rewritten the prompt. Both prompts, and a second example, are at the end of this file.

"Photorealistic, 8k, ultra detailed, cinematic lighting" is how people ask for realism, and it's exactly what makes the image look generated. Those words describe a render, so the model returns one: golden glow, creamy bokeh, a model's face, an empty perfect street. A real photo is different because of how it was taken: who held the phone, from where, what was happening around, what the person was doing. So this skill doesn't add quality adjectives. It rewrites the prompt into a description of a moment someone photographed.

There are two ways to fail, and the skill has to avoid both:

  • Too clean: stock-photo light, a centred model staring at the lens, a spotless empty street, everything perfect. Reads as AI.
  • Too real: grey sky, bins and vans, awkward framing, an unflattering moment. Reads as a bad photo nobody would post.

The target is the photo people actually save: real, and beautiful. A nice place with real life in it, soft or golden light, a person doing something, a good outfit, framed the way a friend frames it.

Workflow

1. Parse the input

Installs
11
GitHub Stars
195
First Seen
Oct 18, 2026