How to Write Portrait Prompts for Photorealistic AI Images

A practical guide to prompt structure for realistic AI portraits — based on the 100 prompts published in this library, generated with the MOSS-Image model.

1. Start With a Photography Genre, Not an Adjective

The single biggest lever: begin your prompt with candid documentary photograph, photorealistic high-resolution photograph or DSLR portrait. Genre words pull the model toward real camera framing; adjective stacks like "beautiful, stunning, perfect" push it toward the glossy AI look everyone recognizes — and skips.

2. Anchor One Person, One Context

Specify a single subject, an age range, a descent, then hair, then wardrobe, then expression, then background. Each clause answers one question a photographer would ask: who, wearing what, feeling what, standing where.

3. Add One Imperfection

Symmetrical, flawless faces are what makes AI portraits feel fake. Our library includes exactly one deliberate flaw per portrait — a beauty mark, a tooth gap, freckles, a small scar. This single clause does more for realism than any quality keyword.

4. Describe Skin Like a Camera Would

Phrases like natural skin texture with visible pores and slight natural imperfections steer the model toward photographic detail rather than painted smoothness.

5. End With Camera & Light

Close with lighting (soft natural lighting, overcast window light) and lens language (shot on a DSLR, sharp focus). Light is half of photorealism.

A Real Example From This Library

Candid documentary photograph of a single real young East Asian woman around 20 years old of Chinese descent, raven-black hair, twin tails, hazel eyes, pale luminous skin. Wearing a mechanic's blue overalls. a confident …

Full prompt and image: see portrait #101 — then hit any detail page's How to Modify section to remix it.

Practice

Copy any prompt from the library, change one clause at a time (hair color → outfit → background), and re-run. One change per generation teaches you what each clause controls — the fastest way to build prompt intuition.