# Photoreal AI image prompts that work on Image One — lens, light, CFG

> Lens and light vocabulary for photoreal AI images, negative prompts, CFG scale, reference photos that hold a face, and 15 example prompts for Image One.

Published 2026-09-05 · eroq.ai — canonical: https://eroq.ai/blog/photoreal-ai-image-prompts


A photoreal prompt is a photographer's shot note, not a wish list. Prompts that come out of [Image One](/models/eroq-image-one) looking like photographs name a lens, a light and a surface; prompts that come out looking like renders say "realistic, 8k, hyperdetailed" and hope. Here is the vocabulary, the settings, and 15 prompts you can paste into [the Image studio](/studio/image) as written.

## The shape of a photoreal prompt

One paragraph, 30 to 80 words, in this order: **subject** with two or three physical details, **setting** in a phrase, **lens** (focal length and aperture), **light** (source and quality), one or two **surface or atmosphere** details, and a two-word **mood**. Skip any part and the model fills it with its own default, which is fine for the setting and fatal for the light.

> Close portrait of a woman in her forties with silver-streaked hair and freckles, 85mm lens at f/1.8, soft window light from camera left, plain gray backdrop, faint film grain, calm and direct gaze.

The words "photo" and "realistic" are not in it. The lens and the light carry the realism.

## Lens vocabulary

Naming a lens tells the model how much of the world to include and how much to blur. You are describing a look, not picking glass.

- **24mm or 28mm** — wide interiors and landscapes, slight stretch at the edges, everything in focus.
- **35mm** — street and environmental portraits, the person and their place.
- **50mm** — the neutral option; use it when you do not know.
- **85mm or 105mm** — portraits, flattering compression, background dissolves.
- **Macro (60mm to 100mm)** — products, textures, food, jewelry.
- **f/1.4 to f/2** — shallow focus, one plane sharp, everything else soft.
- **f/4** — the subject and a little context.
- **f/8 to f/11** — landscapes, products, anything that should be sharp end to end.

Do not add "shallow depth of field" on top of a wide aperture; it says the same thing twice and the blur goes cartoonish.

## Light vocabulary

Light is where photoreal prompts are won. The seven lighting names in the video studio's director's rack are also the most reliable words to type into an image prompt:

- **Window light** — soft, directional, the default for portraits and interiors.
- **Soft cross** — two soft sources at an angle; clean commercial look for faces and products.
- **Contre-jour** — light behind the subject, rim on hair and edges, flare if you ask for it.
- **Practicals** — lamps, candles, screens and signs inside the frame doing the lighting.
- **Overhead fall** — top-down light, shadows under brows and cheekbones; dramatic, unflattering on purpose.
- **Silhouette** — subject dark against a bright ground.
- **Neon glow** — colored signs as the key light; say the colors.

Then qualify it with "hard" or "soft", the time of day, and what the light lands on. "Overcast diffuse light" and "hard noon sun" are two different photographs of the same street.

## Negative prompts and CFG on Image One

All three image models take a negative prompt and a CFG scale. For photoreal work the negative list is short and stable; keep one and reuse it:

> painting, illustration, cartoon, 3d render, plastic skin, waxy, oversaturated, extra fingers, deformed hands, blurry, text, watermark, logo

Add per-shot terms ("glasses", "hat", "crowd") on top rather than growing it forever; a long negative list fights the positive prompt and flattens the image.

CFG sets how literally the model follows your words. Leave it at the default for a first batch, nudge it up if the image ignores a detail you named, down if skin goes crunchy or colors oversaturate. Photoreal work usually wants CFG a notch lower than illustration, because a photograph is allowed to be a little loose. The [CFG scale](/glossary/cfg-scale) and [negative prompt](/glossary/negative-prompt) glossary entries go deeper.

## Holding a face

The same person has to look the same in every render. Two fixes, both in the studio:

- **Reference photos.** Attach one or more photos and the model holds the face across renders. This is the [reference image](/glossary/reference-image) control; use a clean, front-lit source.
- **Characters.** Save a name, a persona and reference photos once in [Characters](/studio/characters), then @mention the character in any prompt. Same face across images and clips, and the character can carry a voice too.

Render in batches of 1 to 4 to compare takes at the same settings; every image is 1024×1024 on Image One at 10 credits, saved to your library with its recipe, and aspect ratios run 1:1, 3:4, 4:3, 9:16 and 16:9. For character products the full workflow is in [image generation for characters](/blog/image-generation-for-characters).

## 15 photoreal prompts for Image One

**Portraits**

1. Close portrait of a woman in her forties with silver-streaked hair and freckles, 85mm lens at f/1.8, soft window light from camera left, plain gray backdrop, faint film grain, calm and direct gaze.
2. A bearded fisherman in a wet orange oilskin laughing on a harbor wall, 50mm lens, overcast diffuse light, salt spray on the lens, boats softly blurred behind him, weathered and warm.
3. Studio headshot of a young man in a charcoal turtleneck, 105mm lens at f/4, soft cross lighting with a gentle rim, dark seamless background, catchlights in both eyes, confident and quiet.
4. A dancer resting on a rehearsal room floor with her back against the mirror, 35mm lens, overhead fall light from a skylight, chalk dust in the air, sweat on her collarbone, tired and satisfied.

**Street**

5. A cyclist waiting at a crosswalk in Tokyo rain at night, 35mm lens at f/2, neon glow from a pharmacy sign reflected in the wet street, motion blur on a passing taxi, restless and bright.
6. Two old men playing chess on a park bench in autumn, 70mm lens, low warm sun through bare trees, long shadows on the gravel, shallow depth of field, patient and companionable.
7. A street vendor's hands slicing mango on a wooden board, 60mm macro lens, midday sun filtered through a striped tarp, juice glistening, focus on the knife edge, vivid and busy.

**Product**

8. A matte black wristwatch on a slab of raw slate, 100mm macro lens at f/8, single soft cross light with a hard rim on the bezel, dark gradient background, precise and dust-free.
9. A glass of iced coffee on a marble counter with condensation beading, 50mm lens at f/2.8, morning window light from behind, faint kettle steam in the background, fresh and clean.
10. A pair of tan leather boots on a wet cobblestone street, 35mm lens at knee height, overcast soft light, puddle reflection beneath them, fresh mud on the welt, rugged and honest.

**Interiors**

11. A reading nook with a worn green armchair under a tall window, 24mm lens, late afternoon window light throwing a warm rectangle on the floor, dust in the beam, books stacked on the sill, lived-in and quiet.
12. A hotel bar at closing time, empty leather stools and one bartender polishing a glass, 35mm lens at f/2, practical lamps and backlit bottles, deep shadows in the corners, tired and elegant.

**Landscape**

13. A gravel road through a Scottish glen after rain, 24mm lens at f/11, contre-jour sun breaking through low cloud, wet stone shining, a few sheep in the middle distance, vast and clean.
14. A lighthouse on a basalt cliff in a winter storm, 70mm lens, gray diffuse light with one warm beam, spray frozen mid-air, long-exposure softness on the waves, fierce and lonely.
15. A desert highway at blue hour with a single pickup, headlights on and far away, 50mm lens, twilight gradient from orange to deep blue, heat haze still on the asphalt, empty and calm.

## FAQ

### Should I write "photo" or "photorealistic" in the prompt?

Rarely. A lens, an aperture and a light source do the job better without pushing the model toward stock-photo stiffness. If a render still looks illustrated, add "film grain" and put "painting, illustration" in the negative prompt.

### Why do faces change between renders?

Because without a reference the model invents a new face every time. Attach reference photos, or save the person as a Character and @mention them; the face then holds across images and, through image-to-video, across clips.

### What does CFG scale do for photoreal images?

It controls how strictly the model obeys the prompt. Higher is more literal and more saturated; lower is looser and softer. For photographs, a notch below the default usually looks more natural than a notch above.

Paste any prompt above into [the Image studio](/studio/image), pick Image One, and render a batch of four.