AI Photorealistic Prompt Guide: How to Use Nano Banana and “A photograph of ~”

By15ml Updated on
A thumbnail image of a woman wearing glasses standing in the city, accompanied by AI-generated prompt guide text in a realistic style.

While latest image generation models like Nano Banana, which can be tested on LMArena, are gaining attention, the fundamental principles of prompting apply regardless of the model.

Fundamentally, there is no single correct answer for prompts.
This is because the weight of prompts varies depending on the model used, and the methods may continue to change in the future. The best prompt is simply one that is easy for you to use and quickly yields the desired results.
Whether it’s a sentence style, a tag style, Korean, or English, it effectively doesn’t matter.

However, if you just can’t get the results you want, it is good to structure your prompts to some extent. Doing so makes them applicable to any model and helps maintain consistent output quality. The content summarized below follows this principle. Refer to it when needed and apply it accordingly.

Go to Photorealistic Prompt Generator

Photorealistic Base Prompts

A photograph of a ~

While expressions like “photo style” or “photorealistic” exist, “A photograph of a ~” actually worked more consistently.
For example, you can use it like this:

A photograph of a Korean woman

Tip: The simpler the prompt, the greater the randomness.

By applying this prompt, you can also photorealize 2D illustrations or 3D character models. Simply attach the image and input the prompt as is. The images below are the results of applying the prompt to a 2D illustration to convert it into a photorealistic style.

Prompt Example

A photograph of a Korean woman closely resembling the attached original character.
Composition:high angle/low angle

2D Photorealism (Ex 1)

2D female character with bobbed hair against a neon city background, original image before photorealistic prompt conversion
Original character image in 2D anime style
Photorealistic image of a woman with bobbed hair generated by Nano Banana AI model, white background and upward gaze
Example of converting a 2D character to a photorealistic style using the Nano Banana AI model
Example of converting a 2D character to a photorealistic style, a woman with bobbed hair in a city background
Image reinterpreted in photorealistic style while retaining the characteristics of the 2D character

2D Photorealism (Ex 2)

Original 2D image of a blonde bob-haired character with glasses standing against a white background
Original 2D Character
Photorealistic image of a blonde woman with glasses generated by Nano Banana AI, Hanok street background
Blonde glasses character generated by Nano Banana AI
Blonde bob-haired woman with glasses, wearing a gray jacket with multiple badges, Hanok street
The badge details are cute

Tip: The same prompt structure can be applied to any model, including Imagen, GPT-Image-1, and Nano Banana.

Natural Scene Direction

candid shot

Just by adding the phrase “candid shot” to your prompt, you can create a natural look as if the subject is going about their daily life without being conscious of the camera.

TIP: Use this expression actively when you want to capture a moment where the subject doesn’t care about the camera.

Auxiliary Prompts

off-camera gaze

Looking away from camera

unposed posture

Unposed posture

hands mid-action

Hands in motion

slight motion blur (natural)

Slight motion blur (natural)

Prompt Example

A photograph of a Korean woman closely resembling the attached original character.
candid shot, cafe

Candid Shot Example

Blonde woman with glasses holding a coffee cup and smiling by a cafe window, candid shot photorealistic example
Let’s stage a natural moment.
Blonde bob-haired woman with glasses sitting by a cafe window holding a coffee cup with a smile, natural expression
A scene capturing a moment of sitting naturally in a cafe

Shot Size

When creating an image, shot size (distance between camera and subject) is a key element that affects the result more than you might think.
However, controlling it with prompts is quite tricky in practice. After some trial and error, I have summarized the expressions that work best based on current standards.

Since the impression changes significantly depending on framing, it is also good to refer to the Framing Prompt Guide to Change AI Image Quality.

Extreme Long Shot

tiny figure in a wide environment

Auxiliary Prompts

camera far away
vast [ENVIRONMENT]

Long Shot / Wide Shot

Full Shot (head to toe) with wide surroundings

Auxiliary Prompts

generous headroom and footroom
28–35mm wide
feet visible
ample footroom

Full Shot

full shot (head to toe)

Auxiliary Prompts

[35–50]mm normal perspective
shallow DOF
medium DOF

Medium-Full Shot

medium-full shot (from knees to head)

Auxiliary Prompts

from mid-thigh to head
knee-up
three-quarter length

Cowboy/American Shot is sometimes used interchangeably, but it is not recommended for LLM-based image generation models like Nano Banana or Imagen.

Medium Shot

medium shot (from waist to head)

Medium Close-Up

medium close-up (from chest to head)

Close-Up

close-up (full face)

Auxiliary Prompts

focus on the eyes
visible catchlights,
both eyes in focus

Close-Up Example

Close-up of a blonde bob-haired woman with glasses staring at the camera, Nano Banana photorealistic prompt example

Extreme Close-Up

extreme close-up (eyes only)
or (lips only) etc. (specify the target part)

Auxiliary Prompts

macro close-up
very shallow DOF
soft diffused light
micro texture detail

Head-and-Shoulders

head and shoulders (from shoulders to top of head)

Practical TIP: Since it works differently depending on the model, it is stable to always specify shot size and framing directly when using models with limited usage counts.

When Possible

full body
knee-up
waist-up
chest-up
head-and-shoulders

Viewpoint (Camera Position)

Basic Angle Prompts

Eye Level
High Angle
Low Angle
Extreme Low / High
Tilted / Dutch Angle

Applying Angle Prompts

Viewpoint (Where is the camera?)

from behind and slightly to camera-right (rear three-quarter view)

from in front and slightly to camera-left (front three-quarter view)

over-the-shoulder shot from behind the subject

Viewpoint Prompt Examples

Blonde woman with glasses sitting by a cafe window holding a coffee cup
Front three-quarter view, slightly tilted left
Back view of a blonde bob-haired woman sitting and looking out the window
Rear three-quarter view, slightly tilted right

Subject Direction (Where body/face is facing)

subject turned three-quarters toward camera-left

subject facing away from camera

subject glancing back over her shoulder

High Angle

Dramatic high-angle shot from the right side
High angle, three-quarter view from the right

From a high-right angle

High angle shot from the right side

Shot from a high angle on the right

Low Angle

Dramatic low-angle shot from the left side
Low angle, three-quarter view from the left

From a low-left angle

Low angle shot from the left side

Shot from a low angle on the left

Perspective/Distortion

high-angle/Low angle (presence up)

Normal Perspective : Realistic proportions

Mild Foreshortening (10~20%) : Slight perspective exaggeration effect

Strong Foreshortening (30~40%) : Dramatic effect where limbs or face pop out significantly

Wide-angle Distortion : Wide-angle lens distortion (24mm, emphasizes spatial sense)

Telephoto Compression : Telephoto lens compression (85mm+, appears flattened, spatial compression)

Example Prompt

Medium-Full Shot (from knees to head), Low angle (presence up), strong foreshortening(~30%)

Depth of Field (DOF)

Deep DOF : Full focus, for landscapes
Shallow DOF : Emphasizes subject
Medium DOF

Lighting

Soft Lighting – Soft and subtle light

Hard Lighting – Strong contrast, distinct shadows

Dramatic Lighting (Rembrandt/Chiaroscuro) – Dramatic chiaroscuro

Natural Light – Sunlight, window side

Fluorescent Lighting / Fluorescent practicals – Fluorescent lights, realistic atmosphere

Are numeric prompts like 80 degrees or 30% effective?

In conclusion, they are very effective. However, they do not work with the mathematical precision we might expect. It is not a structure that executes commands like “tilt camera exactly 80 degrees” or “apply 30% perspective exaggeration” to the AI.

These numbers act more as directives for directionality rather than precise commands to the AI.

The operating principle of AI is based on “association.”

Some AI models understand these numerical expressions better, while others may almost ignore them.

How to use: Use numbers together with auxiliary prompts. To maximize the effect of numeric prompts, it is best to use numbers combined with descriptive expressions.

Bad Example (Using numbers only)

Medium-Full Shot, Low angle 80°, foreshortening 30%

Good Example (Combining numbers and description)

Dramatic low-angle shot, from a steep 80° upward perspective, creating strong foreshortening(around 30%)

Sample Combination Examples

[Perspective Boost: ON] 80% exaggeration
[Camera Angle] Low Angle, about 85° upward perspective
[Shot Size] Medium Full Shot (from knees to head)
[Perspective] Strong Foreshortening (80%)
[Lens] 24mm wide-angle cinematic distortion
[Depth of Field] Shallow DOF, subject sharp, background softly blurred
[Lighting] Dramatic lighting, high contrast shadows

In summary, you can create the desired image by combining composition, distance, and viewpoint. Please refer to the separate post about the latest model, Nano Banana.

Summary FAQ

Starting the prompt with “A photograph of a ~” provides the most stable and consistent photorealistic results. It is often more effective than words like “photorealistic.”

The most reliable way is to specify the range of the shot. Instead of simply writing “Full Shot,” using parentheses to specify the range, like “full shot (head to toe),” helps the AI draw the full body more accurately.

Since most major AI models are trained on English, writing in English is the most stable. While Korean is supported, if the desired scene does not appear correctly, it is recommended to write in English.

Both distort space, but their focus is different. “Wide-angle distortion” distorts the surrounding background widely like a wide-angle lens, emphasizing the “sense of space.” On the other hand, “strong foreshortening” draws parts of the subject close to the camera (e.g., fists, feet) extremely large, emphasizing “three-dimensionality” and “dynamism.”

추천 포스트