A prompt is the written instruction you give an AI image model. It says what the picture should show and how it should look, and the model makes an image from those words. A prompt that names the subject, setting, camera and light leaves the model less to guess than one line of adjectives.
What a prompt is, and where it stops
In practice a prompt is a few sentences typed into a box. It is the main input, but not the only one. In AIGE you pick the model and the aspect ratio from menus, so neither has to go into the words. A picture you made earlier can go in beside the words as a reference, so the next run stays close to it.
A prompt does not return one fixed picture. AIGE picks a fresh seed on every run, so the same words give a new image each time (the seed entry in the glossary explains the mechanism). That is why you judge a prompt over several tries. Each try spends credits. Nano Banana 2 at 1K costs 8 credits, Nano Lite costs 4 credits and Nano Banana Pro at 1K costs 15 credits. The Generate button shows the price before you press it. The AI image generator page lists the models you can pick from.
A template with seven slots
Fill every slot. A slot you leave empty is a decision the model makes for you.
Subject and action. Who or what, doing one specific thing: "a man in a rain jacket checking his phone", not "a person outside".
Setting. A real place with two or three concrete details, such as "a tram stop after rain, wet pavement, a blurred tram behind". "A beautiful background" tells the model nothing.
Camera. A body class and a lens, for example "full-frame mirrorless, 85mm at f/1.8", "35mm at f/4 on a rangefinder" or "phone main camera, 26mm equivalent". Add the depth of field that follows from it: shallow at f/1.8, much deeper at f/8.
Light. One named source with its direction, quality and color: "window light from camera left, soft and diffused, late morning, about 5500K, no fill".
Angle and framing. Camera height and distance: "eye level, chest-up, subject 1.5 m away", or "overhead, 90 degrees, flat lay".
Real-world imperfection. What a real photo has: skin texture, flyaway hair, a crease in a collar, a dust speck, an uneven background.
Constraints. What must stay out of the frame, in one plain line: "no text, signage or logos in frame".
Here is a prompt built from the slots:
A man in his twenties in a rain jacket checks his phone at a tram stop after rain. 35mm lens at f/4 on a rangefinder, slight motion in his free hand. Overcast sky as one large soft source from above, wet pavement reflecting it, about 6500K. Eye level from across the platform, 3 m away. Water beads on the jacket, a scuffed shoe, a blurred tram behind. No readable signage.
Change one slot at a time between runs. Then you know which change did what.
Words to leave out
Skip filler such as "8k", "ultra detailed", "hyperrealistic", "masterpiece", "trending on artstation" and "award-winning". Each one names a quality you want, not something to put in the frame. The same goes for "cinematic" used in place of a light source, and for a string of adjectives where the light should be. Write the source, its direction and its color instead.
Leave the aspect ratio out as well. In AIGE it is a setting in the picker, not part of the sentence, and the aspect ratio guide lists the shapes the image models offer.
Getting more from the prompt box in AIGE
Be firm about direction. If the first results come back with flat, even light when you asked for window light, restate it as a hard constraint, for example "lit from the left, the right cheek a stop darker". Do the same for the subject: if the expression keeps drifting, name the one you want, such as "mouth closed, listening". Expect to throw some frames away.
Two helpers sit by the box. The sparkle button next to Generate opens the AI Prompt Assistant, a chat that can put the prompt it writes into the prompt box. Once a prompt works, save the prompt and model setup as a preset; the presets guide shows how. For ten prompts that were run in AIGE, each with the image it made, see the Nano Banana prompts page.
Honest note: The prompt above shows how to write one, not what any model will draw from it. Each run is a new image, so judge a prompt over several tries.
Quick answers
Question | Answer |
|---|---|
What is a prompt in AI image generation? | The written instruction an image model reads to make a picture. It says what the picture should show and how it should look. |
What should an AI image prompt include? | A subject and action, a setting, the camera and lens, one named light source, the angle and framing, one real-world imperfection and a line of constraints. |
Why does the same prompt give a different image each time? | AIGE picks a fresh seed on every run, so identical words start from a new pattern and end in a new picture. |
Do I put the aspect ratio in the prompt? | Not in AIGE. The aspect ratio is a setting in the picker, so the words only need to describe what is in the frame. |
What should I leave out of a prompt? | Filler like "8k", "ultra detailed" and "masterpiece", and "cinematic" used instead of a named light source. Say what is in the frame. |
What does it cost to run a prompt in AIGE? | Per image: Nano Banana 2 at 1K is 8 credits, Nano Lite is 4 credits and Nano Banana Pro at 1K is 15 credits. The Generate button shows the price first. |
Comments
Be the first to comment.