This article is also available in:

Writing a good prompt

A good prompt describes what you want to see, like a brief to a photographer: the subject, its details, the setting, the lighting and the style. No technical jargon needed: clear sentences are enough.

Building your prompt in four layers

Let's take the example of a cosmetic product and enrich the prompt step by step.

1. The subject

Start with what should be at the center of the image.

Liquid cosmetic product presented as a minimalist studio packshot

The subject alone

2. The details

Materials, colors, outfit, expression: what makes the subject unique.

Copper and gold liquid cosmetic product presented as a minimalist studio packshot

With details

3. The setting and context

Where the subject is, at what time, in what situation.

… in a bathroom

With a setting

4. The style and mood

Lighting, rendering, atmosphere: "cinematic", "natural morning light", "dark mood, fog".

With a style

The choice of model matters as much as the prompt: a photorealistic model will interpret the word "illustration" less well. See Which image model to choose?.

With a Seelab model: describe the scene, not the product

Your Seelab model already knows your product: its shape, its label, its colors. If you describe it in the prompt, you risk distorting it.

Avoid: "a 30 ml amber glass bottle with a gold dropper and a white label, on a marble table"Prefer: "the bottle on a marble table, morning light, leaf shadows, tight framing"

Writing text in the image

Put the text to display in double quotes so it is reproduced exactly:

A man holding a sign that reads "Parlez-vous français ?"

Text in quotes

You can specify the typography: "in capital letters", "handwritten", "in embossed gold letters".

Describing colors

Models understand colors described in words well. They respect the shade of a HEX or RGB code, but not its exact value.

Prefer "dark orange background" to "#FF8C00 background".
To stay within your brand colors, combine the description with the palette (up to 3 colors, next to the prompt).

Words that make a difference

Photorealistic, hyper-detailed: a rendering close to a real photo.
Cinematic: contrasted lighting, film-like composition.
Depth of field: a blurred background that makes the subject stand out.
Natural light, soft light, backlight: the lighting mood.
Tight shot, wide shot, top view: the framing.

Tips

Be precise but concise: one or two well-chosen sentences are better than a long list.
Change one thing at a time from one generation to the next to understand what works.
Type @ in the prompt to insert a model, an image or a color.
The prompt of an image in your feed can be reused in one click: see Your feed and image actions.

Glossary

Prompt: the text that describes the image to generate, up to 3,000 characters.
Framing: the part of the scene visible in the image (tight shot, wide shot…).
Palette: up to 3 colors enforced at generation.
Seelab model: a model trained on your products, characters or styles.

Updated on: 31/10/2026

Was this article helpful?

Thank you!