Skip to content
OpenDream AI

AI generator

Text to Image: Turning a Description into a Picture

Generating from text alone means the words carry everything, so the way you write matters. This page covers how models read a prompt, what order to put details in, and how to fix results that miss the point.

Made with OpenDream

See more in the gallery

How it works

  1. 01

    Lead with the subject

    Put the main subject and what it is doing in the first few words. Models weigh early words more heavily, so 'an old fisherman mending a net' should come before lighting or style.

  2. 02

    Add setting, light and framing

    Follow with where it happens, the light source and the viewpoint: 'on a pier at dawn, low sun from behind, close-up at eye level'.

  3. 03

    End with style, then iterate

    Close with medium and mood, generate, and change one phrase at a time. If the wrong element is emphasized, move it earlier or describe it more concretely.

Why create text to image with OpenDream

How to structure a prompt so the model follows it

A dependable structure is subject, action, setting, light, framing, style. The model does not enforce it, but it mirrors how people describe pictures and stops you from forgetting a layer. 'A girl on a bicycle' becomes 'a teenage girl riding a bicycle with a basket of oranges, cobbled hill street, late afternoon backlight, wide shot from street level, warm film photograph'. Every added phrase removes a guess.

Be concrete about what matters and quiet about the rest. If the bicycle's color is important, state it; if not, leave it out rather than padding with 'beautiful, amazing, highly detailed'. Stacked quality words dilute the prompt and rarely improve results on current models.

Describe what you want rather than what you do not want. 'An empty beach' tends to beat 'a beach with no people', because mentioning people can pull them into the frame. When something keeps appearing, replace it with a positive description of what should fill that space.

For lettering, put the exact words in quotation marks and keep them short, ideally under six words. Say where the text sits and how it looks: 'the words "Night Market" in hand-painted red lettering on a wooden sign above the stall'. Proofread every result, since even strong typography models occasionally swap a letter.

Prompt examples

Start from a template

Frequently asked questions

How long should a prompt be?
Long enough to cover what matters. One to three sentences handles most images. Longer prompts can work on models that follow detail closely, but the important elements should still come first.
Why does the model ignore part of my prompt?
Usually the missing detail was late in the prompt, vague, or in conflict with something else. Move it earlier, make it concrete, or remove the competing instruction.
Can I include a title or words in the picture?
Yes. Quote the exact text and keep it short. Ideogram V3 and GPT Image 2 handle lettering best; check the spelling in every result.
Do I need negative prompts?
Rarely. Describing the scene you want in positive terms is more reliable than listing things to avoid.

Keep exploring

Ready to create?

Pick a plan and you only spend credits on images you actually receive.

Start creating