> For the complete documentation index, see [llms.txt](https://docs.ideogram.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.ideogram.ai/prompting/prompting.md).

# Prompting Basics

Learn how to write Ideogram image prompts, choose plain text or JSON, use Magic Prompt, and separate image descriptions from generation settings.

A prompt is the text you give Ideogram to describe the image you want. It can be one line or a full paragraph, plain or poetic. Start here for prompt tips and how to write prompts that get the image you want. This guide covers the Ideogram models in the Image app: Ideogram 4.5, Ideogram 4.0 and Ideogram 3.0.

## How Ideogram reads a prompt

Ideogram uses your prompt to guide the image's subjects, setting, style and visible text. Specific descriptions give you clearer details to evaluate, but a prompt is a request rather than a guarantee that every detail will be rendered.

> A tall man in a red coat walking through a snowy forest.

From this prompt Ideogram will try to draw the height, the red coat, the walking, the snow and the forest. Leave out a detail and the model picks one for you. Add a word you don't need and it may show up in the image.

A few habits follow from this:

* **Write in sentences.** Tag lists like `man, forest, fire, dramatic, painting` can work, but full sentences tell the model how things relate: who holds what, what sits behind what.
* **Describe what can be seen.** "A sunset over the ocean with orange and pink light on the water" gives the model more to draw than "a beautiful scene." [Prompt Structure](/prompting/prompt-structure.md) covers this in depth.
* **Put the main idea first.** Words near the start of the prompt tend to carry more weight.
* **Say what you want, not what you don't.** "An empty street" works better than "a street with no people."

There is no hidden syntax. Ideogram doesn't read weights like `::1` or `(important)` or flags like `--ar` and `--style` as commands; it treats them as words. Set aspect ratio, seed and quality in the Image app settings instead.

You can write in any language. English gives the most reliable results, most of all when the image includes text. Magic Prompt can translate your prompt into English for you.

## Plain language or JSON

Ideogram 4.5 and 4.0 accept either a plain-language prompt or a structured JSON prompt. Start with plain language on Ideogram 3.0. JSON can describe layout, text placement and requested hex colors in separate fields; it doesn't turn image generation into a fixed layout engine. See [JSON Prompting](/prompting/json-prompting.md).

## What belongs in the prompt and what belongs in settings?

| Put in the prompt                         | Set in the Image app                               |
| ----------------------------------------- | -------------------------------------------------- |
| Subject, action, scene and lighting       | Model and supported reference images               |
| Words that should appear in the image     | Aspect ratio, output size or resolution            |
| Intended lettering and illustration style | Number of images, quality and seed                 |
| What should change in a reference image   | Background transparency, where the model offers it |

Writing **16:9**, **four images** or **transparent PNG** in a description doesn't replace the corresponding app control. Use the model's [generation settings](/create/image-settings.md) and the [download menu](/organize-and-share/download.md) for output and file choices.

## Magic Prompt

Magic Prompt rewrites and expands a short prompt before Ideogram generates the image. On Ideogram 4.5 and 4.0 it can turn a plain-text idea into a structured prompt. Turn it off when you want to evaluate your own wording without that expansion. Availability depends on the model; on Ideogram 4.5 it disappears when references make the prompt an edit instruction. See the [Magic Prompt setting](/create/image-settings.md#magic-prompt).

## Start here

1. Pick a model: Ideogram 4.5, 4.0 or 3.0. See [available models](/create/models.md).
2. Write one sentence that says what kind of image it is and what it shows, such as `A watercolor painting of a fox curled up in snow under a pine tree.`
3. Put any text you want in the image in quotation marks, near the start.
4. Generate, look at the results, then add detail only where the image missed.
5. Change one thing at a time so you can tell what made the difference.

## Next pages

1. [Prompt Structure](/prompting/prompt-structure.md)
2. [Text in Images](/prompting/text-in-images.md)
3. [JSON Prompting](/prompting/json-prompting.md)
4. [Refine and Iterate](/prompting/refine-and-iterate.md)
5. [Fix Common Problems](/prompting/fix-common-problems.md)
6. [Vocabulary Reference](/prompting/vocabulary.md)

## In this topic: creating images

1. [Create Images](/create/create.md): make your first image
2. [Models](/create/models.md): pick the model for the job
3. [Image Settings](/create/image-settings.md): set size, quality and other options
4. **Prompting Basics** (this page): write a prompt that works
5. [Fix Common Problems](/prompting/fix-common-problems.md): fix results that come out wrong