> 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/using-ideogram/getting-started/prompting-guide/in-a-nutshell.md). # Quick Summary ## Ideogram prompting in a nutshell This guide covers everything you need to write effective prompts for Ideogram! I wanted to go over all of the available models, from the basics of **natural language prompting used for Ideogram 2.0 and 3.0** to the full **structured JSON format introduced in Ideogram 4.0**. Whether you're just getting started or looking to unlock the full precision of Ideogram's latest model, this guide walks you through it step by step. *** ## Natural language **Ideogram can understand plain, everyday language.** No need to use weights, parameters, or technical syntax—just describe what you want as clearly as possible, the way you’d describe something to a person. *** ## Be clear and visually grounded The more your prompt describes things that can actually be seen—like shapes, colors, materials, lighting, and background—the better the AI can understand and render them. **Focus on:** * The main subject (who or what is it?) * What it’s doing (pose or action) * Where it is (setting or background) * How it looks (style, color, emotion, lighting) Example: *“A watercolor painting of a fox curled up in snow under a pine tree, with soft blue shadows and falling snowflakes.”* *** ## Use a structure You don’t need perfect grammar, but to get the best results, structure your prompt using a clear format. [Section 3](/using-ideogram/getting-started/prompting-guide/3-prompt-structure.md) walks you through each part, but here's the general idea: * **Image summary**: A concise phrase describing the type of image * **Main subject**: Who or what the image is about * **Pose or action**: What the subject is doing * **Secondary elements**: Anything else around the main subject * **Setting & background**: Where it’s taking place * **Lighting & atmosphere**: How the scene feels * **Framing & composition**: Camera angle or layout (optional) Here’s a complete structure that works well: > *\[Image summary]. \[Main subject details], \[Pose or action], \[Secondary elements], \[Setting & Background], \[Lighting & Atmosphere], \[Framing & Composition]* For example: > *A watercolor painting of a girl flying a kite on a hill at sunset, wearing a red dress, wind blowing through her hair, with trees in the background and soft golden lighting.* No need to use every part—keep it short when exploring, and add more parts/details when you need control. *** ## For maximum control: use JSON in Ideogram 4.0 Ideogram 4.0 was trained on structured JSON captions. If you need precise layout control, exact color palettes, or precise text placement, JSON prompting gives you capabilities that plain text simply can't match. **See Section 4 for the full guide.** [Magic Prompt](/using-ideogram/getting-started/prompting-guide/7-creative-tools-in-ideogram/using-magic-prompt.md) can also expand a natural language text prompt into a full JSON caption automatically — so you can get JSON-quality results without writing JSON yourself. *** ## Main ideas come first Ideogram tends to give slightly more importance to parts placed earlier in the prompt. This is where to write the subject. *** ## Generating text in images If you want text in your image, describe it near the beginning of the prompt — this leads to better results with fewer errors. Use quotes `" "` around the desired text. The longer and more complex the text, the more likely it is to produce errors. **For pixel-perfect text placement, consider using JSON prompting with Ideogram 4.0** (Section 4). *** ## Handle negatives * Don’t say *“no people,” “man without a beard,” “no background”* or *“without trees.”* * Instead, say exactly what you want (*“empty street,” “clean-shaven man,” “a plain white background,” or “desert landscape”*). *** ## Be specific and detailed, but not overloaded Ideogram generally performs best with plain-text prompts **under approximately 150 words** (around 200 tokens). If the prompt is longer, the AI may begin to misinterpret or ignore the content beyond that point. *** ## Experiment, iterate, tweak If something looks off, tweak one thing at a time: * Reword key phrases, try synonyms or analog concepts * Emphasize or remove a detail * Use a different style * Change the aspect ratio to change framing **Prompting is a creative process**—experiment, learn, and have fun. *** ## Use creative tools * **Magic Prompt**: Enhances and expands your original prompt, and can automatically generate a full JSON caption for Ideogram 4.0. * **Reference Style**: Applies the visual style from an image (generated or uploaded) to the image you are generating. * **Random Style**: Same as Reference Style but with a random image. Perfect for wild visual explorations. * **Remix**: Adjust and tweaks existing images using similar or different prompts. Combine all of these for surprising, creative outcomes. *** {% hint style="success" %} That’s it! :tada: Once you’re comfortable, the full guide below can help you go deeper and get even better results. But for now, just start prompting and see what happens. **Have fun!** {% endhint %} Try Ideogram