<aside> 🖼️
What you describe is what you get. AI doesn't read between the lines. It reads the lines. This guide teaches you how to write image prompts that leave nothing to interpretation — so the output matches the intent.
</aside>
Before writing prompts, understand the core terms that shape how AI image generation works.
| Term | What it means |
|---|---|
| Prompt | The text input given to an AI to generate content. Think of it as instructions that tell the AI what you want it to create. |
| Model | A specific AI system trained to perform certain tasks. Different models excel at different types of content creation and have various strengths and limitations. |
| Sampling / Temperature | Controls how random or deterministic outputs are. Higher values produce more creative but less predictable results; lower values create more consistent outputs. |
| Parameters / Settings | Adjustable values that influence how the AI generates content — including image resolution, style emphasis, and generation quality. |
A prompt structure is a framework to organize your requests, ensuring you provide all necessary details for your desired output. A well-structured prompt leads to more consistent and accurate results.
<aside> 🎯
Follow the three-line structure. Separate your prompt into technical details, subject/composition, and style — each on its own line. This keeps instructions clear and easy to tweak.
</aside>
<aside> ✅
Do: 50mm lens, f/1.8 aperture, natural window light, warm color temperature A fluffy orange tabby cat sitting at a small wooden table in a cozy cafe, holding a tiny coffee cup with both paws Photorealistic style, warm earthy tones, soft bokeh, cozy intimate mood
</aside>
<aside> ❌
Don't: "A beautiful, stunning, amazing cat drinking coffee in a nice cozy cafe with gorgeous warm lighting and pretty bokeh and nice vibes."
</aside>
Same subject, same scene — but the first prompt tells the model exactly what to render. The second leaves everything up to guesswork.
[Technical details] + [Lighting]
[Subject / Main concept] + [Perspective / Composition] + [Action / Pose] + [Setting / Environment]
[Style / Aesthetic]
Technical: 50mm lens, f/1.8 aperture, natural window light
Subject: A fluffy orange tabby cat sitting at a small wooden table in a cozy cafe, holding a tiny coffee cup
Style: Photorealistic, warm earthy tones, soft bokeh, cozy mood
<aside> 💡
Sectioning prompts is key. Writing each component on its own line ensures the model processes instructions one piece at a time — not as a single wall of text.
</aside>
| Principle | How to apply it |
|---|---|
| Be specific | Include details about subject, setting, mood, lighting, style, and perspective. |
| Use clear structure | Start with the main concept, then layer in details following the three-line format. |
| Include technical details | Specify camera type, lens, lighting conditions, and other technical aspects for more control. |
| Reference styles | Mention specific art styles, artists, or aesthetics to guide the AI toward a particular look. |
| Use positive instructions | Tell the AI what you want rather than what you don't want (though some models handle negative prompting). |
| Subject-first approach | Begin with the main subject before adding details — "A majestic lion..." rather than starting with style or setting. |
| Weight important elements | Some models allow emphasis using syntax like (concept) or ((concept)) for stronger emphasis. |
| End with quality boosters | Terms like "highly detailed," "professional," or "award-winning" can improve overall output quality. |
Kittl supports two prompting approaches. The preferred structure separates each component onto its own line for better model comprehension.