> ## Documentation Index
> Fetch the complete documentation index at: https://docs.dopaminegirl.com/llms.txt
> Use this file to discover all available pages before exploring further.

# SDXL Prompting Guide

> Write stronger SDXL image prompts with clear structure, negative prompts, and starter recipes.

# SDXL Prompting Guide

SDXL is a flexible text-to-image model for general creative work: photos, product concepts, realistic portraits, design mockups, cinematic scenes, and broad illustration styles.

SDXL usually works best with clear natural language. You do not need a huge pile of quality words. A specific subject, clear composition, and believable lighting usually matter more than repeating `masterpiece` or `ultra high quality`.

> Want anime, cartoon, furry, or tag-heavy character prompts? Use the [Pony Diffusion Prompting Guide](/tutorial/pony-diffusion).

## Prompt Formula

Use this structure:

```txt theme={null}
[subject] + [action/pose] + [scene] + [composition] + [lighting] + [medium/style] + [quality details]
```

Example:

```txt theme={null}
A modern perfume bottle on wet black stone, centered product photography, softbox reflection, shallow depth of field, cool blue rim light, luxury advertising style, crisp glass details
```

For best control, make the first words identify the main subject. Then add the frame, lighting, and visual style.

## Prompt Building Blocks

**Subject**

* Who or what is in the image.
* Add material, clothing, age range, product type, or defining traits.
* Keep one main subject unless you need a complex composition.

**Action or pose**

* Describe what the subject is doing.
* For portraits, use pose and expression: `looking at camera`, `relaxed smile`, `three-quarter view`.
* For products, use placement: `resting on marble`, `floating above a matte pedestal`.

**Scene**

* Name the location and background.
* Include important foreground or background objects only when they matter.

**Composition**

* Use camera and framing terms: `close-up`, `wide shot`, `centered composition`, `rule of thirds`, `negative space`.
* For photography, lens cues help: `35mm`, `50mm`, `85mm`, `macro lens`.

**Lighting**

* Use practical light descriptions: `softbox`, `golden hour`, `neon rim light`, `overcast daylight`, `hard side light`.
* Lighting often improves the image more than generic quality tags.

**Medium and style**

* Examples: `product photography`, `cinematic still`, `editorial illustration`, `fantasy concept art`, `watercolor`, `3D render`.
* Avoid conflicting styles like `photorealistic anime watercolor 3D render` unless you want a hybrid.

## Negative Prompts

Keep negatives short at first:

```txt theme={null}
blurry, low quality, distorted, extra fingers, bad hands, text, watermark, logo
```

For portraits, add:

```txt theme={null}
deformed face, asymmetrical eyes, bad anatomy, extra limbs
```

Avoid massive negative prompts unless you know what each token is doing. Large negatives can remove useful detail and make images look generic.

## Starter Recipes

**Product photo**

```txt theme={null}
Premium wireless headphones resting on brushed aluminum, close-up product photography, soft studio lighting, clean reflections, black and white color palette, sharp edges, commercial advertising composition
```

**Cinematic portrait**

```txt theme={null}
Portrait of a confident fashion model in a rain-soaked city street at night, 85mm lens, shallow depth of field, neon signs reflected in puddles, soft key light, natural skin texture, cinematic color grading
```

**Fantasy concept art**

```txt theme={null}
A lone explorer standing before a colossal stone gate carved into a mountain, wide establishing shot, morning mist, golden rim light, detailed environment, epic fantasy concept art, painterly realism
```

**Editorial illustration**

```txt theme={null}
Minimal editorial illustration of a glass greenhouse floating above a dense city, clean geometric shapes, muted color palette, soft shadows, modern magazine cover composition
```

## Troubleshooting

**The image ignores the subject**

* Put the subject first.
* Remove conflicting style words.
* Make the scene simpler.
* Describe the subject with concrete visual traits.

**The image looks plastic or oversharpened**

* Remove stacked quality tokens.
* Use natural lighting terms instead of only `high detail`.

**The composition is messy**

* Use one clear shot type.
* Reduce the number of subjects and props.
* Add framing language like `centered`, `close-up`, `wide establishing shot`, or `clean negative space`.

**Text in the image looks wrong**

* Avoid asking SDXL to render long exact text.
* Use simple labels only when needed.
* Add exact typography later with an editor when accuracy matters.

## Final Checklist

* Put the subject first.
* Add composition and lighting before generic quality words.
* Use a short negative prompt, then add only the fixes you need.
* Remove conflicting style words before adding more detail.

## References

* [SDXL base model card on Hugging Face](https://huggingface.co/stabilityai/stable-diffusion-xl-base-1.0)
* [Stability AI SDXL AWS documentation](https://stability.ai/sdxl-aws-documentation)
* [SDXL research paper](https://arxiv.org/abs/2307.01952)