> ## 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.

# Basic Prompting Guide

> Learn how to structure positive and negative prompts, control emphasis, and improve image results through focused iteration.

# Basic Prompting Guide

A prompt tells the image model what to generate. Dopamine Girl runs image workflows on ComfyUI, where the positive prompt describes what you want and the negative prompt describes what you want the model to avoid.

ComfyUI passes those prompts to the selected model. The model still determines which words, tags, and settings work best, so start with this general method and then use the guide for your model when you need more control:

* [SDXL Prompting Guide](/tutorial/sdxl) for photos, products, cinematic scenes, and general illustration.
* [Pony Diffusion Prompting Guide](/tutorial/pony-diffusion) for anime, cartoon, furry, and tag-heavy character work.
* [Anima Prompting Guide](/tutorial/anima) for anime and non-photorealistic illustration.

## Start With a Clear Structure

Build the positive prompt in this order:

```txt theme={null}
[subject] + [action or pose] + [environment] + [composition] + [lighting] + [style or medium]
```

Example:

```txt theme={null}
A red fox sitting beside a mountain stream, pine forest in the background, medium shot, soft morning light, detailed storybook illustration
```

Put the main subject first. Add details that can be seen in the image, and remove words that do not change the intended result.

## Build the Prompt One Layer at a Time

Start with the subject:

```txt theme={null}
A red fox
```

Add an action and environment:

```txt theme={null}
A red fox sitting beside a mountain stream, pine forest in the background
```

Add composition and lighting:

```txt theme={null}
A red fox sitting beside a mountain stream, pine forest in the background, medium shot, soft morning light
```

Finish with a visual medium or style:

```txt theme={null}
A red fox sitting beside a mountain stream, pine forest in the background, medium shot, soft morning light, detailed storybook illustration
```

This approach makes problems easier to diagnose. If the final image drifts away from your idea, remove the newest phrase or replace it with a more concrete description.

## Describe Visible Details

Use specific visual language:

| Vague                | More controllable                                                 |
| -------------------- | ----------------------------------------------------------------- |
| `beautiful woman`    | `woman with a short black bob, green jacket, calm expression`     |
| `nice lighting`      | `soft window light from the left, warm rim light`                 |
| `cool background`    | `rainy neon street with blurred signs and wet pavement`           |
| `professional photo` | `studio product photo, centered composition, softbox reflections` |

Useful detail groups include:

* Subject traits: clothing, colors, materials, age range, hair, and expression.
* Action: standing, running, holding an object, or looking in a direction.
* Framing: close-up, portrait, full body, wide shot, overhead view, or centered composition.
* Lighting: golden hour, overcast daylight, softbox, hard side light, or neon rim light.
* Medium: photograph, watercolor, anime illustration, 3D render, or cinematic still.

Avoid stacking contradictory instructions such as `photorealistic watercolor anime 3D render` unless a mixed style is intentional.

## Use Negative Prompts for Exclusions

The negative prompt should identify unwanted content or defects. Start short:

```txt theme={null}
blurry, low quality, text, watermark, distorted anatomy, extra limbs
```

Add a term only when you see a recurring problem. A large generic negative prompt can suppress details you wanted and make results less distinctive.

Write the desired alternative in the positive prompt whenever possible. For example, if you do not want a busy background, add `clean plain background` to the positive prompt instead of relying only on `cluttered background` in the negative prompt.

Some newer model families use negative prompts differently or may ignore them. If changing the negative prompt has no effect, check the model-specific guide rather than adding more terms.

## Emphasize Important Terms

In ComfyUI workflows that use CLIP prompt weighting, this syntax increases or decreases emphasis:

```txt theme={null}
(golden hour:1.2)
(background details:0.8)
```

* Values above `1.0` increase emphasis.
* Values below `1.0` decrease emphasis.
* `1.0` is the normal weight.

Use small adjustments such as `1.1` or `1.2`. Very high weights can create harsh colors, artifacts, or an image that ignores the rest of the prompt. Weighting behavior can differ when a workflow uses another text encoder, so treat it as an optional correction rather than the foundation of the prompt.

## Generate Variations With Prompt Choices

Dopamine Girl supports dynamic prompt choices. Put alternatives inside braces and separate them with `|`:

```txt theme={null}
{option one|option two|option three}
```

Before an image is generated, one alternative is selected from each brace group. This lets one prompt produce different combinations of subjects, outfits, environments, lighting, poses, or styles.

```txt theme={null}
a {ginger|blonde|brunette|black-haired} beautiful girl, {wearing a flowing summer dress|wearing an elegant evening gown|wearing a cozy oversized sweater|wearing a futuristic cyberpunk outfit}, {standing in a flower garden|walking through a neon city street|sitting by a quiet lakeside|posing in a sunlit bedroom}, {soft natural lighting|cinematic golden hour lighting|dreamy pastel lighting|moody dramatic lighting}, highly detailed, beautiful face, expressive eyes, realistic, 8k
```

This example has four hair choices, four outfit choices, four scene choices, and four lighting choices, allowing up to 256 combinations.

Keep each group focused on one type of detail:

```txt theme={null}
Portrait of a {smiling|serious|thoughtful} astronaut, {close-up|medium shot}, {inside a spacecraft|standing on a lunar surface}, {cinematic rim light|soft window light}
```

Use prompt choices when you want variety across generations. Use a normal fixed phrase when a detail must stay consistent. Start with a small number of alternatives so you can identify which combinations work well before expanding the prompt.

## Iterate Without Losing a Good Result

Use a controlled loop:

1. Generate with a simple positive prompt and a short negative prompt.
2. Keep the same seed while changing one phrase.
3. Compare the result and keep the change only if it helps.
4. Change the seed after the prompt is working to explore new compositions.

The seed initializes the generation noise. Reusing it helps you compare prompt changes against a similar starting composition; changing it produces a new variation.

Do not change the prompt, seed, sampler, steps, and guidance at the same time. You will not know which change improved or damaged the image.

## Understand the Main Generation Controls

Prompt quality matters, but sampling controls also affect the result:

| Control               | What it changes                        | Practical use                                                                                                |
| --------------------- | -------------------------------------- | ------------------------------------------------------------------------------------------------------------ |
| Seed                  | Initial random noise                   | Lock it while testing a prompt; change it for variations.                                                    |
| Steps                 | Number of denoising iterations         | More steps take longer and do not always improve a finished image.                                           |
| CFG or guidance       | Strength of prompt adherence           | Increase carefully; excessive guidance can reduce image quality.                                             |
| Sampler and scheduler | How noise is removed                   | Start with the workflow default or use the [sampler and scheduler guide](/tutorial/scheduler-sampler-combo). |
| Denoise               | How strongly an input image is changed | `1.0` is full denoising; lower values preserve more of an input image in image-to-image workflows.           |

Model defaults matter. A CFG value that works for one checkpoint may be wrong for another, so do not apply one universal setting to every workflow.

## Starter Prompts

### Portrait

```txt theme={null}
Portrait of a woman with a short black bob wearing a dark green jacket, looking at the camera, rainy city street at night, 85mm lens, shallow depth of field, soft neon rim light, cinematic photography
```

### Product Image

```txt theme={null}
White ceramic perfume bottle on black stone, centered close-up product photo, clean background, softbox reflections, cool blue edge light, luxury advertising style
```

### Fantasy Illustration

```txt theme={null}
A small airship approaching a city built into red cliffs, wide establishing shot, clouds below the city, late afternoon sunlight, detailed fantasy concept art
```

### Anime Character

```txt theme={null}
1girl, solo, short silver hair, blue eyes, red winter coat, standing at a quiet train platform, falling snow, looking at viewer, soft evening light, detailed anime illustration
```

For anime checkpoints trained on tags, use the exact quality, source, and safety tags recommended in the relevant model guide.

## Fix Common Problems

### The model ignores the main subject

* Move the subject to the beginning.
* Remove secondary characters and unnecessary objects.
* Replace vague adjectives with visible traits.
* Reduce competing styles and weighted terms.

### The composition is wrong

* Add one framing instruction such as `close-up`, `full body`, or `wide shot`.
* State where the subject is placed: `centered`, `on the left`, or `in the foreground`.
* Keep the seed fixed while testing composition changes.

### The image looks overprocessed

* Remove repeated quality terms.
* Lower strong prompt weights.
* Reduce CFG or guidance if the selected model supports that control.
* Use a clear lighting setup instead of stacking `masterpiece`, `8k`, and `ultra detailed`.

### Important details keep changing

* Put identity details near the subject.
* Use consistent wording for colors, clothing, and accessories.
* Simplify the scene before adding more prompt terms.
* Use a suitable LoRA or reference-image workflow when text alone is not enough.

## Final Checklist

* Name the main subject first.
* Describe action, environment, composition, lighting, and style with visible terms.
* Keep the negative prompt focused on problems you actually see.
* Use prompt weights sparingly and only when the workflow supports them.
* Use `{choice one|choice two}` groups when you want automatic prompt variations.
* Keep the seed fixed and change one variable at a time.
* Follow the selected model's prompting guide for model-specific syntax.

## References

* [ComfyUI text-to-image workflow](https://docs.comfy.org/tutorials/basic/text-to-image)
* [ComfyUI KSampler node documentation](https://docs.comfy.org/built-in-nodes/sampling/ksampler)