# Image Generation
### **Basic Settings**
**Generation Model** – The engine used to generate images. Choosing a different model can drastically change the style of the image.
Here you can select either easy models or advanced models.
**Image Prompt** – The description of your image. For easy models, simply type a description of the image you want in the prompt. For advanced models, use tags or keywords for illustrious/pony models and natural language for Flux.
**Upload Base Image** – Allows you to create a similar image. It is recommended to use the same aspect ratio as the base image.
**Prompt Enhancer** – Automatically modifies your prompt for better results, automatically adds correct quality tags for corresponding models. Turn this off if you're an experienced user and want more control over your image.
**Tokens** – The currency used to generate images. 1 token is spent per image. Free users receive 20 tokens daily, while FL+ users receive 200 tokens daily.
### **Advanced Settings**
**Generation Steps** – The number of iterations the engine performs to generate the image. More steps can improve details but may also introduce significantly more artifacts. The recommended range is 20–40 steps, depending on the sampler.
**CFG Scale** – Controls how closely the image follows the prompt. Higher values may lead to artifacts. The recommended range is 3–10.
**Negative Prompt** – A prompt that specifies what to exclude from image generation. It is commonly used to remove undesirable elements like "low quality" or to reinforce a character's gender (e.g., using "1girl" in the negative prompt).
**Samplers** – These affect how the image is generated from random noise. Recommended samplers are "Euler a" and "DPM++2M SDE Karras"
**Clip Skip** – Can be used to skip layers in the encoder. Higher values will produce more abstract results.
**Generation Seed** – A random number used for image generation. It can be used to create similar images but does not guarantee the same character in different positions.
### Advanced Generation
For the advanced generation example we're gonna use illustrious and Pony-based models because Flux models use more simple and less structured natural language. SDXL also uses the same tag system as IL and Pony. \
\
You can see what the model is based on in the left upper corner when browsing all models.
#### **1. Basics**
IL and Pony use a tag-based system, meaning their prompts are divided into keywords. These keywords are based on tags from the Danbooru image board. Generally you can look there for tag groups, such as: styles, attires, poses, hair, camera angles, characters.\
\
General rules are:\
-Divide tags with commas: `best quality, 1girl, black hair` \
-Don't use capitalisation if it isn't used in the tag originally: `Masterpiece` ≠ `masterpiece` \
-Avoid repeating the tags\
-Generating multiple characters is extremely hard, generate them separately and use other image editing services to combine them.\
-Some of the tags have this format `(detailed fire forest background:1.2)` . It means they are weighted. Weights can be used to increase the effect of certain prompts. Be careful and only use the provided format, the number can be changed to reduce or increase the weight.
#### 2.Tag structure
Tag order and structure is very important for a good generation. I will use the prompt from the image above as an example and will divide it into small parts. This example is extremely large for the purpose of showing the tag examples; however, even a small prompt can result in great image gens.
* Quality
masterpiece, best quality, amazing quality, very aesthetic, 4k, high resolution, ultra-detailed, absurdres, newest, scenery
These are used to improve a general quality of the image and set up a style for it. Usually, 3-7 tags is more than enough. However, if you're generating something more niche - consider reducing the number of quality tags, it's gonna make generations more diverse. Pony-based models can also use quality tags as `score_10, score_9` , don't use them for IL-based models
* Body
1girl, furry, anthro, phoenix fox girl, orange fur, long hair, multicolored glowing hair,gradient hair, flame hair, glowing orange hair, floating hair, hair between eyes, gradient orange yellow eyes,very large eyes, uneven eyes
They are used to describe the overall character appearance: gender, hair, eyes, height, build, species.
* Expression
facing to the side, looking at viewer, upturned eyes,intimidating, smirk, smug, grin, crazy smile, long snout, sharp teeth
Tags used for detailed facial expressions. Aren't necessary in most of the cases, but can be used for further detailing if necessary
* Clothing
black headband with fire pattern, black collar with white spikes, cropped leather flaming red jacket over collared impossible white shirt, underbust belt
The most customisable type of tags, can be combined and adjusted to change certain details.
* Poses
```
portrait, upper body, close-up, foreshortening, dutch angle
```
Tags for various positions, can be combined with furniture and other characters.
* Background and effects.
(detailed fire forest background, fiery background:1.2), (blue fire:1.2) AND red fire,
dark ambient, glow, detailed eyes, depth of field, detailed fluffy fur, volumetric lighting
Used for background descriptions and additional effects.
#### 3. Negative tags
{% code overflow="wrap" %}
```
particles, 3d, rendered, human, multiple tails, old, oldest, cartoon, graphic, text, painting, crayon, graphite, abstract, glitch, deformed, mutated, ugly, disfigured, long body, lowres, bad anatomy, bad hands, missing fingers, extra digits, fewer digits, very displeasing, (worst quality, bad quality:1.2), bad anatomy, sketch, jpeg artifacts, signature, watermark, username, simple background, conjoined, bad ai-generated
```
{% endcode %}
These are the negative tags. For our current example, they were used to remove all remaining human features and to reinforce the modern art style. The common practice is to add negative quality tags here to improve the overall result even further.
---
# Agent Instructions: Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.
Perform an HTTP GET request on the current page URL with the `ask` query parameter:
```
GET https://fictionlab.gitbook.io/fictionlab/detailed-guides/image-generation.md?ask=
```
The question should be specific, self-contained, and written in natural language.
The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.
Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.