NovelAI Image Generation Prompt Tips: A Comprehensive Guide

NovelAI has revolutionized the field of artificial intelligence by harnessing the power of language models to generate human-like text. With its impressive capabilities in generating coherent and contextually relevant content, NovelAI has become an invaluable tool for various applications such as writing, storytelling, and even customer service. However, NovelAI’s capabilities extend beyond text generation, as it also possesses the ability to generate images.

In this blog post, we will explore NovelAI’s image generation capabilities and delve into the exciting advancements it has made in this domain. While NovelAI’s image generation results have been impressive, there is always room for improvement. We will acknowledge the generic nature of the images generated by AI models and discuss the need for enhanced image quality.

Throughout this post, we will provide an overview of the content, covering various aspects of NovelAI’s image generation. We will explore the underlying technology behind image generation, discuss the challenges faced in producing high-quality images, and examine the potential applications of NovelAI’s image generation capabilities. Additionally, we will highlight the ongoing research and developments aimed at further enhancing image generation in NovelAI.

Join us on this exploration of NovelAI’s image generation as we uncover the current capabilities, address the limitations, and envision the future possibilities of AI-generated images.

Quick Tips for Better Image Generation

Generating high-quality and visually appealing images with NovelAI requires a combination of creative input and an understanding of the platform’s functionalities. In this section, we will provide you with some quick tips to enhance your image generation experience.

1. Be Descriptive in Prompts: When providing a prompt to generate an image, it is essential to be as descriptive as possible. Instead of simply stating “a cat,” try providing specific details such as “a fluffy white cat playing with a red ball of yarn on a sunny window sill.” The more information you provide, the better NovelAI can understand your desired image and generate a result that aligns with your expectations.
2. Utilize Curly Brackets and Parentheses: NovelAI allows you to use curly brackets ({}) and parentheses (()) in your prompts to give emphasis and provide additional instructions. For example, you can specify the attributes of an object within curly brackets like “{a tall building}” or indicate the desired action within parentheses such as “(smiling).” These formatting techniques help guide the model’s attention and provide more context for generating the image.
3. Save Your Images: It’s important to note that images generated through NovelAI are not automatically saved on the NovelAI website. To ensure you don’t lose the images you generate, make sure to save them locally on your device. After generating an image, you can simply right-click on it and choose the “Save image as” option to save it to a desired location on your computer or device.

By following these quick tips, you can improve the quality and relevance of the images generated by NovelAI. Remember, the more precise and descriptive you are in your prompts, the better the chances of getting results that meet your expectations.

Utilizing Danbooru for Training Data

To create the remarkable image generation capabilities of NovelAI, a vast amount of training data is required. One valuable source of such data is Danbooru, the largest anime imageboard on the internet. In this section, we will explore the significance of Danbooru as a training data resource and discuss its robust tagging system.

1. Introduction to Danbooru: Danbooru is an imageboard primarily focused on anime and manga-related content. It hosts an extensive collection of high-quality images, spanning various genres, characters, and art styles. With a dedicated community of contributors, Danbooru has become an invaluable resource for artists, enthusiasts, and researchers alike.
2. The Importance of Danbooru’s Tagging System: One of the standout features of Danbooru is its robust tagging system. Each image on Danbooru is accompanied by a comprehensive set of tags, which describe various aspects of the content, including characters, settings, actions, emotions, and much more. These tags are manually assigned by the community, ensuring a consistent and detailed categorization of the images.

The tagging system plays a vital role in training AI models like NovelAI. By leveraging the wealth of labeled data available on Danbooru, the models can learn to associate specific tags with visual features and generate images that align with the given prompts. This enables users of NovelAI to create custom images based on specific themes, characters, or scenarios, all derived from the collective knowledge and creativity of the Danbooru community.

3. Resources for Studying Danbooru Tags: If you wish to explore the diverse tags used on Danbooru and gain a better understanding of how they can be utilized in your image generation prompts, there are resources available to aid you in your studies. Websites like “Safebooru” and “Danbooru Preview” provide platforms where you can search for tags, explore related images, and delve into the depths of Danbooru’s tagging system.

By harnessing the power of Danbooru’s extensive image database and its meticulous tagging system, NovelAI can generate images that align more closely with the prompts and capture the desired aesthetic or narrative. Understanding the nuances of Danbooru’s tags can greatly enhance your image generation experience and allow you to create more tailored and engaging visual content.

Understanding NovelAI Settings

Steps:

Think of the “Steps” setting as the amount of time the AI spends working on the image. Generally, increasing the number of steps results in higher-quality images. However, there are diminishing returns as you go higher in step count.

To avoid additional charges (for Opus plan users), it’s recommended to start with 28 steps since NovelAI charges Anlas for 29 steps or more. Keep in mind that the maximum step limit is 50, except for Stable Diffusion, which has a higher limit.

For experimentation, start with 28 steps and adjust as needed. If you find a prompt or seed that you like, raise the steps to 50 and try again for potentially improved results.

Scale (CFG):

Scale, also known as CFG scale, determines how strictly the AI adheres to your prompt. Higher values result in the AI being more focused on your prompt, while lower values give it more freedom to deviate and provide unexpected results. A recommended range to start with is 5-15.

Personally, using a lower scale of around 4-7 is preferred by some users as it allows the AI to offer results that are beyond their expectations.

Sampling:

Sampling refers to how the AI thinks about your image. The k_euler method is popular because it produces consistent and predictable results. On the other hand, the ddim method tends to be more expressive and artistic, introducing greater variation.

Seed:

Generators in NovelAI start with a noise image and progressively transform it into an image based on the model’s understanding. Each seed value corresponds to a unique starting noise image. This means that if you use the same settings, prompt, and seed as another user, you can generate the same image with a high degree of accuracy (90%-100%).

Keeping the seed the same while adjusting other settings allows you to see how each setting impacts the final image. Knowing the seed is crucial if you want to replicate another user’s image exactly.

To summarize, think of the “Steps” setting as the AI’s working time on the image. Scale (CFG) determines how strictly the AI follows your prompt. Sampling methods, such as k_euler and ddim, offer different approaches to image generation. The seed value corresponds to a unique starting point for image creation. Experimentation with different seeds and settings allows you to explore and fine-tune your desired results efficiently.

Improving Image Quality: Starting with a Generic Template

Starting With a Generic Template:

To begin, here’s a template that you can customize according to your preferences:

Prompt:

{{{masterpiece}}}, {{{best quality}}}, {{ultra-detailed}}, {cinematic lighting}, {illustration}, {beautiful detailed eyes}, {1girl}, upper body, looking at viewer, depth of field

Undesired Content (Negative Prompt):

lowres, bad anatomy, bad hands, text, error, missing fingers, extra digit, fewer digits, cropped, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, blurry, artist name

Settings:

• Training set: NAI Diffusion Anime (Full)
• Steps: 40 (start lower at 28 so you don’t burn your Anlas)
• Scale: 7
• Sampling: k_euler

Modifying it:

I recommend you put together your own generic template that produces a style that you like, that you can start or end all your prompts with.

(for inspiration, check out Danbooru tags. The model is trained on Danbooru images, so a lot of these tags will work well)

Using Quality & Detail Tags

There’s a ridiculous number of quality tags you can try and also mix together for interesting results:

• detailed
• hyper
• intricate
• wonderful
• accuracy
• amazing
• wonder
• finely
• super
• exquisitely

Here’s an example where I’ll keep the other settings exactly the same but change the quality descriptor.

• Undesired content (negative prompts): None
• Training set: NAI Diffusion Anime (Full)
• Add Quality Tags: Off (we’ll be using our own)
• Steps: 28
• Scale: 12
• Sampling: k_euler
• Seed: 2000000

Remember to enjoy the experience and embrace the variation in generated images.

Using More Emphasis Tags

If your images don’t look as good as you want them to, try adding more curly braces {} around your masterpiece tag.

Curly braces add emphasis and we’re just telling the AI we want a higher-quality image.

You can also add them to quality descriptors such as best quality and ultra-detailed, as well as things you feel are underemphasized in your generations.

{{{{{{{{{{{masterpiece}}}}}}}}}}},  {{{{{{swimsuit}}}}}}, {{{{{ultra-detailed}}}}}, {cinematic lighting}, {illustration}, {beautiful detailed eyes}, {1girl}, upper body, looking at viewer

• Training set: NAI Diffusion Anime (Full)
• Steps: 50
• Scale: 7
• Sampling: k_euler

Note: the masterpiece and best quality and other quality tags will steer the AI away from things that are considered undesirable as well as things that are less common. If you are trying to generate a niche subject, you should actually avoid these tags!

Undesired Content (Negative Prompts)

Getting too many mutated body parts?

Here are some negative prompts you can use to avoid different things (undesired content field). Click on headers to expand sections:Whole body (avoid mutated)Hands (avoid mutated)Hand held weapons (avoid fusing)Face (avoid poor quality)Leg (avoid mutated)Clothes (avoid fusing)

Art Styles

There are as many art styles you can try as there are stars in the sky. Here’s the same character with the same settings, in different art styles:

Settings:

• Undesired content (negative prompts): None
• Training set: NAI Diffusion Anime (Full)
• Add Quality Tags: Off (we’ll be using our own)
• Steps: 28
• Scale: 12
• Sampling: k_euler
• Seed: 121000000

Base prompt:

Results with base prompt + art style modifier:

{ gouache }

{ oil painting }

{ etching }

{ flat color }

{ line art }

{ 1990s }

{ hyperrealistic }

Understanding the Randomness in Image Generation:

It’s crucial to acknowledge that image generation involves a significant random component (RNG). When browsing through images shared by others online, keep in mind that they typically showcase the best results they have obtained.

In this guide, I have also selected and presented the better images for reference. However, it’s important to note that alongside the desirable outcomes, you may encounter numerous subpar results as well. This is an inherent part of the image generation experience.

Even if you use the exact prompts you find online, there’s no guarantee of replicating the same exceptional results. The RNG introduces variability and unpredictability into the generated images.

To manage your Anlas effectively, it’s advisable to set your steps to a lower value initially. This approach allows you to explore different possibilities without depleting your Anlas allocation too quickly.

Embrace the diversity of generated images and enjoy the creative journey, knowing that each outcome carries its own unique charm.

Character Prompting

Existing Characters

Results will vary wildly depending on the character.

For some, all you have to do is prompt character name for great results.

Others, you will have to prompt the character’s name as well as describe their characteristics.

There is only one correct way to write the character’s name. you want to use the name the character is tagged with on Danbooru.

eg. using yorha no. 2 type b will produce good results, however using 2b will not.

(You can see what version of the name NovelAI uses by letting it auto-complete the tag.)

Here’s a list of the most popular characters on Danbooru. You can assume if the character has a substantial number of Danbooru images, NovelAI will generate them well. Remember to remove the underscores from the character’s name when you are using them in tags.

• Undesired content (negative prompts): Low Quality + Bad Anatomy (Preset)
• Training set: NAI Diffusion Anime (Full)
• Add Quality Tags: On
• Steps: 28
• Scale: 15
• Sampling: k_euler

saber, masterpiece

yorha no. 2 type b, blindfold, black dress, black stockings, frills, {masterpiece}

akiyama mio, masterpiece

asuna (sao), masterpiece

hatsune miku, masterpiece

izayoi sakuya, masterpiece

ayanami rei, masterpiece

souryuu asuka langley, masterpiece

Multiple Characters

You can make multiple characters appear with this prompt:

2girls_A_and_B, Agirl have blonde twintails and blue eyes, {master piece}, {extremely detailed}, Bgirl have short black hair and red eyes

• Undesired content (negative prompts): Low Quality + Bad Anatomy (Preset)
• Training set: NAI Diffusion Anime (Full)
• Add Quality Tags: On
• Steps: 28
• Scale: 15
• Sampling: k_euler

As usual, the more details you add the less stable the generations become (more propensity for weirdness).

You can see the eye colors are swapped from my prompt. This one will take a few rolls to get right.

Faces

You can try the following expressions:

happy

evil smile

serious

unamused

surprised

Poses

You can try the following poses:

• looking back
• head rest
• on all fours
• wariza
• hugging own legs

Environment Prompting

Backgrounds

• classroom
• stained glass
• greenhouse
• bedroom
• cafe
• bookstore
• bakery

Outdoors:

• Forest
• lunar surface
• stratosphere
• underwater
• snowy
• atrium
• sunlight filtering through the trees

Composition / Perspective

• close-up
• portrait
• upper body
• cowboy shot
• full body
• lower body

Camera Effects

Depth of field: a good all-round addition to create the clear subject/blurry background photography effect.

{{depth of field}}, hatsune miku, outside, sunny day, {masterpiece}

You can also specify different camera lens, starting from 20mm and going to 200mm

You can see that more zoom not only gives you camera effects, but increases the realism of the image as well:

20mm, portrait, girl, outside, sunny day, {masterpiece}

35mm

100mm

200mm

NAI Diffusion paved the way for many other models such as Anything V3. More models and techniques continue to come out every day.

Conclusion:

In conclusion, we have covered several important tips and techniques for generating high-quality anime images using NovelAI. By following these guidelines, you can enhance your image generation experience and obtain impressive results. Let’s recap the key points:

1. Be descriptive in your prompts: Providing detailed descriptions helps guide the AI and improve the quality of generated images.
2. Utilize curly brackets and parentheses for emphasis: By using these symbols strategically, you can emphasize certain aspects of your prompts and influence the AI’s focus.
3. Remember to save your images: As images are not automatically saved on the NovelAI website, make sure to save your desired creations for future reference.
4. Consider utilizing Danbooru for training data: Danbooru, the largest anime imageboard, offers a robust tagging system that can enhance the AI’s understanding of your prompts.
5. Understand NovelAI settings: Familiarize yourself with settings such as “Steps” to control image quality, choosing the appropriate scale (CFG) value, and selecting the sampling method that aligns with your preferences.
6. Experiment and explore: Don’t be afraid to try different prompts, settings, and combinations to discover unique and captivating results.

As you embark on your image generation journey, we encourage you to experiment and push the boundaries of creativity. NovelAI provides a platform for endless possibilities, allowing you to bring your anime-inspired visions to life.

We invite you to share your generated images and experiences with the NovelAI community. By sharing your creations, you not only inspire others but also contribute to the collective exploration of this remarkable AI technology.

Unleash your imagination, generate stunning anime images, and let your creativity shine with NovelAI!