Self-Introduction
Nice to meet you and hello. I am TohoFrog (@tohofrog), and the NovelAI team asked me to share my method of creating images using NovelAI Diffusion.
First of all, let me briefly introduce myself. I am a very ordinary frog living in a port town in Japan. My relationship with drawing and computers goes back a long way, and I have been drawing digital illustrations since I was four years old. My mother used to say that I was a genius in my childhood.
I am now a dull and mediocre frog, but I like to create things, so I have been creating digital illustrations for many years in an equally mediocre way. When I was a student, I drew derivative works based on my favorite anime and games, and submitted original illustrations to exhibitions.
Recently, however, I have been spending more time at work and with my family, which has severely reduced the time that I have to create illustrations. Illustrating can be compared to sports; if I don’t practice, it doesn’t take long for my skills to deteriorate and weaken. It was getting increasingly difficult for me to draw in the way I wanted to, and I withdrew from the art of illustration.
For me, NovelAI was the perfect tool to help continue my creative endeavors. Using NovelAI, I was able to replace the manual aspect of “drawing” with “composing”, and was excited to have a new way to explore my own creativity.
Currently I am using NovelAI as my tool to communicate and share with others, and give form to my imagination.
Goal of this Article:
Image generation may seem like an all powerful magic, but in order to make the AI generate the images as you imagine you will need to know a few key points.
Next, I’ll walk you through the three primary aspects of NovelAI I plan to cover, aimed at guiding both beginners and intermediate users.. I hope that the reader will be able to achieve these three objectives:
•Gain an understanding of the basic usage of NovelAI.
•Generate high quality images
•Convey the vision within your mind.
After reading this article, you should be able to create an image like the one below.
NovelAI’s Basic Usage
Setting to Japanese Language:
NovelAI supports Japanese notation. If you would prefer to use Japanese, please change the language settings from the goose icon menu.
Filling in the Prompt:
Using text in the “Prompt” box, enter some specifics of the image you wish to generate.
Try typing the following prompt as a test:
close-up face, 1girl, glasses, make up, see-through jacket on tanktop, silver hair, short hair, dutch angle, white background
Once you have filled in the box, press the “Generate 1 Image” button.
Generating Images:
Around 20 seconds after pressing the “Generate 1 Image” button, an image like the one shown below will be displayed.
If you don’t like it, press the button again to generate a different new image. If you create an image that you like and wish to keep, click on the floppy disk icon to download it. All downloads keep the image metadata, allowing you to recreate the same image via the previous prompt and settings used.
⚠ If you refresh or close the web page or window, the image will disappear.
If you would like to change the character design, or would like to produce the same character in a series, please check out this article where you can learn how to easily specify the visuals of a character.
Creating a High-Quality Image:
Changing the Sampler Settings (SMEA DYN)
Click on the ▲symbol to open detailed information.
Check the SMEA and DYN boxes in the Sampler section.
Now let’s try to generate the image with the following prompt (the same prompt used in the previous section):
close-up face, 1girl, glasses, make up, see-through jacket on tanktop, silver hair, short hair, white background
Adding tags:
The text below are examples of the quality tags NovelAI trained into NovelAI Diffusion that have the effect of improving the quality of the image. They are called “quality tags” and “aesthetic tags”, you can read about them here.
best quality, amazing quality, very aesthetic
Let’s add these tags to the prompt!
Next, try and generate the following prompt (the {} symbols on either side of the tag will be explained later).
close-up face, 1girl, glasses, make up, see-through jacket on tanktop, silver hair, short hair, white background,{{{{{best quality, amazing quality, very aesthetic}}}}}
⚠If you want to generate images other than anime characters, you must modify the dedicated tags!
Adapting to match your vision:
From here onwards, things may get a little difficult.
Before creating an image, you must first prepare a concept. You will need to think about what you want to express in your image, what you would want people to focus on, and what emotion you want to give the image. Preparing a concept will determine the course you will need to take to adjust future work.
In this example, we will set our concept theme to focus on expressing the beauty of transparency.
Removing Unnecessary Elements:
To help define and clarify the concept, you can write undesired content into the input box for the elements you wish to exclude. The entry field will appear by clicking the “Undesired Content” tab. You can read up on how the Undesired Content box works here.
Even though it is sexy to see cleavage, it is not necessary for this concept, so let’s remove it by adding cleavage to the undesired content box.
Describing in More Detail:
If we give specific details to the AI, we can get the generated image to match our concept.
Transparency is an expression of light. The following is a list of related words: prism, reflection, refraction, transparent
Convert these words into more specific text.
prism effects, reflection other on glasses, beautiful light transparent, rays refraction
Then add these newly picked tags to the prompt. Let’s try and generate using the following prompt:
close-up face, 1girl, round glasses, opaled eyes, make up, see-through jacket on tanktop, silver hair, short hair, prism effects, reflection other on glasses, beautiful light transparent, rays refraction, white background, {{{{{best quality, amazing quality, very aesthetic}}}}}
Adding specific descriptions will help describe the transparency in more detail.
Adjusting the Text’s Emphasis:
If the prompt text is not reflecting as you would like in the generated image, please use the following method to adjust it.
Strengthening & Weakening Vectors:
Enclose the prompt text in parentheses to adjust its effect on the image. {} is for emphasis and [] is for suppression, read up on the Strengthening & Weakening of vectors here. These effects can be overlapped as well.
Sorting:
Rearranging the order of the text can significantly change the results. The first text tends to be
more heavily weighted during generation so your main subjects should be placed there.
Examples of Images and Prompts
Example #1:
Prompt: {{{{close-up face}}}}, [[dutch angle]], 1girl, looking at viewer, round glasses, opaled eyes, make up, see-through jacket on tanktop, silver hair, short hair, prism effects, reflection other on glasses, beautiful light transparent, rays refraction, white background, {{{{{best quality, amazing quality, very aesthetic}}}}}
Undesired Content: cleavage
Emphasizing the close-up face allows for a better view of the face. The newly added “Dutch Angle” is also designed to create an angle as if the camera were tilted.
Example #2:
Prompt: best quality, {{{{close-up face}}}}, [[dutch angle]], 1girl, looking at viewer, round glasses, opaled eyes, make up, see-through jacket on tanktop, silver hair, short hair, prism effects, reflection other on glasses, beautiful light transparent, rays refraction, white background, [[water drop]], backlighting, {{{{{best quality, amazing quality, very aesthetic}}}}}
Undesired Content: cleavage, leather jacket, sketch, necklace, long neck, monochrome, top of head, round image, through the hole, chubby
I added water drops and backlighting to create transparency. To stabilize the result, I added multiple tags to the elements we want to exclude.
Example 3:
Prompt: very aesthetic, ultra detailed, {{close-up face, 1girl}}, year 2022, round glasses, opaled eyes, long eyelashes, [[make up]], {{see-through jacket}} on black tanktop, silver hair, short hair, {{small diamond earrings}}, {{{prism effects, reflection other on glasses, distorted transparent, rays refraction}}}, white background, [[water drop]], iridescent, backlight, color shadow, [[dutch angle]], {{{{year 2021}}}}, {{{{{best quality, amazing quality, very aesthetic}}}}}, {{{{{ultra detailed, photo}}}}}
Undesired Content: cleavage, leather jacket, sketch, necklace, long neck, monochrome, chubby, flat, oil painting (medium), [[[[[ai-generated]]]]], top of head, round image, through the hole
Further modifications to the prompt can produce images depicting complex light and shadow. small diamond earrings were added to accentuate the image. The text of the prompt has been extended, so we have increased the emphasis on quality/aesthetic tags, which had become relatively less influential as the overall prompt length increased.
Conclusion
Thank you for taking the time to read my article. I will go over the points one more time.
•NovelAI’s Fundamental Usage
-Setting in Japanese Language
-Entering Prompts
-Generating Images
•Generating High Quality Images
-Changing the Sampler Settings
-Adding Dedicated Tags
•Adjusting to Fit your Imagination
-Removing Undesired Elements
-Describing in More Detail
-Adjusting the Text’s Emphasis
This article was designed to support beginner to intermediate users, and I would appreciate any feedback or questions that you may have!
As I mentioned before, conceptualization plays an important role in image generation.
Part of the fun is in the process of planning the concept and designing the prompts to achieve your goal. Although image creation tends to be an easy process, it is also important to take the time to think through the process in depth.
In addition, by working through your own concept you will create a unique identity for your generated images, which will provide an opportunity to further deepen your interactions with others and elicit positive responses.
NovelAI will support your creative activities and provide you with a new world of expression. I hope this article will help enrich your artistic endeavors!
