+!!! danger STILL A W.I.P

Guide for Novel AI image generation


Intro

Image description Novel AI. What is it? Pulled word for word from the NovelAI site.
Novel AI status
Novel AI updates New NovelAI updates

Use powerful image models to depict characters and moments from your stories, with the leading Anime Art AI and other AI models. No censorship or guidelines - anything goes!

Anything? Yes. ANYTHING

NovelAI is another LLM and an image generation service. You might be asking why should I pay for an image generation tool when there are free options out there?

  • The 3 paid tiers offer credits for you to generate images, starting at 1000 anlas (tablet tier), 1000 anlas (scroll tier), and then 10,000 anlas (Opus tier, with Unlimited gens outside of the 10,000 anlas: Steps 28, and any of the normal/small image settings you can gen for 0 anlas)
  • They have a great furry model (I'm not one for genning that stuff but I've seen people generate amazing stuff)
  • They updated to V4 and hooollyyy fuck the extra character options, and being able to manually edit characters positions?!?!?? A GODSEND
  • If you buy any of their tiers, you also get access to their text generation LLM (which is pretty dope might I add)

Useful resources!

Danbooru artist tags experiments This showcases a SHIT TON of artists. Note that it showcases with only females and not males.
Feldherren's home page
Feldherren's male tag experimentation Please note that this is still a W.I.P
Feldherren's medium tag experiementation Please note that this is still a W.I.P
Novelai V4 This goes over the new changes in Novelai V4 (includes action tags, character positions, text integration)
Novelai V4 diffusion medium article This introduces the new changes, but also goes over the new quality and year tags, new special tags and phrases.
Creating consistent characters This is technically NovelAI V3, but it goes over a ton of hairstyles, body tags, angles, clothing, ect. Very handy for beginnings!
Danbooru No seriously, looking at the images and seeing how they tag it is super helpful!
Image description

The basics

So when you load into Novelai and click on images, you're greeted with this.

Image description

What is it? Well that blank box is where you put your prompts! This is where the 'main prompt' or solo character prompt goes. (We'll delve more into that). You'll also see options to add a base img, character prompts (add characters), image settings, and that bar at the bottom that says 'steps, guidance, and, sampler.'

Right! Now lets delve into how to access your steps, guidance, seed, and samplers. Then I'll do a brief description of what each one does.

image descriptionimage description

  • Steps define the number of iterations the AI should refine from the initial creation of your prompt
    • Too many Steps may also be counterproductive, barely improving the generation. Along with going above 28 steps gets rid of your generations costing 0 anlas (on opus tier), so if you set your steps at 29, your anlas cost will go up.
  • The Prompt Guidance value indicates how much the AI will respond to your prompt.
    • Sometimes the Prompt Guidance value will adjust how refined and strong the effect will be: a lower Guidance number can be more painterly and softer, while a higher Guidance value provides more sharp details.

image description

  • Samplers are used to tell the AI how it should start generating visual patterns from the initial noise.
    • Different Samplers generate different images, but the differences may not always be obvious or predictable.

DPM++ 2m and Euler Ancestral are their recommended samplers, due to their consistent, high-quality generations in combination with NovelAI Diffusion. I'd highly recommend leaving the Sampler setting as is, unless you have a deeper knowledge of Image Generation.

image description

  • Now lets get into our noise samplers. Noise Schedule affects the detail quality in general. I always stick to Karras, but I do honestly recommending trying it out for yourself as we might all have differing opinions.

  • The seed section is where a number that settles the generation's randomness. Please leave it empty to get different gen, or copy and paste a seed from your previous generation (make sure to use the same prompt and more or less the same generation parameters) to get the either the exact same gen or as close as it can get. There's an amount of changes you can make to the prompt for the seed to keep the overall composition same, but it's not consistent.

Generating images

There are prompt settings that automatically add quality tags, you can view this by pressing the gear in the prompt section. If you turn this off, it will remove the quality tags automatically as well. There is also a undesired content preset, you can hit none and it will remove that preset for you. I have mine set to heavy undesired content (UC for short) preset. image description You'll also note that there is a 'disable tag suggestions', tag suggestions pops up possible tags. So if you start typing hair, it will bring up a few hair prompts that you could choose from. I don't personally turn this off as it does help me locate a tag without popping up Danbooru.

Now let's get into generating an image! So you have your prompt section and your undesired content section. Your prompt section is where you put the prompts for the image you wish to generate. The undesired content is where you put what you don't want. I'll provide a bare basics undesired content prompt below.

Undesired bare basic preset I use:
blurry, lowres, error, film grain, scan artifacts, jpeg artifacts, very displeasing, chromatic aberration, logo, dated, signature, normal quality, bad quality, worst quality

How to prompt

Novelai uses the Danbooru tagging system to create prompts, so you can achieve your results with danbooru tags. Novelai v4 has added natural language integration into generating an image, however I still prefer to use the Danbooru tagging system. First off, let's go into the bare basics of a prompt:

  • Gender?
  • Skin tone?
  • Hair color/Hair length/Hair style?
  • Eye color/Pupils?
  • Clothes?
  • POV
  • Background?

These are all the essential questions on creating a more in-depth character. Let's just say you don't care what they look like, and just want a female or a male character and let want to let the model random the rest of their features, that's ok too! (There is also a random prompt button but we'll get into that later).


So let's craft a simple starter prompt just describing these things: Gender, skin tone, hair color, eye color, clothes. (Steps: 23. Prompt Guidance: 6)
1girl, pale skin, blue eyes, black hair, tracksuit,
That's it. That's the prompt. Now let me show you two generations it produced.

image descriptionimage description


Now to showcase a more in-depth prompt, I am going to be using my main persona I use to roleplay. Here is what this prompt will include: Gender, skin tone, hair color/length/style, eyecolor, clothes, background, POV focus.
1girl, dark skin, dark brown hair, long hair, messy hair, sidelocks, wolf cut, dark brown eyes, purple knit sweater, black leggings, chubby, plump, big breasts, thick thighs, waving, grinning, white outline, beautiful background, garden, blurry background,

image descriptionimage description

Each tag tells the model what you want and what to do. I also included the tags that go along with these examples, along with ways you could tag the background.

  • Gender = Are they a female (1girl), male (1boy), other (1other)?
  • Skin tone = Are they pale (Pale skin), dark skinned (Dark skin), are they alien so they have purple skin (purple skin)?
  • Hair color/length/style = What's the color of their hair (blue/black/brown hair), is it short (short hair), in-between (medium hair), or freakishly long (absurdly long hair)? Is is curly (curly hair), wavy (wavy hair), straight (straight hair)? Do they wear it in a pony tail (ponytail), a bun (bun), braids (braid)?
  • Eyecolor/pupils = Do they have dazzling blue eyes (blue eyes), dark green (dark green eyes)? Do they have heart pupils (heart pupils) for that perfect yandere look, or maybe star pupils (star pupils) to show excitement? Slit pupils (slit pupils) because they're a cat kemonomimi?
  • Clothes = Are they in a simple tracksuit (tracksuit)? A dress with leggings (dress, black leggings)? A maid outfit (maid outfit)?
  • POV = Is this from the shoulder up (portrait)? Everything but their feet are in frame (Feet out of frame)? Are they far away (wide shot)?
  • Background = Are they in a seedy bar (bar, dark, foggy)? Maybe out in the woods (forest, sunlight, winter)? Are they swimming in a pool (swimming pool, water, in water)?

A good thing to note is the more you add to each category the more refined your generations are. I'll provide an example by showcasing the process I went through on generating my Star Wars persona. I wanted her to be in a dress, and for the life of me, it was giving me a dress that hugged her body which isn't what I wanted. I wanted the skirt to flare out at either the waist or underneath her bust. This was my original prompt: 1girl, cowboy shot, dark skin, long hair, dark brown hair, messy hair, wavy, sidelocks, dark brown eyes, thick lashes, round face, curvy, plump, chubby, medium breasts, thick thighs, wide hips, black dress, long sleeves, bell sleeves, long dress, frilled dress, chronopattern dress,

image description
As you can tell it doesn't flare out where I wanted it to, AND if you note my prompt I didn't include anything that would indicate this. I also included 'chronopattern dress' which didn't even show up in the end image. So I went back to the drawing board, did some searching just using google to try and figure out the names of what that skirt type was called.

After I found out I went back to my prompt, removed the 'chronopattern dress' and added the different tags. So now, my new workshopped prompt looks like this: 1girl, cowboy shot, dark skin, long hair, dark brown hair, messy hair, wavy, sidelocks, dark brown eyes, thick lashes, round face, curvy, plump, chubby, medium breasts, thick thighs, wide hips, black dress, long sleeves, bell sleeves, long dress, empire skirt, flare skirt,

image description
Notice the difference? The skirt flares out, an empire skirt features a high waistline that sits right underneath the bust. Combined with the skirt flare, the dresses weren't clinging to her body and gave her the dress that I wanted it too.

But I still wanted more, I had the skirt shape I wanted, but now I wanted her face to be covered. If you want your characters facial features not to come through, such as the eyes are covered, then you NEED to remove that correlating tag. So here is the tweaked prompt: 1girl, cowboy shot, dark skin, long hair, dark brown hair, messy hair, wavy, sidelocks, black veil, over face, curvy, plump, chubby, medium breasts, thick thighs, wide hips, black dress, long sleeves, bell sleeves, lace trim on sleeves, long dress, empire skirt, flare skirt,

image descriptionimage description

As you can tell, she's wearing a veil, you can't see her facial features (which gives that eerie vibe to the whole gen), and while it didn't give me the exact lace trim I wanted, it still included the lace into the arms. It kept the empire waistline, and kept the skirt flare. Now for the sake of showing that, the right side gen does have that eye tag. The veil now doesn't cover her face, her eyes, eyebrows, and her mouth are visible. Here is the prompt with the eye tags still included: 1girl, cowboy shot, dark skin, long hair, dark brown hair, messy hair, wavy, sidelocks, dark brown eyes, thick lashes, round face ,black veil, over face, curvy, plump, chubby, medium breasts, thick thighs, wide hips, black dress, long sleeves, bell sleeves, lace trim on sleeves, long dress, empire skirt, flare skirt,

The more detail you add to your prompt, the more consistent images you'll get. Don't get me wrong, sometimes less is better, like bunny suit will almost ALWAYS generate a playboy bunny costume. So you don't have to include a ton of detail or even go super in depth.

Style tags/Medium tags

Now lets get into style tags ( I.E: Sketch, pixel, monochrome, pen, play cards, tarot cards, watercolor, ink), or this can be year tags (I.E: year 1990, year 2024, year 2010). Novelai added these new medium tags, along with some simple style tags.

For the ease of showcasing medium and style tags, I will be using this prompt: 1girl, long hair, black hair, red eyes, fangs, (steps: 28. Guidance:6)

MEDIUM TAGS

  • photo (medium), photographic doll, fumo (doll) : will generate a plushy in the fumo style.
  • photo (medium), figure : will generate characters in the form of a plastic figure.
  • {insert medium here} (medium) : is the correct way to tag your mediums for generations

Order of Medium tag example images:

Left side Right side
Photo (medium), photographic doll, fumo (doll) Photo (medium), figure
traditional media, pen (medium) card (medium)
traditional media, watercolor (medium) traditional media, ink (medium)

Fumo dollFigure
PenCard
WatercolorInk

STYLE TAGS

  • Highly finished :indicates that the image is not unfinished in any way, linework is clean, shading is fully complete.
  • Digital illustration : can overall increase the quality of your generation in some cases and push towards a less realistic style.
  • Anime style : can specifically push towards more anime-like styles, which can be useful in some cases.
  • Smooth shading : indicates a style of shading that is smooth and less “cell shading”-like.
  • Smooth, shiny shading : additionally emphasizes shiny highlights.
  • {Insert style tag}. Ex: pixel art, sketch, line art, monochrome

Order of the Style tags example images:

Left side Right side
Highly finished Digital illustration
Anime style Smooth shading
Smooth, shiny shading Year 1990
Sketch, lineart, monochrome Pixel art

highly finishedDigital illustration
Anime styleSmooth shading
Smooth, shiny shadingYear 1990
Sketch, lineart, monochromePixel art

Artist tags

Artist tags can emulate an artists style, I will be using four artist tags to show you the difference it can make. I will be using these artists: Ainiwaffles (This is the artist that actually personifies NovelAI's text generation models), Incase, kimidake, and kafun as examples. You can simply tag artist names as, artist:Ainiwaffles or ainiwaffles. I also will be combining ainiwaffles, kimidake, and incase, kafun, to show you that you can blend artist tags to create a unique mix. For the ease of showcasing the artist tags, I will be using this prompt: 1girl, short hair, curly hair, blue hair, red eyes, heart pupils, bunny outfit, heart hands (steps: 28. Guidance:6)

Order of the artist tags example images:

Left side Right side
Ainiwaffles Incase
kimidake kafun
ainiwaffles, kimidake incase, kafun

ainiwafflesincase
kimidakekafun
ainiwaffles,kimidakeincase,cafun

Multi-character prompting, action tags, and character positioning

Emphasis, text in generation, natural language prompting, and metadata

I decided to put these together instead of dedicating a section to each as they are all smaller things that can be easily explained. I also will be showing a gen with an example for each.

Emphasis and de-emphasis

You can emphasize and de-emphasize your prompt in two ways. You can use {insert tag here} to emphasize OR you can do 1.1::insert tag here:: (anything above 1.0 will emphasize, as 1.0 is the starting baseline of the weight of any tag.) To de-emphasize you can use [insert tag here] OR you can do 0.9::insert tag here:: (as anything below 1.0 will de-emphasize as stated above it is the baselines of the tag weight.)

You can view this by pressing the gear in the prompt section, where you will see 'highlight emphasis'. Keeping it on allows you to see how emphasis affects a tag: It turns red when you increase its weight, and blue when you reduce it. The color darkens depending on how much you emphasis the 'do this more' or 'do this less'. You can turn this off if you don't wish to see the colors. image (Ignore the bad quality of the image please)

Now I will be showcasing how emphasizing and de-emphasizing affects a prompt. This will be the base prompt:1girl, cowboy shot, pale skin, medium hair, platinum hair, messy hair, wolf cut, sidelocks, purple eyes, black and red button down flannel, unbuttoned, white undershirt, black ripped skinny jeans, black leather gloves, curvy, small breasts, thick thighs, small belly, wide hips, looking at up, crazy smile, night sky, standing in rain, dripping, highly finished, To provide an example of what to reference it with, I will include a base gen without any emphasis. no emphasis

I will be adding emphasize/de-emphasize to the messy hair tag along to showcase the difference between the tag weight at it's base weight and then showing

Left side Right side
Emphasis at: 1.8::messy hair:: De-emphasis at: 0.3::messy hair::

EmphasisDe-emphasis

Adding text to your images

Novelai added text integration into your images. While it won't spit out novel level of writing, its definitely much easier. The BEST way to prompt for it is to include 'text, english text, into your prompt. Once your prompt is finished, end it on a period instead of a comma, then follow it by Text: insert text. I've provided two different ways you can prompt it, using the normal tagging system, and using natural language prompting.

Left side Right side
Normal promping: 1girl, button up, dress pants, black pants, text, english text, speech bubble, white background, holding sign, wooden sign. Text: HELLO <3 Natural language: 1girl, text, english text, speech bubble, white background. A pastel pink speech bubble with yellow handwritten text says “HAVE FUN!”. Text: HAVE FUN!

Normal promptingNatural language

Natural language prompting

Novelai has introduced natural language prompting! With this comes an extended prompt context of 512 tokens, you can write your prompt fully in natural language or you can combine normal tags with natural language prompting. I will be showcasing both a fully natural language simple prompt and a more indepth prompt. I also will be showcasing using both normal tagging and natural language prompting with a simple prompt and a more indepth prompt.

Simple NL prompt: One girl looking at viewer wearing a cute sweater and jean shorts. She is smiling happily.
Indepth NL prompt: One girl looking at viewer wearing a blue sweater with shoulder cutouts. She is also wearing a black and white pleated skirt with knee high dark brown leather boots. She has her hand up and is doing a peace sign.
------ | ------
Simple NL prompt | Indepth NL prompt
simple NL promptindepth NL prompt

Meta data

How to generate furries

Kemonomimi

NSFW

Sprucing up your generations

Now once you generate an image you might notice icon at the top of the screen above it.
image description image description image description image description image description image description

Edit

Pub: 22 Mar 2025 01:24 UTC

Edit: 10 Sep 2025 00:28 UTC

Views: 414