Foreword

• This guide is a work in progress, which means it’s not a completed, full-fledged product. I have been using NAI for a short time and this document is where I’ll share my otherwise little knowledge with you in order to give you a few tips and tags to work with size scenarios. In case you find out more tags and tips which could be useful for this same purpose and you want to share them, please write your suggestions in the thread and I’ll add all those here.
• NovelAI’s image generator is only accessible through a paid subscription. Currently there are 3 available subscription tiers: Tablet, Scroll and Opus. We recommend getting Opus, the most expensive one (25$ / 23€ per month), as it grants the most convenient benefits for image generation. Opus-tier allows the user both to use all the available and future Diffusion models provided by NAI and to generate an unlimited amount of pics choosing from 3 standard resolutions (1024x1024, 832x1216, 1216x832). The user also receives 10K Anlas points each month, which can be deployed to add further improvements to the pics. In my experience, the most useful of these improvements is Enhance, which generally "improves" the overall quality of the generated pic based on the prompt: personally I use Enhance to create pics at a slightly larger resolution and fix broken/bad limbs in one try.

• When writing a verbal prompt to create a pic on NovelAI, it’s useless to try to be descriptive and write grammatically correct sentences: what you want to do, instead, is to copy-paste tags and separate them with commas. As a general rule, the more posts a tag has, the more chances you have to make NovelAI recognize the thing you have in mind. While placing tags in the prompt, keep in mind that the sooner they are, the more important they’re considered by the AI.

Recommended settings for OPUS-tier subscribers

• Steps: 28 (max amount of steps for OPUS users to generate pics for free)
• Guidance: 5
• Sampler: Euler
• Add quality tags: Toggle
• Undesired content preset: Heavy

Furry V3

This section contains some tips which should be useful for you to create size scenarios using NAI's Diffusion Furry V3 model. This model is based on e621's tags. Compared to the Anime V3 model, this model offers some notable advantages for the creation of size scenarios:

• more and better size interactions, including internal and entrapment shots ;
• better management of multiple characters of different sizes, which makes inpainting necessary only to fix small things;
• more accurate, more faithful and overall better representation of sexual organs;
• more options for views and size control tags.

However, this newly released model also has some disadvantages:

• since e621 is strongly skewed towards furry subjects, it's more difficult to make completely human characters; (check this section of the present guide to learn about some simple workarounds for this)
• as it's been reported by NAI's devs, artist tags do not work.

Despite the con's, I do recommend to start experimenting with this model over the Anime V3 one, especially if you don't mind putting fantasy creatures in your pics.

Tags

Since e621's wiki is rich and generally well-written, it would be redundant and tedious for me to copy-paste all the possible tags. Instead, all I'll do is to copy-paste the wiki pages which you can read to draw all the informations you need. Good reading!

General tags (views, framing)

• For rear view, low-angle view and the other views, check this page and this one;
• for first person view and all kinds of POV sequences, check this page;
• for full-length portrait and other kinds of framings to use in case only one character is represented in the pic, check this page.

Size tags

• For macro, micro and all the possible size interactions, start from this page;
• for vore and everything related to it, including internal parts, check this page;
• for larger female, smaller male, size difference and all the tags which can be used to define and control the size of multiple characters, check this page and this one.

General tips

Human characters

In order to push the AI to draw human characters, I've found that the best solution is to place the no furry tag right at the beginning of the prompt along with other tags such as human or humanoid. For more consistent results, you can also place the anthro tag in the Undesired content area of the prompt.

Anime V3

This section contains tags and tips which should be useful for you to create size scenarios using NAI's Diffusion Anime V3 model. This model is based on Danbooru's tags. The only advantage this model grants over the newly released Furry V3 model for size pictures is that it supports artist tags, so my recommendation is to choose this model if you want to emulate an artist's artsyle and you don't need the AI to draw a complex size scenario or you have basic needs.

Useful tags for size scenarios

General (camera angles, framings, art styles...)

• from above from below from behind from side dutch angle
• blurry background depth of field
• close-up foreshortening wide shot
• official art official style anime screencap (use these tags to replicate a screenshot from an anime show: check on Danbooru that the anime series is tagged in no fewer than 400 posts; you can see how well NAI can emulate the art style of some popular anime shows from the pic right below, which I personally created as an experiment)
  

• realistic photorealistic
• monochrome greyscale
• traditional media colored pencil (medium) watercolor (medium) graphite (medium) (these are only some of the possible medium tags for traditional media: check this page on Danbooru to learn about the other ones)
• year xxxx 19x0s (style)
• nsfw
• full body cowboy shot upper body lower body

Artist tags

Artist tags play a pivotal role in NAI, mainly for two reasons:

1. they improve the general look of the whole pic;
2. they make the overall aspect of more pics much more consistent across different scenarios and characters;
3. by not putting an artist tag into your prompt, you're basically preventing yourself from benefitting of one of (if not the) most important advantage NAI grants over a local Stable Diffusion installation.

In order to maximize the effect of artist tags, though, you're recommended to keep these things in mind.

1. Artist tags work best when they're placed right at the beginning of the prompt, just in front of the general tags which we've shown in the former paragraph.
2. Make sure to mention artist tags of artists who are tagged in no less than 100-150 posts on NAI: you can refer to this list on Danbooru which sorts all available artists by post counts. The obvious consequence is that the more posts an artist is tagged in, the more accurately your pic will resemble the original artist's artstyle.

3. When picking a specific artist, you should check that the scenario or characters or aesthetical features you have in mind either appear in the artworks of that artist or, at least, aren't too niché or too far from the artist's scenarios. Example: asking the NAI to draw a foot sole by mentioning an artist who has seldomly drawn feet or who's tagged in very few posts on NAI which feature feet will seldomly yield good results, as the feet will probably look extremely hideous or, in the best case, the NAI will draw the foot from a different artist, making the things look out of place. In other words, if you want to pick a specific artist for your pic, you should try to think about a scenario where that artist is particularly good at.

Size-artists' tags

The last point we've written in the previous paragraph should make it clear why selecting a specific size artist for generating size-based pics is usually the best choice. Some size artists, like Karbo, for example, who excel at drawing landscape and at emphasizing the size difference between a character and their surroundings will cause tags such as size difference, miniboy, giantess, landscape etc.etc. to work more efficiently. Do not take this last assertion as a golden rule, though: NAI will still struggle to draw a good size scenario with uncommon size tags, even if you mention a size artist; on the other hand, NAI might be able to generate good size pics for non-size artists who have a high count on Danbooru. The best you can do is to experiment and see how well a certain artist tag might combine with specific tags.
As for the specific size-artists whom NAI can recognize quite easily, I've singled out these few ones, so far:

• karbo
• kuro oolong
• utopia
• uru (uru0000)
• seo tatsuya
• terada ochiko

Non-size-artists' tags

Even though mentioning a size artist is usually preferable, there are still quite a few artists who can work quite well with size scenarios, at least according to my experience. Here are their tags and the things they're good for in round parenthesis:

• sciamano240 (western style, both female and male characters, NSFW, phantasy characters, feet, almost all body types: basically anything works with this artist!)
• ilya kuvshinov (both female and male characters, NSFW, stylish/pretty characters, thin/medium body type, feet; pics genned with this artist tag might look gloomy/dark)
• riz (both female and male characters, NSFW, phantasy characters, thick body type)
• lewdamone (female characters, thick body type)
• asura (asurauser) (both female and male characters, NSFW, feet)
• synecdoche (female characters, thick/plump body types, elves)
• kairunoburogu (both female and male characters, NSFW, feet)
• arttoru (female characters, thick body type, feet)
• bigxixi (female characters, young look, feet / feet in pantyhose or socks)
• artist:nona_moth (both female and male characters, NSFW, feet; use artist: in order to avoid moths and insects in your pic)
• donburi (donburikazoku) (female characters, thick body type)
• shexyo (western style, female characters, thick body type, NFSW)
• abubu (similar to sciamano240, even though it's usually better at thick body types)
• agawa ryou (female characters, NSFW, thick body type)
• seraziel (mostly male and futa characters, NSFW)
• fizintine (both female and male characters, NSFW, feet)
• nottytiffy (female characters)
• cutesexyrobutts (western style, female characters, thick body type)
• rakeemspoon (female characters, thick body type)
• mina cream (female characters, NSFW, feet)
• nyantcha (both male and female characters, NSFW, lots of body types and poses)
• blushyspicy (female characters, thick body type, feet)
• dandon_fuga (western style, female characters, lots of body types and poses)
• artist:afrobull (both male and female characters, NSFW, use artist: to avoid the afro haircut)
• luminyu (both male and female characters, NSFW)
• exlic (both male and female characters, futa, NSFW)
• echo saber (female characters, thick body type)
• cheshirrr (both male and female characters, feet, NSFW)
• combos & doodles (both male and female characters, NSFW)
• artist:automatic_giraffe (female characters)
• ashiomi masato (female characters, NSFW, thick body type)
• nekoda (maoda) (female characters, feet)
• ishikei (both female and male characters, NSFW)
• as109 (both female and male characters, NSFW)
• sincos (both female and male chatacters, NSFW)
• rariatto (ganguri) (both female and male characters, NSFW)
• kinkymation (both female and male characters, NSFW)
• kajin (kajinman) (mostly female characters, NSFW, finger nails, feet)
• shigatake (both female and male characters, NSFW, cute)

Giant person

• giant giant male
• giantess
• size difference height difference
• foreshortening
• building
• skyscraper
• city
• cityscape landscape continent
• growth torn clothes twitching motion lines
• destruction debris ruins rubble
• foot up standing on one leg foot focus
• soles
• stomping imminent stomping
• walking
• gigantic breasts gigantic penis
• pointing pointing at another pinching
• vore imminent vore
• swallowing
• lifting person
• femdom assertive female
• holding
• hands up
• open hand reaching reaching towards viewer
• clenched hand own hands together

Shrunken person

• miniboy minigirl mini person
• stepped on
• struggling
• blood stain
• role reversal
• between breasts
• lying on stomach on side on back
• sitting
• in palm
• shrinking
• breast smother face to breasts
• sandwiched
• deep skin
• vaginal object insertion anal object insertion vaginal object insertion

Size control

Since size tags like giantess, miniboy etc.etc. are extremely generic, the only way to control the size of a character in NAI can be done by using specific tags for the surrounding environment.

Minigiant (10 - 40 feet tall)

This one is probably the trickiest scenario to make in NAI, as it can easily overlap in the following one and it's highly dependent on luck. However, it's still possible to decrease the chances of failure by following all of or some of these few workarounds.

1. Combining giantess, giant male with size difference or height difference usually works, although I've noticed that it's more reliable with indoors tag and rooms than with outdoors and open spaces. According to an anon, also the fake big 1girl, big 1boy tags might work, although personally I didn't have any luck with this. It might be worth a try, though!
2. NSFW scenarios, especially when involving sexual acts (check this page on Danbooru to see the tags), usually show a higher chance of success than SFW ones. In a NSFW scenario, it might happen that the sexual organs are too small or too big compared to one of the characters involved: in that case, inpainting is the only solution. I've noticed that also the wide shot tag might be helpful to show the smaller character in NSFW scenarios.
3. Don't insert more than 3-4 characters.

Standard giant (interacts with smaller people or buildings)

This kind of giant is probably the easiest to make as there are many tags available to create suitable scenarios in NAI. I'm going to write down only the most common ones: street, car, city, town, village, building, skyscraper, people, crowd. When a from below tag is used in conjunction with buildings, it might be necessary to insert the giantess, giant male tag inside multiple curly brackets to force the AI to draw them in the foreground and leave the giant character in the background.

Mega giant

For this kind of giant, it's usually enough to combine the giantess or giant male tags with continent, landscape, cityscape, mountain.

Galactic giant

This interaction requires the space tag along with earth (planet) or planet tags.

Size scenarios with shrunken people: use inpainting!

When trying to create size scenarios featuring a normal-sized character interacting with a smaller one, it’s best to avoid writing “miniboy, minigirl” and other related tags (see the previous page) immediately in the first attempt as the AI generally struggles to create multiple characters and distinguish between them in these kinds of situations. My recommendation, instead, is to create a pic with only one character in a specific pose and setting and then edit the pic through inpainting and place the shrunken character along with the proper tags. You can follow the steps which are described in the following pages.

Step 1

Create a pic featuring only one character in the position and setting you like.

Step 2

Select the “inpaint” tool on the top bar, above the pic you’ve just generated.

Step 3

Draw the area where you want to place the tiny person. Once you’re done, click on Save & Close at the top right.

Step 4

Add mini person and other tags which are related to the tiny person at the beginning of the prompt and insert them into multiple {} (I recommend 6, at least). Click on Generate and you should be able to see a tiny person in your pic now!

Further tips on inpainting and manual editing (by fairy anon)

1. If there's a small error (like hand/fingers/other minor stuff) you can often just mask that area with inpaint and reroll with the same prompt. Inpaint takes into account your prompt and the image as a whole to determine what to draw
2. If you want to add some thing for a person to hold/wear/etc you can also just mask that area and change prompt to include the thing that you want to add and it should work most of the time
3. Usually try to be generous with the buffer zone of the mask, if you make it too restrictive it may not properly add the thing that you want
4. If AI in no way wants to listen to you, you can brute force it by drawing some pixels of the element that you want to be displayed, mask that area and "keep" some of the pixels of the element you just drew so that AI is forced to draw something that resembles the thing with the color that you drew. For example, for yellow clothes to be green, you can draw some green paint over the clothes, then mask the clothes and keep some green pixels on resulting image and most of the time AI will draw that (also don't forget to change prompt to accommodate for that).
5. Sometimes when you have a complex idea that is not really describable by prompt, you can draw some scribbles resembling the thing (just make sure to have proper colors for it), then make a "good enough" prompt for it and try to inpaint with 0.3~0.6 strength and see the result. If let's say on first iteration you got a bit "better" resembling composition, you can use that image as base and go again with the same inpaint strength and see if the next image becomes better. This is just trial and error I think but for really complex stuff this works well.