NovelAI character generation tips/prompts

[Corty]

I was thinking about where to put this, but I went with the general chat in the end. Sorry if it was a miss.

Anyway, have any of you here used NovelAI's image creator tool? And if yes, care enough to share some wisdom? Pretty please?

I was playing around with tags, prompts, whatever it is called, using the brackets { } to enhance wanted and unwanted tags to creep in. What I am struggling with is creating a character with it with different eye colors. Heterochromia usually comes back with green and yellow eyes, and even if including those in the unwanted section, they still creep in if I add the tag of heterochromia.

I am also planning to go around and create some scenic backgrounds (fantasy themed). I am yet to try that out. Has any of you have some tips on what to look out for?

Here are some samples I was satisfied with in the end, after playing with it for a few hours:

Spoiler: Images

As you can see in the last image, the eyes get jumbled up when trying to prompt out a character sheet-like image. Also, I just can't pinpoint which of her eyes gets red or blue. I know that it can be fixed in photoshop by image flipping, but still. If someone has experience in it, I would appreciate some tips!

 

[DollarForADevil]

Huh never heard of this makes me wonder if you use this to help you write lol. Eh I mean I'd probably work with these kinds of things, people don't really read the novels they read here, all it needs now is an ASCII Generator that keeps track of the current level or progression of the character.

 

[Corty]

Nope. I would not trust an AI to write a coherent story that is also enjoyable.

Plus I would argue with that people don't read novels here... or I wouldn't post them here in the first place...

 

[Syringe]

Heterochromia is one of the hit or misses with AI generation. BUT there is a really consistent way I've found.

Used some of my Anlas to generate this one for you. Instead of Heterochromia, put in {{right blue eye}}, {{left red eye}} - Or whatever orientation you want. I found it far more consistent.

The full generation for this was:

(masterpiece, highres, ultra detailed face, blonde hair, long hair, twin tails, {{right blue eye}}, {{left red eye}}, full body, wearing royal robes, dynamic pose, small chest, running in a fantasy forest).

Don't just use tags. The AI is flexible enough where you can say: Sleeping on the bed, fighting a monster, etc - Like how I used 'wearing royal robes' here.

Also as a rule of thumb - Set undesired contents to none, and replace the filters with this: 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

Scenery is very easy to do. Just add ____ in the background or and action like "running in ____".

ALSO and most importantly, use image to image. If you generate an image your REALLY like, save it use it as a template for the AI to draw from.

Alternatively, you can even edit the template and COLOR in the eyes. From there you really just have to tweak the strength.

65-75 keeps it really close. 80-90 adds creativity. 99 is basically giving the wheels to the AI.

I hope this helps!

 

[Corty]

Heterochromia is one of the hit or misses with AI generation. BUT there is a really consistent way I've found.

View attachment 16669

Used some of my Anlas to generate this one for you. Instead of Heterochromia, put in {{right blue eye}}, {{left red eye}} - Or whatever orientation you want. I found it far more consistent.

The full generation for this was:

(masterpiece, highres, ultra detailed face, blonde hair, long hair, twin tails, {{right blue eye}}, {{left red eye}}, full body, wearing royal robes, dynamic pose, small chest, running in a fantasy forest).

Don't just use tags. The AI is flexible enough where you can say: Sleeping on the bed, fighting a monster, etc - Like how I used 'wearing royal robes' here.

Also as a rule of thumb - Set undesired contents to none, and replace the filters with this: 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

Scenery is very easy to do. Just add ____ in the background or and action like "running in ____".

ALSO and most importantly, use image to image. If you generate an image your REALLY like, save it use it as a template for the AI to draw from.
View attachment 16671
Alternatively, you can even edit the template and COLOR in the eyes. From there you really just have to tweak the strength.

View attachment 16672
View attachment 16670

65-75 keeps it really close. 80-90 adds creativity. 99 is basically giving the wheels to the AI.

I hope this helps!

Woah! Thank you very much! That is awsome and helpful! Luckily I saved all the images, seeds and tags for future creations so I can go at it again any time. Thanks again!

 

[DollarForADevil]

Ah this is by far my favorite AI Generated Image, I'm thinking of a story just for it.

 

[ACertainPassingUser]

Ah this is by far my favorite AI Generated Image, I'm thinking of a story just for it.

This morning Liam awake as Trash can.
He lives alone in apartment, and no other family member bothering him.

Liam nonchalantly walk out of home and the street. The surrounding is Quiet and empty, no sound of street vendors and pedestrian passing by. Not even the sound of cars on the back road nor the train on the bridge.

The world is suddenly empty and the last human is genderbended into trashcan.

 

[Corty]

This morning Liam awake as Trash can.
He lives alone in apartment, and no other family member bothering him.

Liam nonchalantly walk out of home and the street. The surrounding is Quiet and empty, no sound of street vendors and pedestrian passing by. Not even the sound of cars on the back road nor the train on the bridge.

The world is suddenly empty and the last human is genderbended into trashcan.

I bet this story would have a vore undertone after a chapter or so.

 

[melchi]

I was thinking about where to put this, but I went with the general chat in the end. Sorry if it was a miss.

Anyway, have any of you here used NovelAI's image creator tool? And if yes, care enough to share some wisdom? Pretty please?

I was playing around with tags, prompts, whatever it is called, using the brackets { } to enhance wanted and unwanted tags to creep in. What I am struggling with is creating a character with it with different eye colors. Heterochromia usually comes back with green and yellow eyes, and even if including those in the unwanted section, they still creep in if I add the tag of heterochromia.

I am also planning to go around and create some scenic backgrounds (fantasy themed). I am yet to try that out. Has any of you have some tips on what to look out for?

Here are some samples I was satisfied with in the end, after playing with it for a few hours:

Spoiler: Images

As you can see in the last image, the eyes get jumbled up when trying to prompt out a character sheet-like image. Also, I just can't pinpoint which of her eyes gets red or blue. I know that it can be fixed in photoshop by image flipping, but still. If someone has experience in it, I would appreciate some tips!

If prompts don't work I'd suggest making a sketch and uploading it. Also like what @Syringe said try just writing out what you want. Perhaps a blond girl with blue and red eyes running.

For a background I found out on reddit stuff like "woods(background)" for the prompt works well.

 

[RaizarP]

I've been using NovelAI image generation since they released it.
I've tried fantasy image with NovelAI too. Don't just use tag and use your imagination and type whatever you want.

These two are something that I created while experiencing with background. The prompts are:
masterpiece, best quality, {masterpiece}, 1girl, unkempt long black hair, green eyes, fantasy knight, knight uniform, shorts, fantasy city background

Also, don't forget to tweak the Scale and Steps too. I used Scale: 4 and Steps: 34 for these images.

You can also add undesired content for lowering the chance of something unnecesarry. For example I used:
nsfw, 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, lowres, {bad anatomy}, {{{{bad hands}}}}, text, error, deformed legs, extra legs, missing fingers, missing arms, missing hand, {deformed hands}, multiple limbs, deformed limbs, extra digit, fewer digits, {{cropped}}, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, {{blurry}}

For heterochromia, you can use something like {heterochromia, red eye right, blue eye left} in one bracket like this. It worked wonderfully.

For example is this!

The prompt I used for that one is:
ultra detailed, best quality, amazing, 1girl, small breast, finely detail, depth of field, extremely detailed CG unity 8k wallpaper, masterpiece, mature style, androgynous, {short red hair}, {{heterochromia, green eyes left, red eyes right}}, sharp eyes, floating hair, light shafts, soft focus, character focus, disheveled hair, flowing hair, floating, noble dress, luxurious library, writer

Steps: 28 Scale: 4

Don't forget to use the NAI Diffusion Anime (Full) to get a better image! Just add NSFW as your undesired content and you will be ready to create some AI images.

I hope this help

 

[Corty]

I've been using NovelAI image generation since they released it.
I've tried fantasy image with NovelAI too. Don't just use tag and use your imagination and type whatever you want.

These two are something that I created while experiencing with background. The prompts are:
masterpiece, best quality, {masterpiece}, 1girl, unkempt long black hair, green eyes, fantasy knight, knight uniform, shorts, fantasy city background

Also, don't forget to tweak the Scale and Steps too. I used Scale: 4 and Steps: 34 for these images.

You can also add undesired content for lowering the chance of something unnecesarry. For example I used:
nsfw, 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, lowres, {bad anatomy}, {{{{bad hands}}}}, text, error, deformed legs, extra legs, missing fingers, missing arms, missing hand, {deformed hands}, multiple limbs, deformed limbs, extra digit, fewer digits, {{cropped}}, worst quality, low quality, normal quality, jpeg artifacts, signature, watermark, username, {{blurry}}

View attachment 16673View attachment 16674

For heterochromia, you can use something like {heterochromia, red eye right, blue eye left} in one bracket like this. It worked wonderfully.

For example is this!
View attachment 16675

The prompt I used for that one is:
ultra detailed, best quality, amazing, 1girl, small breast, finely detail, depth of field, extremely detailed CG unity 8k wallpaper, masterpiece, mature style, androgynous, {short red hair}, {{heterochromia, green eyes left, red eyes right}}, sharp eyes, floating hair, light shafts, soft focus, character focus, disheveled hair, flowing hair, floating, noble dress, luxurious library, writer

Steps: 28 Scale: 4

Don't forget to use the NAI Diffusion Anime (Full) to get a better image! Just add NSFW as your undesired content and you will be ready to create some AI images.

View attachment 16676

I hope this help

This helps a LOT, thank you very much!

 

[CarburetorThompson]

Just do it normally then put the image into gimp/photoshop/procreate and edit the hue of one of the eyes.

 

[TheEldritchGod]

Nope. I would not trust an AI to write a coherent story that is also enjoyable.

Plus I would argue with that people don't read novels here... or I wouldn't post them here in the first place...

The ai novel writer socks, but it is good as an automated second pair of eyes. Feed it one paragraph at a time and ask it to rewrite and it gives a 5th graders version of what you wrote.

I find it works well for pointing out minor writing errors you'd miss otherwise. Like, here is a weak dull version of what you wrote with every writing rule followed. Great for fixing parsing errors.

 

[Erios909]

 

[melchi]

Figured I'd try as well just for grins.

 

[greyblob]

I've been using the novelai model locally for some time. it's probably what they're running on their website or some variation of it. best results i got were when i looked up generated images similar to what i want and copied the prompt and tweaked it a bit.
any gallery is fine. i used this

https://lexica.art/

i only have these 2 pics on my phone and I'm not getting out of bed.