AI Art and Learning Thread

by **BellyHungerLady** » Wed Oct 26, 2022 7:44 am

I did some AI art, inspired by Sarus' thread. However, it didn't turn pretty well.

That's why I encourage you to feel free to give tips, examples and ideas for improving the generation of vore related pics.

I will post some pics as sets, as they look similar to each other.

Consider to post your own creation and the process you followed to do them

Have a vorific day :gulp:

by **Chessa** » Wed Oct 26, 2022 8:03 am

One tip I have for you immediately is to use a baseline of tags.

You're using Novel AI right?
In that case use solo, female, (Add each characteristic as a tag), (Add one style as a tag)
This is what I do and it works well, though you do have to fidget and alter tags to get a good result still.

When you get a good result, copy the tag and put it at the bottom in the tag box and try generating more with that seed.

Oh and surrounding tags like {{{Like this,}}} (With more or fewer brackets for priority) helps a lot.

by **BellyHungerLady** » Wed Oct 26, 2022 8:07 am

Chessa wrote:One tip I have for you immediately is to use a baseline of tags.

You're using Novel AI right?
In that case use solo, female, (Add each characteristic as a tag), (Add one style as a tag)
This is what I do and it works well, though you do have to fidget and alter tags to get a good result still.

When you get a good result, copy the tag and put it at the bottom in the tag box and try generating more with that seed.

Oh and surrounding tags like {{{Like this,}}} (With more or fewer brackets for priority) helps a lot.

What kind of tags are considered as "styles"?

The seed and the tags are the same? When I tried to write on seeds, I got no results (I couldn't write on it)

by **anonpls** » Wed Oct 26, 2022 8:07 am

I recommend making 2 Tag lists.
1 List of tags that create succesfull Art
1 List of tags that create unsuccesfull Art.

Also by being very specific and add alot of tags for the generation will create more succesfull results.
The A.I. will just randomise some results if left with room for interpretation.

A.I. Art is not just training A.I. also learning what works and what not.

by **BellyHungerLady** » Wed Oct 26, 2022 8:10 am

anonpls wrote:I recommend making 2 Tag lists.
1 List of tags that create succesfull Art
1 List of tags that create unsuccesfull Art.

Also by being very specific and add alot of tags for the generation will create more succesfull results.
The A.I. will just randomise some results if left with room for interpretation.

A.I. Art is not just training A.I. also learning what works and what not.

So if I make a list of, for example, 20 tags, would it be easier to work in a good way? I supposed it to be worse for generation, so I used around 3 to 6 tags

by **anonpls** » Wed Oct 26, 2022 8:57 am

It´s try and error. When using a long list of tags, the chance that there are conflicting tags increases. First find a couple of tags that work together then experminent with increasing the list.

by **AlluringPredation** » Wed Oct 26, 2022 11:16 am

OP looks like they're using some form of Stable Diffusion locally, maybe trying to force it to produce anime as that is not NovelAI style imagery. Actual vore does not lend well to AI as the tricks used to produce good output place in negative prompting will attempt to fix anatomical issues, which vore can not occur in real life so it's going to be an uphill battle.

For those that use NovelAI, I suggest https://danbooru.donmai.us/wiki_pages/tag_groups Notice that it is avoidant in producing explicit images unless you enter NSFW as a tag at some point in the prompt. Be aware that explicit tags will not be suggested, but often work.

As for artists who may browse this thread. Best use of the AI is to provide instant references, or to create variation references on your current art piece to see different directions that your image could take, especially useful if you hit a roadbock; Using elements you like as a reference to keep your art pure and untainted by the filthy AI.

And for those with even the tiniest bit of artistic knowhow and the tools for the job, image2image is a powerful tool giving greater context to the AI than words alone can achieve. Of course the more you know the better the result will be.

by **BellyHungerLady** » Wed Oct 26, 2022 2:48 pm

That's a good bunch of information! Thank you

by **taylorsm** » Thu Oct 27, 2022 12:01 am

If you're interested, I've generated some nice internal views:
nsfw:

Spoiler: show

Explanation here:
viewtopic.php?f=18&t=63790&p=2928205#p2928205

by **BellyHungerLady** » Thu Oct 27, 2022 12:11 am

taylorsm wrote:If you're interested, I've generated some nice internal views:
nsfw:
Spoiler: show

00248-3504280838-(((masterpiece))),best quality, illustration, straight body, (detailed face), {messy hair}, squeezed 1girl sliding down wet hole.png

00880-3331200333-(((masterpiece))),best quality, illustration, straight body, (detailed face), {messy hair}, squeezed 1girl, restrained, (fetal p.png

Explanation here:
viewtopic.php?f=18&t=63790&p=2928205#p2928205