Jump to content
Search In
  • More options...
Find results that contain...
Find results in...
Existing user? Sign In

Sign In



Sign Up

AI generated omorashi pictures - Anyone willing to make a custom training dataset?


Recommended Posts

I saw some other topics with samples of AI generated art and pictures and decided to try it myself. Spoiler: it's nightmare-inducing bad!

For people who are not familiar, you give it a text prompt and an image matching it will be generated from scratch.

The problem is that the training sets, which are used to "teach" the AI what is what, most likely contain very few pee related images. For example, you can see in the prompts I tried below that the AI loosely associates "pee" with toilets, pants down, legs showing, and crouching.

I propose we create a custom set of training data, images and labels, with pictures of pee accidents, wet pants, etc. Of course we would need permission to use those images, which is the hardest part. The easy part is that not that many images seem to be necessary for greatly improving the model.

These are some of the prompts I tried:

woman peeing in her pants in the park

1040938075_womanpeeinginherpantsinthepark.png.a6f02ef2cde21cdf365bd2510dbf5aaf.png

woman peeing her panties in the bedroom

7453880_womanpeeingherpantiesinthebedroom.png.e3c16f3455a5044812f891989c748f46.png

woman wearing diapers in a busy street in the city at night

1936670061_womanwearingdiapersinabusystreetinthecityatnight.png.68c57fcd5e1585db5f5b6e36a3f20c37.png

woman wetting her pants in a busy train

1875904739_womanwetingherpantsinabusytrain.png.c90f8b22b28c205f45d99dbfe77f57b2.png

adult person peeing in their clothes

1353363979_adultpersonpeeingintheirclothes.png.180f38edbf46a5185bc445cb60eab89f.png

 

Edited by JulesH (see edit history)
Link to comment

Getting the community together for a custom-made training set sounds like a great idea! I've seen AI produce incredible results with sets as small as 50 images, and I'm sure we could go bigger than that with enough effort (but that was for image labeling rather than image generation so it may be different).

If we were to do this, we should do our best to diversify the input data. If we feed in 50 images of the same person in the same outfit in the same location, then we can't expect very good output. I'd love to see the result with many people's input though!

Link to comment
  • 2 weeks later...

https://www.craiyon.com/ (formly known as DALL-E Mini) seems to have some minor concept of pee. If you put "a women needs to pee" you might get an image that looks like a woman that needs to pee, but mostly not. It knows what urine is, but can't combine it with other ideas as the urine goes away. Maybe something to watch for the future because they'll also working on a larger AI model previously called DALL-E Mega. 

craiyon_2022-6-24_0-2-49.png

Link to comment

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Paste as plain text instead

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

  • Recently Browsing   0 members

    • No registered users viewing this page.
×
×
  • Create New...