Jump to content
Existing user? Sign In

Sign In



Sign Up

Using an AI to write descriptions of videos for HD Wetting and HD Diapers.


Recommended Posts

One of the time sucks in running HD Wetting and HD Diapers is just the time spent having to write the video descriptions for the sites.  So, I thought wouldn't it be great if an AI could write the descriptions for me?  To facilitate this I started developing a program based on the GPT-2 language model from OpenAI to take key words or events that I feed it and write a coherent description of the video from that.

The results are interesting, but not exactly what I need.  For example, I tried to use it to write a description for a video that features Alisha intentionally peeing in her pull-on diaper.  The biggest issue I am having is the AI tends to use its knowledge of language and story to bring in other characters and elements that I didn't feed it, and weren't in the video.  Here is the description it wrote-

Quote

Wearing nothing but a diaper, Alisha intentionally pees on a woman while her boyfriend (whose name is not yet known) watches and laughs. Then, Alisha pees on another woman.

When a woman complains, she is threatened by Alisha with a gun: "That's what happens if you don't go along with me and don't pay up!"

After the man's ex-girlfriend tells him she is pregnant, Alisha sneaks into the woman's house and pees on her.

This time, the man has a gun and kills Alisha.

"The defendant is in fact not just a murderer but a sexual predator who preyed on young, vulnerable victims, repeatedly and violently," Acting District Attorney Mary Keenan said.

Alisha, who was 17 at the time, will be sentenced March 8 in Multnomah County Circuit Court.

 

Link to comment
38 minutes ago, TVGuy said:

One of the time sucks in running HD Wetting and HD Diapers is just the time spent having to write the video descriptions for the sites.  So, I thought wouldn't it be great if an AI could write the descriptions for me?  To facilitate this I started developing a program based on the GPT-2 language model from OpenAI to take key words or events that I feed it and write a coherent description of the video from that.

The results are interesting, but not exactly what I need.  For example, I tried to use it to write a description for a video that features Alisha intentionally peeing in her pull-on diaper.  The biggest issue I am having is the AI tends to use its knowledge of language and story to bring in other characters and elements that I didn't feed it, and weren't in the video.  Here is the description it wrote-

 

I highly doubt this is real but if so that's fucking hilarious.

Link to comment
51 minutes ago, Bulge_Lover said:

I highly doubt this is real but if so that's fucking hilarious.

It is real... And not all that difficult to implement.  The language model itself its called GPT-2 and is from the OpenAI initiative.  I used the Python library developed by Hugging Face for GPT-2, as it seems to be the most accessible and popular library for such purposes available.  You can read more about it here- https://medium.com/huggingface/100-times-faster-natural-language-processing-in-python-ee32033bdced

Link to comment
52 minutes ago, Redd said:

I'm guessing the actual video isn't that dramatic, lol. 

No, not nearly.  It was really just a simple diaper wetting video with just Alisha.  She didn't pee on anyone else, there was no boyfriend character, all of that was the invention of the AI.

While I'm not sure that I'll be able to turn to AI to write descriptions anytime in the near future, I may possibly turn to it to get some ideas for a video.

Link to comment

TVGuy,

Maybe a better solution would be if people here contributed video descriptions from which you would produce the videos! For example,

Alisha and a group of friends sit around talking in a pub, and the conversation somehow comes round to bed wetting. John mentions that he has heard that some people wet their beds for fun. The two girls giggle and blush.

Then John goes off, and the two girls walk home discussing the idea of wetting the bed for fun. Neither of them wants to say how much the idea excites them.

Then we see them in bed dressed in nightdresses, teasing each other to pee a little bit.

Finally they roll back the cover to reveal a huge wet patch, and move into a 69 position...

 

Edited by David_E (see edit history)
Link to comment

TVGuy,

I am kind of curious about any application of AI, because I am not sure it is really all it is cracked up to be. I mean, what is the AI program actually given - the video would surely be too hard for it to understand. Does it try to comprehend the audio channel? Does it already know about Omorashi, and do you supply a suitable vocabulary?

Leaving that aside, I think I have a great alternative. Why not supply a link to one complete video right here, and let people supply the description. Then you can select the best effort and give them the job of creating descriptions for your subsequent videos!

Link to comment
On 11/15/2019 at 3:49 PM, David_E said:

TVGuy,

Maybe a better solution would be if people here contributed video descriptions from which you would produce the videos! For example,

Alisha and a group of friends sit around talking in a pub, and the conversation somehow comes round to bed wetting. John mentions that he has heard that some people wet their beds for fun. The two girls giggle and blush.

Then John goes off, and the two girls walk home discussing the idea of wetting the bed for fun. Neither of them wants to say how much the idea excites them.

Then we see them in bed dressed in nightdresses, teasing each other to pee a little bit.

Finally they roll back the cover to reveal a huge wet patch, and move into a 69 position...

 

I do like this idea!  And, I have in fact in the past produced videos based on stories that people have posted here (with the author's permission, of course).  The problem is that it is often difficult to be aware of all of the constraints of a particular shoot more than a couple of weeks in advance.  So, it can be difficult, for example, to count on shooting a video where the action will take place on a bed if, shortly before the scheduled shoot date, our main location drops out and our backup doesn't offer a bedroom or bed for filming.

On 11/17/2019 at 2:28 PM, Ranpalan said:

Interesting, though I'm not sure what kind of results you can expect even perfect AI to come up with if it doesn't actually have access to the video.

I wonder if taking random screenshots of the video, getting labels for what's on them, and then feeding that into a model that strings them into some kind of text would work better.

I was not expecting perfect results.  Rather, I was hoping for something that would give me a good start and only need a small bit of editing.

21 minutes ago, David_E said:

TVGuy,

I am kind of curious about any application of AI, because I am not sure it is really all it is cracked up to be. I mean, what is the AI program actually given - the video would surely be too hard for it to understand. Does it try to comprehend the audio channel? Does it already know about Omorashi, and do you supply a suitable vocabulary?

Leaving that aside, I think I have a great alternative. Why not supply a link to one complete video right here, and let people supply the description. Then you can select the best effort and give them the job of creating descriptions for your subsequent videos! 

In this case, the input I gave the AI was a simple list of the actions in the video that I typed out.  The intent was for the AI to take that list of actions and flush it out as full sentences formatted into paragraphs with more colorful language and descriptions.  The AI itself had been trained on various samples of text from around the web and was not Omorashi specific.

For me, AI is a tool, and I do use it in other way s with my work.  For example, I often use automated, AI driven tools to remove unwanted objects that might creep into a shoot.  Otherwise, you might accidentally see a microphone come into the scene or a production assistant eating a donut in the background.  AI also helps in post with matching up the recorded video with the separately recorded audio, and in categorizing clips based on their content to speed up editing.

Link to comment
Guest fluidfox

Just a thought but perhaps it would be easier to hand code this with python instead of relying on the AI. I'm willing to bet that a lot of your work follows patterns and there are certain repeating themes and locations. For example:

actress_name has gone to visit her partner_gender for a occasion_name at a location, but when she arrives she desperately needs to pee_verb + gender_reference...

Lucy has gone to visit her boyfriend for a hot weekend at a luxurious hotel, but when she arrives she desperately needs to relieve herself...

This is a very simple example but you get the idea.

If you wrote a dozen of these scripts and then used a combination of predetermined variables and randomly generated verbs you could probably get a pretty accurate description, or at least one that only requires minor editing and they are different enough that people won't notice the repetition. Throw in some conditionals for the numbers of actors or whatever et voila! 

Sure it would take a few hours of coding but I bet there would be hundreds of possible variations.

Like I said, just a thought...

Link to comment
On 11/19/2019 at 6:29 PM, TVGuy said:

I do like this idea!  And, I have in fact in the past produced videos based on stories that people have posted here (with the author's permission, of course).  The problem is that it is often difficult to be aware of all of the constraints of a particular shoot more than a couple of weeks in advance.  So, it can be difficult, for example, to count on shooting a video where the action will take place on a bed if, shortly before the scheduled shoot date, our main location drops out and our backup doesn't offer a bedroom or bed for filming.

LOL - you shouldn't let constraints like this cripple your artistic flair!

Why not spread stories like this over several shooting sessions - do the pub scene one week, then maybe the walking home scene is filmed on another occasion, and the bed scene on yet another. Once you got that started, there would be no reason why you wouldn't produce just as many videos as before, and they might seem more varied and unpredictable.

On 11/19/2019 at 6:29 PM, TVGuy said:

For me, AI is a tool, and I do use it in other way s with my work.  For example, I often use automated, AI driven tools to remove unwanted objects that might creep into a shoot.  Otherwise, you might accidentally see a microphone come into the scene or a production assistant eating a donut in the background. 

Doesn't that mean that you have to train the AI up on all possible unwanted objects?

Link to comment
8 hours ago, fluidfox said:

Just a thought but perhaps it would be easier to hand code this with python instead of relying on the AI. I'm willing to bet that a lot of your work follows patterns and there are certain repeating themes and locations. For example:

actress_name has gone to visit her partner_gender for a occasion_name at a location, but when she arrives she desperately needs to pee_verb + gender_reference...

Lucy has gone to visit her boyfriend for a hot weekend at a luxurious hotel, but when she arrives she desperately needs to relieve herself...

This is a very simple example but you get the idea.

If you wrote a dozen of these scripts and then used a combination of predetermined variables and randomly generated verbs you could probably get a pretty accurate description, or at least one that only requires minor editing and they are different enough that people won't notice the repetition. Throw in some conditionals for the numbers of actors or whatever et voila! 

Sure it would take a few hours of coding but I bet there would be hundreds of possible variations.

Like I said, just a thought...

That is a great thought @fluidfox, and I am already working on something kind of like it.  For the past year or so I have been working on a long term project, rewriting and reengineering HD Wetting from the ground up, and the new version does indeed feature a system much as you described.  Certain variable elements for the video can be selected via dropdowns, with randomly selected pre-written components and structures, in order to generate a unique description for each scene.  The description generated would then be able to be edited, but it would give me a head start.  I had just figured that trying this with AI instead might save me a bit of time, at least if the results work out.

 
 
 
 
20 minutes ago, David_E said:

Doesn't that mean that you have to train the AI up on all possible unwanted objects?

Nope, I simply have to paint a selection over the unwanted object.  The AI can than use the other content in the frame, as well as content from adjacent frames, to fill in the painted area.  It can also track motion if the unwanted object or the camera movies.

Link to comment
19 hours ago, TVGuy said:

Nope, I simply have to paint a selection over the unwanted object.  The AI can than use the other content in the frame, as well as content from adjacent frames, to fill in the painted area.  It can also track motion if the unwanted object or the camera movies.

OK - but that means you have to do quite a bit of work by hand.

What I still don't understand, is that surely the task of describing a wetting scene is minute compared with everything else - devising a good 'excuse' for a wetting, getting the scene right - setting up the lighting, controlling the camera action cleaning up the mess (!!), interviewing potential actresses, etc etc.

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...