Hi!
I created AI-powered search engine for iStripper images. There are no predefined tags, just describe what can be seen on the image in English. Try to keep it short, simple, separate with commas and experiment with different wordings. It can even find celebrity look-alikes.

Query examples:

• tight pants
• bikini
• bikini, standing girl, full-size photo
• redhead, face close up, eye contact
• smiling brunette
• shy girl
• Kim Kardashian

By pressing on the image, you’ll enter slideshow mode. Model name and photoset are indicated there.

Keep in mind:
• There are only 47 unique models, 130 photosets, about 7500 images for algorithm to choose from. So, if you don’t get proper results, it may just not exist among this limited number of images or it may not contain enough exact results.
• Algorithm will always output 10 results per query. The best matches are at the top from left to right. Even if you write complete nonsense, you’ll get some results.
• Results are selected only from unique photosets, models may repeat.
• Queries should be in English.
• If nobody is using the service for more than 15 minutes, it enters sleep mode. So, if you are the first one to wake it up, just wait about 10-15 seconds until page loads. From then on, it works fast.

And finally, here is the link: http://bit.ly/aistripper

In order to avoid any legal issues with Totem: this is purely my pet project, I don’t get any money out of it, link is provided only in this forum, all images contain original watermark.

@nowhereman1965 certainly works and interesting results. I searched for

Short shorts
Sexy shoes
Tiny tits
Gloria

Obviously the first three worked fine, but not even one photo of Gloria in the last set? 😕
Never mind I know where to find her without the AI. 😊
Nice idea and I guess in time you might add more data to it perhaps? 👍

@wrightsayswow Regarding searching by model names: AI model was pretrained on millions of different images all over the Internet, then fine-tuned on smaller dataset with girls from iStripper. Girls' names weren't used for training AI model.

It works with Kim Kardashian and other celebrities because their photos were in the initial huge set of images and their names were mentioned there. As there are no Kim's photos among my 7500 images, it just tries to find most similar among them. And it doesn't work for Gloria, because it's just common name and she is not that famous in the Internet compared to non-porn celebrities.

If I were to make production version of it, I would just hardcode girls' names for each image and search for it directly.

No problem. I know it's not a commercial venture, and searching for outfits and body type seemed to work quite well. And I see you click on each image to find which show it is. Very good.
Clever stuff indeed. 👍

This Topic should be maybe pinned in this Section as well. Good Job @nowhereman1965 👍

That's excellent. Good work. Could it be possible to make it as a standalone engine to run on our local collection? Or maybe provide a possibility to export more exhaustive output (more than 10 results) as a text form including set reference and photos filenames for us to retrieve the files in our personal collection? Again very good and useful job! Thank you. 😋

Talking about training so I guess this is an application of the so called "machine learning"?

Thank you for kind words and support!
@Gorfa91 yes, it is machine learning, but more precisely - deep learning.
Deep learning is part of machine learning and they both constitute part of Artificial Intelligence field.
Usually, when model uses neural networks, people call it deep learning as there are a lot of layers stacked on top of each other.

Machine learning models exist for quite some time, truth be told. Depending on your background, you may have heard about linear regression or logistic regression. In modern books they are called machine learning models. Neural nets are next major step in the field of AI.

Regarding local application: maybe, it will be my next project as I don't have much experience here and there are things for me to learn.

Such an engine could be a way to integrate the many missing tags of software, and it could help to provide cards (specially old ones) with right tags! 😵

Thanks for the detailed explanation. As an IT guy I have some basic understanding of the stuff. But I never practiced it just read some slides. My practice in AI field was some programming in Prolog 20+ years ago. That was called "expert systems" a fact database, giving facts and asking questions for the machine to solve. Typical exercices are genealogy, weather forecast, medical diagnosis, etc. 😎

This is a brilliant little project! It seems a bit random in its responses, but that's to be expected with AI.

Deep learning would be an excellent sales tool, if it could be integrated into the product. As @HansSachs says, it elegantly augments the categories and tags, to help users find the shows that suit their particular tastes - and there is quite a variety of particular-ness amongst this community!

Presumably you trained it on the girls' images themselves, as well as publicly available images, so you could include a "Looks like Gloria" function? And would it be a big deal for somebody with a full collection to train it and make it available online? I guess that, since it displays stills from cards that users might not have purchased, Totem would have to support such a venture. And could it include existing tags ("pro dancers that look like Gloria")? And could you supply it integrated with the @Dorsai Playlist Manager (Make me a playlist of pornstars that look like Kim Kardashian)?

Hey, you don't need to answer all these questions. It's just me, rambling. Anyway, I know the answers - it's possible but you haven't done it yet because it's just a pet project and nobody's paying you to do it. But it's still the most interesting innovation in iStripperLand for years. ***** brilliant.

'Useful' Would seem to be an understatment here. 😎👍

@pantalone yes, it's all possible. To make it truly useful for finding new model cards, you need to have the whole collection and add images to it regularly.

Regarding tags: this is just reverse process. Currently we are looking for images that correspond to text, but we can also look for texts corresponding to image. You pass image, have N number of predefined tags and algorithm returns most similar tags in order of relevance.

Also it is possible to find similar images for provided image. Multilingual search without using translation service is also possible to make.

@nowhereman1965 you are a scholar & a gentleman sir! Thank you for your work! You deserve some kinda' award too! ✌💔😎

Could not resist having another play with this. All sorts of searches work very well.
"Eyes", "Smiling, fully nude", "Sitting on haunches, fully nude" and "Bare feet" are shown.
Others I tried:
"Legs, high heels", "Ass up", "Fully nude", "Brunette, fully nude", "Nude, ass up", "Nude, on back"
Most 100% accurate results, some the AI I guess is still learning. Sitting on haunches is one I think it's learning still, and bare feet still showing with some shoes, although at least the type that show plenty of the foot.
The results are the same if repeated, but if the available images are added to that will change. And as @Gorfa91 suggested if the results per search are increased over the current 10 that would be good.
Love it! 👍

Sadly it is not available anymore. Wish Totem could implement such engine to search through our collections. 👍