Viridian's AI Experiments

Artificial Intelligence is here! Really! Anything created with AI assistance, including stories, should be posted here.
User avatar
bogbud
Posts: 724
Joined: Sat May 30, 2015 6:43 am
Location: Stuck and sinking

Re: Viridian's AI Experiments

Postby bogbud » Wed Dec 07, 2022 8:45 pm

A whole lot them looks really good.
Although i guess the A.I. (yet) has no idea about the right (distressed) attitude the models should have. Most of them look like they are posing or having a good time.
Is it possible to create a wader-booted girl like my avatar? :)
I'm already chindeep in this mudbog and every desperate attempt to move my stuck legs only drives me deeper in. The thick mud slowly swamps my waders and my arms have nothing to hold onto.
I'm feeling home.

Viridian
Posts: 1587
Joined: Wed Apr 15, 2009 10:03 am

Re: Viridian's AI Experiments

Postby Viridian » Thu Dec 08, 2022 8:36 am

A few that turned out really well, based on my artwork.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/

Viridian
Posts: 1587
Joined: Wed Apr 15, 2009 10:03 am

Re: Viridian's AI Experiments

Postby Viridian » Thu Dec 08, 2022 9:10 am

bogbud wrote:A whole lot them looks really good.
Although i guess the A.I. (yet) has no idea about the right (distressed) attitude the models should have. Most of them look like they are posing or having a good time.
Is it possible to create a wader-booted girl like my avatar? :)


Part of it is that the AI needs to be told what the emotion should be. Like a search engine, it provides the best match based on algorithm data. For example, an image with tags like "wet shirt" and "nipples showing" may have a chance of returning a character that has a blushing face, but it might not. If you say "embarrassed", it's an abstract feeling that a machine doesn't feel, but its algorithm is likely to bring up blushing, maybe even turn the character's head in shame, but it isn't the same as specifying it. The more specific the parameters, the more precise the image result.

The better you are at describing specific things in incredibly specific detail, the more the AI will work for you. As I've discovered, it's really hard to capture what we think of as "quicksand", so I've had better luck with "mud" and "sand", but there's no way that I can see to make the AI submerge to a specific level or body part, hence the need for source images, but the AI can struggle to decide how far it strays from it depending on Strength - lower means it replicates the original closer, but the style is horrible because it tries to duplicate the exact style without transforming it. Too much freedom, and it completely changes it (e.g. changes perspective, angle, depth, etc.).

From my experience so far, it's very good at generating ideas for you. For example, in reading a novel recently,I was trying to visualise a particular character who walks into a room realising that their thin t-shirt is see-through and her nipples are visible.I plugged in the parameters with what I interpreted to be in the scene, and it generated something that looked very close to what I imagined it to be. However, since I didn't have a specific vision of what they had to look like, I wasn't as fussed about the final result.

Anyway, I ran your image a few times. These were the best results.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/

Viridian
Posts: 1587
Joined: Wed Apr 15, 2009 10:03 am

Re: Viridian's AI Experiments

Postby Viridian » Thu Dec 08, 2022 9:16 am

And just to illustrate the point even further, in my tags, I specified a dozen things, but had only included "bright yellow waterproof coat" for attire. So I ran my best image through it again, this time including "bright yellow waterproof PANTS", and these the results.

The art to using AI generators is in finding the right combination and seed and adding different parameters to fine-tune the product.

I then ran the same image, changed all the parameters - blonde hair, wet white tshirt, denim ripped jeans - and got the final image.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/

OBI-wan
Posts: 1323
Joined: Tue Nov 24, 2009 3:46 am

Re: Viridian's AI Experiments

Postby OBI-wan » Thu Dec 08, 2022 2:04 pm

Here’s a few AI’sI cooked up.
You do not have the required permissions to view the files attached to this post.
"Who's the more foolish, the fool, or the fool who follows him?"

User avatar
MadMax359
Posts: 4214
Joined: Sun Apr 19, 2009 5:07 am
Location: north carolina

Re: Viridian's AI Experiments

Postby MadMax359 » Thu Dec 08, 2022 3:49 pm

these are amazing, keep up the good work!
and all these are Medium Breasts?? they are fine by me! :twisted:
The strong do what they want, the weak do what they must

Viridian
Posts: 1587
Joined: Wed Apr 15, 2009 10:03 am

Re: Viridian's AI Experiments

Postby Viridian » Thu Dec 08, 2022 4:26 pm

I've figured it out. I've been busy "training" the AI - and it turns out that my "Quicksand Studies" series (https://www.deviantart.com/viridianqs/g ... nd-studies) has been incredibly useful in getting the parameters for the visual source. My art style seems to make it easier for the AI to realise what I'm trying to depict. I am now able to submerge my characters.

This is the testing I did.

I used the base image, my QS study, which examined a character in a specific position or depth. My base parameters were always "realistic, mud, quicksand" - and in most cases I kept the "large breasts" parameter for the sake of boobs. I mean consistency. I then added some physical description - typically the hair colour and length, maybe the facial expression, but otherwise I let the AI take it away. I would then change the attire, background and other accessories as desired to see how well it adapted.

My first image attempt was always to replicate my original image. In this set, I then changed to shirt/jeans/boots + forest background, then wet sweater + boots + jeans + grasslands background, and finally white blouse + green skirt + canyon background.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/

User avatar
bogbud
Posts: 724
Joined: Sat May 30, 2015 6:43 am
Location: Stuck and sinking

Re: Viridian's AI Experiments

Postby bogbud » Thu Dec 08, 2022 5:36 pm

Viridian wrote:
bogbud wrote:A whole lot them looks really good.
Although i guess the A.I. (yet) has no idea about the right (distressed) attitude the models should have. Most of them look like they are posing or having a good time.
Is it possible to create a wader-booted girl like my avatar? :)


Part of it is that the AI needs to be told what the emotion should be. Like a search engine, it provides the best match based on algorithm data. For example, an image with tags like "wet shirt" and "nipples showing" may have a chance of returning a character that has a blushing face, but it might not. If you say "embarrassed", it's an abstract feeling that a machine doesn't feel, but its algorithm is likely to bring up blushing, maybe even turn the character's head in shame, but it isn't the same as specifying it. The more specific the parameters, the more precise the image result.

The better you are at describing specific things in incredibly specific detail, the more the AI will work for you. As I've discovered, it's really hard to capture what we think of as "quicksand", so I've had better luck with "mud" and "sand", but there's no way that I can see to make the AI submerge to a specific level or body part, hence the need for source images, but the AI can struggle to decide how far it strays from it depending on Strength - lower means it replicates the original closer, but the style is horrible because it tries to duplicate the exact style without transforming it. Too much freedom, and it completely changes it (e.g. changes perspective, angle, depth, etc.).

From my experience so far, it's very good at generating ideas for you. For example, in reading a novel recently,I was trying to visualise a particular character who walks into a room realising that their thin t-shirt is see-through and her nipples are visible.I plugged in the parameters with what I interpreted to be in the scene, and it generated something that looked very close to what I imagined it to be. However, since I didn't have a specific vision of what they had to look like, I wasn't as fussed about the final result.

Anyway, I ran your image a few times. These were the best results.


Thanks for these, they are looking great and depth, level of stuckness and still visible boot-tops are well done.
I may provide you with other source-images from time to time from now on ;)
I'm already chindeep in this mudbog and every desperate attempt to move my stuck legs only drives me deeper in. The thick mud slowly swamps my waders and my arms have nothing to hold onto.
I'm feeling home.

Viridian
Posts: 1587
Joined: Wed Apr 15, 2009 10:03 am

Re: Viridian's AI Experiments

Postby Viridian » Fri Dec 09, 2022 6:03 am

Next set is based on my "Knees" item. First image replicates the original, then I swapped the parameters for different costumes and settings, including an explorer in an ancient ruin and a school uniform in a park.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/

Viridian
Posts: 1587
Joined: Wed Apr 15, 2009 10:03 am

Re: Viridian's AI Experiments

Postby Viridian » Fri Dec 09, 2022 6:18 am

Next one is "Thighs". As you can see, the AI copies poses very well. Original recreated, then swapping out parameters. #2 is school uniform in construction site. #3 was a shipwrecked theme, with red blouse, black panties, beach background and shipwreck. #4 included quarry background, wet red blouse, black skirt and rain. I had some challenges in switching out colours. The AI is too attached to the original colours and brute-forces them in despite specific instruction for a different colour shirt, etc.
You do not have the required permissions to view the files attached to this post.
Viridian @ deviantART: http://viridianqs.deviantart.com/


Return to “AI Creations”

Who is online

Users browsing this forum: arrdub and 0 guests