How does your avatar look when enhanced by AI?

[Istelathis]

This one was fun, I played around with a lot of prompts.

Original:

ControlNet (boosted brightness I'm not sure if that helps or not)

 

Picture produced:

I'm not sure why it came out so bright.

Prompt: beautiful lady, hyperdetailed photography, realistic, photo-realistic,(8k, RAW photo, best quality:1.3), blue pants, black top, red hair

 

Prompt: beautiful lady, anime coloring, anime screencap

 

--------------------------------------------

Both used the below settings, including seed.

Seed: 3076121221, Dimensions: 512x512, Sampler: euler_a, Inference Steps: 25, Guidance Scale: 7.5, Model: cyberrealistic_v33, Prompt Strength: 0.8, Preserve Color Profile: true, ControlNet Model: control_v11p_sd15_canny

Images upscaled at AnyRec

[Liv Simondsen]

is there a tutorial on how to edit images ? because i know how to generate them, but i don't know how could i edit a SL photo. Thanks in advance!

[Bagnu]

I trained a model of my Sandor AV, and this is what AI came up with. I'm finding it to be like commissioning artwork. I give the idea, and someone else interprets it in their own vision. It's great for getting ideas, but it's not a photo filter.

I uploaded 38 images of my AV to Gencraft, rotated around my (SL) Z axis to train the model.

Interestingly enough, AI added a beard. Which my AV originally had, but I removed, because Caitlin preferred my AV clean shaven.  AI was thinking similarly to me.

Edited December 2, 2023 by Bagnu

[Scylla Rhiadra]

22 minutes ago, Bagnu said:

I'm finding it to be like commissioning artwork.

Except that no one is receiving a "commission," i.e., getting paid for their work, other than possibly the engineers and shareholders at Gencraft (and Midjourney, etc.). Certainly, the working artists whose work is being plundered to "train" and then generate these images aren't.

Obviously, that's not such a large issue when you're simply producing work for your own entertainment rather than profit, but there is still an element of complicity involved in using this kind of generative AI at all, because you are in effect assisting in "training" it. And there are lots of other ethical issues beyond violation of copyright involved as well.

I just mention this in passing -- I'm not going to get into a debate about it because, honestly, I don't want to spoil people's fun -- but it is important to at least be aware, if not acknowledge, the social costs of this kind of thing. And there are some AI programmes that are more ethical than others.

https://www.datacamp.com/tutorial/ethics-in-generative-ai

Edited December 2, 2023 by Scylla Rhiadra

[Bagnu]

40 minutes ago, Scylla Rhiadra said:

Except that no one is receiving a "commission," i.e., getting paid for their work, other than possibly the engineers and shareholders at Gencraft (and Midjourney, etc.). Certainly, the working artists whose work is being plundered to "train" and then generate these images aren't.

Obviously, that's not such a large issue when you're simply producing work for your own entertainment rather than profit, but there is still an element of complicity involved in using this kind of generative AI at all, because you are in effect assisting in "training" it. And there are lots of other ethical issues beyond violation of copyright involved as well.

I just mention this in passing -- I'm not going to get into a debate about it because, honestly, I don't want to spoil people's fun -- but it is important to at least be aware, if not acknowledge, the social costs of this kind of thing. And there are some AI programmes that are more ethical than others.

https://www.datacamp.com/tutorial/ethics-in-generative-ai

I DO understand what you are talking about, and I have many concerns as well. AI is here though, and it won't go away. Only by experimenting with it , and seeing it's shortcomings, can we actually understand the risks better. 

I can see a huge risk to the entire entertainment industry. One single person could create an entire movie without the need for any actors or anyone else involved. Not even a script writer. It's not quite there yet, but it will be. Same with music. No band needed. No musical skills necessary. Same again with art. The list is endless. Just by writing this, my concerns grow deeper.

The scariest thought of all, is that humanity will become dependant on AI. I can actually see this happening.

Edited December 2, 2023 by Bagnu

[Sid Nagy]

Artists have always inspired themselves and each other by looking and learning from each others work, haven't they?
So that AI learns from other artists is kind of a natural process IMHO.
AI will simply become a competitor for all arts, whether it is painting, music, poetry, fashion, film, photography, literature, architecture  .....

A new era, we have to adapt to that.
Photography did not ruin painting, but changed it.
Photographers weren't pleased with digital photography and digital processing at first either.
Now almost everyone embraces these techniques and uses them.

AI will not ruin human arts, but it will change it too.  Life evolves, we will have to learn to use the new situation. AI is one of them.

 

Edited December 2, 2023 by Sid Nagy
Added a few. Changed bits. I need an AI ghostwriter. 😋

[Scylla Rhiadra]

6 minutes ago, Sid Nagy said:

Artists have always inspired themselves and each other by looking and learning from each others work, haven't they?
So that AI learns from other artists is kind of a natural process IMHO.

Sure, but there's an enormous difference between "being inspired" by or "learning" from a work, and literally ripping parts of it for one's own use. If someone ripped textures or parts of one of your in-world creations for use in their own product for sale, you'd not be pleased, I'd imagine?

There's also (as I have to keep reminding my students) a difference between reference, allusion, citation, or quotation, and plagiarism. I reference or allude to a painting by Andrew Wyeth in one of my own images, but the point of that reference is that that reference not merely acknowledges, but is intended to draw attention to Wyeth's work: part of the "meaning" of my own pic is to be found in a comparison with Wyeth's.

[Sid Nagy]

It is a fascinating discussion already. But I don't think this is the right thread for it, maybe not even the right forum.
Although SL and its users will have to deal with AI as well.

Edited December 2, 2023 by Sid Nagy
It is Saturday, so I have to edit.

[Scylla Rhiadra]

1 minute ago, Sid Nagy said:

It is a fascination discussion already. But I don't think this is the right thread for it, maybe not even the right forum.
Although SL and its users will have to deal with AI as well.

I think it's a discussion we need to have in the context of SL, because AI is going to become more and more important here in lots of spheres. It's going to be used by creators to produce mesh and texturing for those products, as well as other things, such as landscape. In terms of the arts community, there are already galleries, including a number of major ones, that as policy do not accept AI-generated art.

However, I'll agree that this is not the thread for it. I just thought there should be an acknowledgement here of the issue.

[Bagnu]

Gasp, AI is almost as bad as when stockings were replaced by pantyhose!!!

[Marigold Devin]

On 11/16/2023 at 5:52 PM, Bagnu said:

Now here's @Cinnamon Mistwoodenhanced by AI. The first pic is the reference image, which is a crop of a pic I posted earlier when she was together with me and Caitlin. I used the same "prompts" as for Caitlin, except I typed in " red hair".

 

 

 

 

I am imagining her typing with that left hand (gorgeous face though)

[Bagnu]

16 minutes ago, Marigold Devin said:

I am imagining her typing with that left hand (gorgeous face though)

Thanks!!! But AI made her face. I can't take credit.

LOL, six fingers would make it faster. I wish I could type with more than two!!!

Edited December 2, 2023 by Bagnu

[Cristiano Midnight]

[Istelathis]

On 12/2/2023 at 1:32 PM, Scylla Rhiadra said:

Sure, but there's an enormous difference between "being inspired" by or "learning" from a work, and literally ripping parts of it for one's own use. If someone ripped textures or parts of one of your in-world creations for use in their own product for sale, you'd not be pleased, I'd imagine?

This is an interesting topic, but it does worry me.  Not AI generated imagery, mind you, but people starting to sue and fine one another and ultimately leading to stifled creativity as it becomes part of our culture to actively seek out artwork in an attempt to score some of that sweet royalty money.

It reminds me of youtube, once a place where we could share music, clips from movies, shows, whatever.  Now there are people that will actively try to find videos, that they can capitalize on and issue out legal notices to remove videos if royalties are not paid.  It becomes this dystopic capitalistic nightmare, where we are all cornered into pre determined types of behavior, or risk legal woes.  See a pattern that looks close to something you have created?  Sue them!  A voice that sound similar to your own, sue them!  

I really hope this is not the path humanity takes, where a few people dictate the creativity of others, and we no longer have the ability to stand on the shoulders of others, where innovation becomes stifled, ya know?

 

Anyway, I was playing around again with diffusion and thought some of this might be of interest:

Here is the original image and Control/Net

The ControlNet produces this image, using in this instance, Cranny:

I'm not entirely sure how it works, but it appears that diffusion will use the above image, and the original image to produce the result.  You can create your own ControlNet image to get better results, diffusion, it seems, will fill it in for you.  You set the value of various weights, and I see it as more so trying to fill within the lines 🤣 Those settings, the weights, they will let the software know how much it can deviate from it.

I imagine, if I wanted better eyes I could modify the ControlNet to produce better looking eyes.  I could also inpaint the eyes, and keep using different seeds to hope for better results.

I wanted to try to capture my avatar in a strange pose, so I used a poser to move my arms and legs around in unnatural ways, to see just what the result would be.  Action shots were discussed earlier, and I wanted to see what I could get accomplished in pictures you would not normally see others drawing in, and to show how flexible it all is.  Easy Diffusion (and others) give you a lot of flexibility, to produce what you want, using various weights, images, controlnets, and so on.  You can see it attempts to maintain the controlnet, fill in the lines, and so on.

 

Here is another one, playing around with prompts and changing the guidance scale.

No pants, but no vulgarity either.  You can see how it tried to fill in the foot with a flesh like material.  Overall, it tried to stay in the lines .  If I wanted to, I could go in with GIMP and fix that, I could draw some pants on it, make the bottom of a foot look more like a sole of a shoe, fix the left leg coloring,  which would make for a better looking image.  Right now it could be interpreted as the stuff of nightmares, but with a little tweaking it could look so much better.  

Adding new prompts ( red pants, black sole) with the same settings and seed

There is a lot of flexibility here, and I see it more of a tool for people.  

Edited December 3, 2023 by Istelathis

[Istelathis]

Playing around a bit more with the top image, I wanted to give it a theme.  The original animation had no intention behind it but to test out action shots, I thought she looked as though she were falling so I added a few prompts to give a suitable background.

I need to be more careful in my spacecraft!

[Bagnu]

3 hours ago, Istelathis said:

This is an interesting topic, but it does worry me.  Not AI generated imagery, mind you, but people starting to sue and fine one another and ultimately leading to stifled creativity as it becomes part of our culture to actively seek out artwork in an attempt to score some of that sweet royalty money.

It reminds me of youtube, once a place where we could share music, clips from movies, shows, whatever.  Now there are people that will actively try to find videos, that they can capitalize on and issue out legal notices to remove videos if royalties are not paid.  It becomes this dystopic capitalistic nightmare, where we are all cornered into pre determined types of behavior, or risk legal woes.  See a pattern that looks close to something you have created?  Sue them!  A voice that sound similar to your own, sue them!  

 

AI is scouring Youtube all the time. I made an anniversary video for Caitlin, and it was blocked for copyright infringement outside of Canada. The only similarity is with the falling leaves to my falling rose petals. The movie is a children's movie. My video is not. I disputed it, and the company dropped the copyright claim. But I shouldn't have had to do that. I posted both mine, and the trailer to my alleged copyright infringement. So yes, it's a worry what AI can come up with.

 

 

Edited December 3, 2023 by Bagnu

[Cristiano Midnight]

What becomes of the brokenhearted?

 

[Jasmine Hexem]

These are mine

[Istelathis]

Original

Boop the nose!

Anime face did not want to pop 

 

Original

Trying for more anime in above

 

 

It is fun, but I do like the SL versions better.  So far, I haven't had much luck with this avatar when trying to produce AI images.  

Edited December 5, 2023 by Istelathis

[Zeta Vandyke]

Using trained Stable Diffusion models of my avie to do an image to image creation of my SL portrait, and the outdoor one just using a prompt with the model and no img source.

[Louise Fyrewik]

My picture from SL:

My render from Leonardo.ai:

[Istelathis]

Checking out screenshots using the latest official viewer, the screen appears more clear to me but it could just be a virtual placebo I'm experiencing 🤣

Original and ControlNet

Realistic

Watercolor

Pencil (upscaling made it look too smooth)

 

Ink

 

Comic Book

 

Street Art

Lastly Cartoon

All done in easy diffusion, the first couple were upscaled with AnyRec - after that the upscaler would make the images to smooth looking, which did not work well with the prompts used to create the images in the first place.

No Lora used, just the controlnet and the CyberRealistic model.  I had to change the guidance scale for the more artistic ones, I believe that is why you will see subtle shifts such as where the eyes are facing.  From my understanding, it should have for the most part, kept in line with the below image:

 

 

Edited December 6, 2023 by Istelathis

[Cristiano Midnight]

noblesse oblige

 

[Istelathis]

Original/ControlNet

Street Art 🤗

Realistic:

 

Watercolor

Edited December 7, 2023 by Istelathis

[Love Zhaoying]

On 12/5/2023 at 12:13 PM, Istelathis said:

Original

Boop the nose!

Anime face did not want to pop 

 

Original

Trying for more anime in above

 

 

It is fun, but I do like the SL versions better.  So far, I haven't had much luck with this avatar when trying to produce AI images.  

A couple are very reminiscent of Full Metal Alchemist!