How does your avatar look when enhanced by AI?

[Danielle Atheria]

Inspired by Quartz Mole. Thank you hon for the suggestion.

I just actually made this about an hour ago after a friend shared a link.  If you scroll after this you'll see the original in world pic.

This is my first attempt

Original 

 

Edited November 6, 2023 by Danielle Atheria

[UnilWay SpiritWeaver]

Here:

 

Original:

I do a lot of work with Stable Diffusion actually.

This one was more automated than my usual. I usually take images back and forth between Stable Diffusion and Gimp (an open source replacement to Photoshop).

Instead... 50 passes through img2img in Stable Diffusion because it didn't understand that top and leotard she's wearing and kept giving me chubbier and chubbier results that I had to tone back with creative prompts and some of my LoRa presets (because the moment I put in furry it tried to convert her and the trees into a bare chested furry boys and I needed to some tricks to get around that).

Edited November 6, 2023 by UnilWay SpiritWeaver

[UnilWay SpiritWeaver]

Here's another one with my main. Managed to find an image with clothes on.

Original:

[UnilWay SpiritWeaver]

Here's my main again. The same image, but enhanced with a tool called 'outpaint' that expands the image based on prompts and existing content.

I did this to get a 4x3 - Going to bring it into SL as a picture in my home.

Outpaint gets a little wild on the edges at times - obviously log homes don't fade out like that on the left, but the effect looks cool.

This was actually a really cool idea to take an SL avatar and run it through AI tools. I'm going to have to keep this one.

 

Edited November 6, 2023 by UnilWay SpiritWeaver

[Marigold Devin]

I haven't really got to grips with this idea yet, but can see why, to someone on another thread, we all look like "monsters" to them in Second Life! There is such a difference between what we can actually create in Second Life and what artificial intelligence can do for us.

As I am not able to spend much time on this currently, must to my frustration, I described my Second Life avatar into AI software, and it came up with quite a nice version of "her". I asked for blonde hair with pink tips, it gave me the right blonde hair and put the pink tips on the trees.

The AI version of my avatar is certainly a lot easier on the eye.

 

Edited November 6, 2023 by Marigold Devin

[MrsSeren]

I prefer the original.  

[UnilWay SpiritWeaver]

16 hours ago, Danielle Atheria said:

Inspired by Quartz Mole.

I missed Quartz's note. Did Quartz have a 'how to get started in this' guide?

If not...:

Slight tangent for a second but this is a guide to installing AI art tools on your own computer, and where to get training resources for them. It's about a year old but still points to the best sources and goes through some complex things (like installing python and running a git command to download something) in very simple step by step terms. As long as you follow the steps exactly and don't rush ahead, it works:

The training models site can be used to get everything from hyper-real styles, oil painting, anime, etc. It does have a comically large amount of anime like content, but the other things are there as well...

The tool in stable diffusion for making an AI image of an SL image is 'img2img' - you put your SL image into a window on one side, put in some prompts to customize, load up some 'training models' to filter the style, and have it do the work.

I imagine there are online website to do this also.

 

Edited November 6, 2023 by UnilWay SpiritWeaver

[Istelathis]

Using Easy Diffusion

Initial Image (img2img):

I did not use any prompts, other than the image modifiers you can select from within the interface.

Seed: 0

Dimensions: 968x512

Sampler: dpmpp_2m

Inference Steps: 25

Guidance Scale: 7.5

Model: cyberrealistic_v33

Prompt Strength: 0.8

Preserve Color Profile: true

=====================================================

Image Modifier: 8-bit

 

Image Modifier: Anime

 

Image Modifier: Photo

 

Image Modifier: Realistic

 

Image Modifier: Watercolor

 

 

I'm not sure how to do a better job of img2img, I found that setting the seed to 0 was necessary or else I would get something completely different, I did not want to add a prompt so I intentionally left it blank.  I'll have to do a bit more research on how to perform a better job with easy diffusion later on, I have had better luck at changing pictures using online services but don't like the limitations they make, I don't want to spend extra money when I can hopefully just learn how to do it locally on my computer for free.

[UnilWay SpiritWeaver]

8 minutes ago, Istelathis said:

Using Easy Diffusion

I'm not sure how to do a better job of img2img, 

Nah those are insanely cool. I especially like the 'photo' and 'watercolor' ones.

 

Edited November 6, 2023 by UnilWay SpiritWeaver

[Love Zhaoying]

28 minutes ago, Istelathis said:

Image Modifier: Photo

 

Image Modifier: Realistic

Interesting to me that "photo" looks more realistic (to me) than "realistic"!

Is that a "mountain lion" / "cougar"?

Or a lion cub?

Asking for myself!
 

 

[Istelathis]

37 minutes ago, Love Zhaoying said:

Interesting to me that "photo" looks more realistic (to me) than "realistic"!

Is that a "mountain lion" / "cougar"?

Or a lion cub?

Asking for myself!
 

 

I think it is a puma, it surprised me when it showed up though 🤣  I thought you might notice that one.

I decided to see what would happen if I changed the model to the default, using realistic and keeping all of the other settings the same and this is what it created:

Model: sd-v1-4

Image Modifier: Realistic

 

I don't like the model that easy diffusion comes with by default it makes pretty scary images 😭😨🤣

 

Using the model deliberate_v3 with the same settings just created a room, which looked nice but had no avatar in it.  I switched the model to  dreamshaper_8 and this is what I get for Realistic:

 

I have a bunch of other models, but the HDD I have them stored on is slow as molasses and it takes forever for them to load into memory so I can't try out too many of them for now, I will goof with a few other settings though and see what happens.

Changing the Guidance Scale from 7.5 to 1.1 resulted in this:

I'm not sure what that creature is, I think it is friendly though.

Changing the prompt strength in Easy Diffusion seems to be pretty important:

 

Bleh, so many things to fiddle around with.  I could spend hours trying to figure it out, and I most likely will.  I'll do one more before going outside and touching some grass 🤣

 

Seed: 0, Dimensions: 968x512, Sampler: dpmpp_2m, Inference Steps: 25, Guidance Scale: 1.1, Model: dreamshaper_8, Prompt Strength: 0.5, Preserve Color Profile: true
Processed 1 images in 1 minute 13 seconds

Edited November 6, 2023 by Istelathis

[Love Zhaoying]

6 minutes ago, Istelathis said:

I think it is a puma

Cool! Apparently (I had to refresh myself via Google), the "mountain lion" and "cougar" and "panther" are all the same critter.

My school teams until college were all "Cougars".. 🙂

So, it just puts random things in the background alongside you (according to the template, I suppose)!

[Love Zhaoying]

14 minutes ago, Istelathis said:

 

I don't like the model that easy diffusion comes with by default it makes pretty scary images 😭😨🤣

Picasso would be jealous!

[SpiritSparrow Skydancer]

before and after just using simple generative fill AI in photoshop.  Keep in mind, this is just me messing around! I am not sure what y'all are using for the other photos, which are absolutely stunning!!!

 

 

 

[ValKalAstra]

Oh, me, me! I've been dabbling in AI stuff too. My first tries were with Midjourney more or less a year ago and when I could no longer afford that - I stumbled accross Stable Diffusion. That has been my go-to and I have spent a lot of time experimenting and playing with it. I've since experimented a lot in trying to bring my looks over into other concepts, such as drawings, paintings etc. So it's not going to be a direct Image 2 Image but rather something created with probably the... 8th or 9th generation of LORA and custom blend models.

Either way, some of my experiments with it. As comparison my Avi from SL first.

In case someone warrants it, my signature stare in custom sticker form - I need to create more of these!

A shot of Anime Nina and vibrant water colour!

Nina has played too much Eldenring!

And Nina got style - at least in the eyes of the all spooky AI.

[ValKalAstra]

1 hour ago, Istelathis said:

I think it is a puma, it surprised me when it showed up though 🤣  I thought you might notice that one.

I decided to see what would happen if I changed the model to the default, using realistic and keeping all of the other settings the same and this is what it created:

Model: sd-v1-4

Image Modifier: Realistic

 

I don't like the model that easy diffusion comes with by default it makes pretty scary images 😭😨🤣

 

Using the model deliberate_v3 with the same settings just created a room, which looked nice but had no avatar in it.  I switched the model to  dreamshaper_8 and this is what I get for Realistic:

 

I have a bunch of other models, but the HDD I have them stored on is slow as molasses and it takes forever for them to load into memory so I can't try out too many of them for now, I will goof with a few other settings though and see what happens.

Changing the Guidance Scale from 7.5 to 1.1 resulted in this:

I'm not sure what that creature is, I think it is friendly though.

Changing the prompt strength in Easy Diffusion seems to be pretty important:

 

Bleh, so many things to fiddle around with.  I could spend hours trying to figure it out, and I most likely will.  I'll do one more before going outside and touching some grass 🤣

 

Seed: 0, Dimensions: 968x512, Sampler: dpmpp_2m, Inference Steps: 25, Guidance Scale: 1.1, Model: dreamshaper_8, Prompt Strength: 0.5, Preserve Color Profile: true
Processed 1 images in 1 minute 13 seconds

These are cool though. Did you try using a Canny and Depth Control.net, perhaps? Although seeing the inference times, it might be a bit taxing on the system? I haven't checked EasyDiffusion, I'm more of an automatic1111 girl.

And yeah, totally - the sd 1.4 model is absolutely ancient at this point, basically to the point of being no longer useful. Deliberate is a good general purpose model, dreamshaper works well for fantasy images. Another decent model is RevAnimated - although you absolutely WILL want to add nude to the negative prompts there. It's thirsty - but produces great clothed images too. Just needs convincing.

 

[Orwar]

16 minutes ago, ValKalAstra said:

Nina has played too much Eldenring!

   Dang it, are you the one who's been running around killing my minions? Harrumphs.

[Janet Voxel]

[SpiritSparrow Skydancer]

3 minutes ago, Janet Voxel said:

That is outstanding!! wow

[Marigold Devin]

1 hour ago, ValKalAstra said:

 

 

 

This one is my favourite. Those eyes are like liquid, and just the combination of the colours is causing dopamine to be released in my brain.

[Marigold Devin]

12 minutes ago, Janet Voxel said:

Very "Bridgerton" and I agree with Spirit wholeheartedly. Very hard to take your eyes away from such a stunning work of art.

[Istelathis]

1 hour ago, ValKalAstra said:

These are cool though. Did you try using a Canny and Depth Control.net, perhaps? Although seeing the inference times, it might be a bit taxing on the system? I

I've tried them, but my laptop is pretty old.  I get memory errors when I give it a shot unfortunately.  Eventually I will upgrade my computer and have a bit more wiggle room to play around with it all.  

[JeromFranzic]

Original...

 

AI... Created from a random tungsten.run AI filter.
 

Edited November 7, 2023 by JeromFranzic

[SpiritSparrow Skydancer]

21 hours ago, UnilWay SpiritWeaver said:

Here:

 

Original:

I do a lot of work with Stable Diffusion actually.

This one was more automated than my usual. I usually take images back and forth between Stable Diffusion and Gimp (an open source replacement to Photoshop).

Instead... 50 passes through img2img in Stable Diffusion because it didn't understand that top and leotard she's wearing and kept giving me chubbier and chubbier results that I had to tone back with creative prompts and some of my LoRa presets (because the moment I put in furry it tried to convert her and the trees into a bare chested furry boys and I needed to some tricks to get around that).

I am still trying to understand img2img   I get some really weird results...

[SpiritSparrow Skydancer]

So I did that img2img and I got these are my results ,  I used nightmare shaper which makes no sense but there it is lol