How does your avatar look when enhanced by AI?

[Riley Brinner]

[KenzieFoxx]

These are really cool!

[Gopi Passiflora]

Started playing with sdxl model found on Replicate.

Original Picture:

 

Two attempts: (Prompt: "full body picture of a woman with very long hair")

 

I'm not totally satisfied with the results but I will continue playing around with the tool.

Edited November 17, 2023 by Gopi Passiflora
Added prompt for img2img

[Cristiano Midnight]

[mika Mesmeriser]

[Istelathis]

I've been playing a lot with easy diffusion today, goofing around with settings.  I initially was loading an image at lower resolution believing it would speed up the process and be less of a strain on my video card but that assumption was completely incorrect.  At 768x400 for the initial image and control net, the AI rendered images would be far less quality, the faces would be really horrible and I couldn't get it right at all.

Initial image, and controlnet (system avatar)

Dimensions: 512x512, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 5, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

I did not like the background of the above one, I added a new prompt for the next one which was supposed to be a line drawing but it did not quite follow through as expected.

Dimensions: 512x512, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 5, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

Second shot, using higher resolution image this time for initial image and control net.

Below is using a LORA

Dimensions: 768x400, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 1.1, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Lora Model: detailSliderALT2, Lora Strength: 0.24, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

This picture below is not using a Lora

Dimensions: 768x400, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 1.1, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

Edited November 17, 2023 by Istelathis

[ValKalAstra]

21 hours ago, Janet Voxel said:

I think the confusion comes from progress that has been made with AI. It has been with us for a while. Before you’d use AI to “Enhance” your picture. Really what it did was run your picture through a series of filters and voilà….you had an AI enhanced photo.

Over the last couple years the advancement is really, that it’s doing its own thing. The better ones allow you to put how much it’ll actually alter your photo. The real magic is you can just type something, a few sentences and you’ll get a fully rendered drawing, painting or photo.

I’m not a techie person, so I may not have explained it properly. I think the misunderstanding is that it’s supposed to enhance what you put in there as a reference. It’s really taking what you put in there and making something new.

Personally, I enjoy playing around with it.

I'm the resident techie adjacent gal and I give this my seal of approval. Good explanation!

 

27 minutes ago, Istelathis said:

I've been playing a lot with easy diffusion today, goofing around with settings.  I initially was loading an image at lower resolution believing it would speed up the process and be less of a strain on my video card but that assumption was completely incorrect.  At 768x400 for the initial image and control net, the AI rendered images would be far less quality, the faces would be really horrible and I couldn't get it right at all.

Initial image, and controlnet (default SL avatar)

Dimensions: 512x512, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 5, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

I did not like the background of the above one, I added a new prompt for the next one which was supposed to be a line drawing but it did not quite follow through as expected.

Dimensions: 512x512, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 5, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

Second shot, using higher resolution image this time for initial image and control net.

Below is using a LORA

Dimensions: 768x400, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 1.1, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Lora Model: detailSliderALT2, Lora Strength: 0.24, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

This picture below is not using a Lora

Dimensions: 768x400, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 1.1, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

Woah - you've made leaps and bounds with these already! Yah, having the correct resolution is very important. SD 1.5 works best in the range of 512-768. Deviate too strongly and you're going to lose a considerable amount of quality and cohesion. The crux with that is that you really want a higher resolution for things like faces and hands.

There's a little trick you can do and that is head on over to inpaint, set it at a low denoise of give or take 0.4, switch to "only masked", mask out the face and then run the same prompt at the highest resolution you can. This will give that small area more resolution to work with (and for those that would like to automate that: adetailer extension does this for you under a1111).

[rasterscan]

Incredible. The power of deceptional AI enhancement. Great eye candy though.

[BJoyful]

20 hours ago, Orwar said:

The day we can prompt an AI to 'mesh, rig, and texture some fancy clothes for SL', I'll be opening a store selling that sort of stuff!

That has already happened... ChatGPT has been used to create mesh in SL and its really good... you do need to know how to prompt it correctly, though.

[Sabrina Nebula]

How do you guys do that? All I can do is describe what I want to have as a picture but I can't find something where I can put an image of my Avatar in and say "Make it look real" (for free if possible ^^).

I made pics but that's not what I'm looking for, she's looking gorgeous tho:

 

Edit: I think I found a way. The results are disturbing but I'll get there one day haha

Edited November 17, 2023 by Sabrina Nebula

[Istelathis]

2 hours ago, ValKalAstra said:

There's a little trick you can do and that is head on over to inpaint, set it at a low denoise of give or take 0.4, switch to "only masked", mask out the face and then run the same prompt at the highest resolution you can. This will give that small area more resolution to work with (and for those that would like to automate that: adetailer extension does this for you under a1111).

Thank you for the kind words, and advice.  My poor graphics card throws out error messages even when only inpaint is used to produce higher resolution, the card on this laptop only has 4 gigs on it.  I think it might also have to be due to using easy diffusion, the settings are not as configurable as stable diffusion or a1111.  I actually downloaded JuggernautXL for a model because it looked really impressive for making more realistic pictures and it just about killed my laptop 🤣 Trying to load a 6gb model was probably not the best of ideas I had today.

I did manage to get it to work at producing a slightly higher resolution, using inpaint and I was impressed with some of the changes, I'll have to play around with it a bit more to see if I can get better results.  The face was inpainted, but there are no advanced settings by how much I can change it by.  There are preset icons I can click on for the options of opacity and sharpness.

When zoomed in, it looks like the nose is broken though, lol.  It was less noticeable before I upscaled it.

Hopefully, I will get a newer laptop in a few months, if I do,  I'll try using just stable diffusion as I hope to have become more familiar with the generating AI art.  I think easy diffusion is great, for starting with to become familiar with everything, and it is capable but from what I have seen not nearly as configurable.  Regardless, it is a lot of fun learning how this all works, and trying to get this old laptop to produce all that it can.  

Edited November 17, 2023 by Istelathis

[Ceka Cianci]

I can't even figure out what changed on artbreeder.. I can't even find where to upload my image to get started so I can do any of this fun stuff.. hehehe

 

ETA: Always two seconds after I'm ready to give up, things land right in my lap.. Found it!! \o/

My advise in this world is, don't look for things, just stumble around until you trip over them.. lol

Edited November 18, 2023 by Ceka Cianci

[AmeliaJ08]

On 11/16/2023 at 10:29 PM, Orwar said:

   The day we can prompt an AI to 'mesh, rig, and texture some fancy clothes for SL', I'll be opening a store selling that sort of stuff!

Yep. It's undoubtedly coming too.

I'm sure it already exists in some form, I'd genuinely love to see what this type of technology is capable of when it comes to generating 3D art.

edit: saw someone up there mention ChatGPT as having done this, would be curious to see it. If you look at a collada file in a text editor you will see how possible this is, it's laid out in a surprisingly meatbag-readable way, there's a lot of it for anything truly complex but to a language AI like ChatGPT that can already write code (kinda) there's no reason it couldn't string together a file like this.

 

 

Edited November 18, 2023 by AmeliaJ08

[Danielle Atheria]

Pic i took last year redone

 

[Cristiano Midnight]

A sky full of stars

 

[Istelathis]

 

Initial Image/Controlnet

 

Photograph

 

Watercolor

 

Anime

Line Art (eye goofed on this one prompt strength raised to .99)

 

Crosshatch(added "closed eyes" in negative prompt, PS .99)

With the exception of prompts, seeds, or otherwise indicated under each picture the settings were as follows:

Dimensions: 512x512, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 5, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

Upscaling really messes with my pictures, I downloaded another upscaler but run into memory limitations on my video card.

-------------------------EDIT--------------------------

Out of curiosity,  I went to https://stable-diffusion.site/image-to-prompt/ to convert the initial image to a text prompt, which was:

"a beautiful young woman sitting on top of a bed, inspired by Allan Linder, cg society contest winner, at the park, firefly lights, pretty face!!, not cropped"

I changed it to:

"a beautiful woman sitting on top of a bed, inspired by Allan Linder, cg society contest winner, at the park, firefly lights, pretty face!!, not cropped, symmetrical eyes"

 

Using this prompt, in easy diffusion without an initial image or ControlNet, this is what it produced:

Edited November 19, 2023 by Istelathis

[Marigold Devin]

On 11/17/2023 at 1:53 AM, Cristiano Midnight said:

Blimey! All I can think of is how much my mother loved firemen! What a stunning piece of art you are.

[Cristiano Midnight]

[Cristiano Midnight]

6 hours ago, Marigold Devin said:

Blimey! All I can think of is how much my mother loved firemen! What a stunning piece of art you are.

That is very kind of you thank you ❤️

[Ajay McDowwll]

This is the original picture:

And here's the AI variant:

I am not at all good at this lol -- it's entertaining to see what the AI decided to do with my original jewelry.  And I think they took my crow's wing and made it into hair? 

It's cool, though

[Leora Greenwood]

22 hours ago, Istelathis said:

 

Initial Image/Controlnet

 

Photograph

 

Watercolor

 

Anime

Line Art (eye goofed on this one prompt strength raised to .99)

 

Crosshatch(added "closed eyes" in negative prompt, PS .99)

With the exception of prompts, seeds, or otherwise indicated under each picture the settings were as follows:

Dimensions: 512x512, Sampler: dpmpp_2m_sde, Inference Steps: 25, Guidance Scale: 5, Model: cyberrealistic_v33, VAE: vae-ft-mse-840000-ema-pruned, Prompt Strength: 0.6, Preserve Color Profile: false, ControlNet Model: control_v11p_sd15_canny

Upscaling really messes with my pictures, I downloaded another upscaler but run into memory limitations on my video card.

-------------------------EDIT--------------------------

Out of curiosity,  I went to https://stable-diffusion.site/image-to-prompt/ to convert the initial image to a text prompt, which was:

"a beautiful young woman sitting on top of a bed, inspired by Allan Linder, cg society contest winner, at the park, firefly lights, pretty face!!, not cropped"

I changed it to:

"a beautiful woman sitting on top of a bed, inspired by Allan Linder, cg society contest winner, at the park, firefly lights, pretty face!!, not cropped, symmetrical eyes"

 

Using this prompt, in easy diffusion without an initial image or ControlNet, this is what it produced:

So interesting but it also looks quite painful; the AI seems to have fused your two hands into one in at least two of those images.  🙀

[Love Zhaoying]

1 minute ago, Leora Greenwood said:

So interesting but it also looks quite painful; the AI seems to have fused your two hands into one in at least two of those images.  🙀

One of those cases where if the AI was "smart", it would think "hey, lucky me - she's hiding her hands so I don't have to draw them in any generated pictures!" Instead of, "hey, lucky me - a chance to generate more Nightmare Fuel".

[Krystina Ferraris]

On 11/16/2023 at 4:38 PM, Bagnu said:

I did a bit of fiddling . Here's how AI interpreted Caitlin:. The original is first. I didn't make any alterations to the AI generated images. The original was the reference image. I was able to crop the original a bit in the Gencraft online AI I used, so it would pick up her features better, but I can't download that from the website.

I love the 1950s vibe but I am not sure in pic #2, what that lump  between the trapezium and radius of her left hand?

I love the pics though!

[Krystina Ferraris]

On 11/16/2023 at 10:29 PM, Orwar said:

   The day we can prompt an AI to 'mesh, rig, and texture some fancy clothes for SL', I'll be opening a store selling that sort of stuff!

Me too but I guess, by the time this will be available, everyone will have figure out how to do it and opened a store. My granpa used to say "If I become a hatter, men will be born headless". LOL

[Bagnu]

48 minutes ago, Krystina Ferraris said:

I love the 1950s vibe but I am not sure in pic #2, what that lump  between the trapezium and radius of her left hand?

I love the pics though!

Thanks!!! AI seems to have a problem with hands. I didn't fix any of the pics, so the AI errors can be seen. 

Caitlin liked the pics too.