Your AI-enhanced images (tutorial with your pictures)

[who did this ?]

I am no expert, but here you can post pictures, your idea what you want to see and i will try to do it as a tutorial.
Good idea ?

Here are the links i posted earlier:

https://www.reddit.com/r/StableDiffusio ... _analysis/
https://moritz.pm/posts/parameters
https://photogpedia.com/stable-diffusio ... nd-tricks/

I am using the NMKD Gui for stable diffusion, it's really one of the best tools i found in a long time :
https://nmkd.itch.io/t2i-gui

Important : Stable Diffusion 2.0 is out, it can generate better pictures, but at the moment it lacks NSFW, famous people and art styles of known artists. Also you need much more GPU-RAM. I'll stick with the older version until there are community based enhancements...

[xrattletrapx]

Hi Who. It's a fantastic idea. I've thought of a few potential scenarios.

This is something I made a few years ago (I'll edit this post to show the image later).

https://www.deviantart.com/xrattletrapx ... -854894847

I was wondering how you were able to isolate changes. So if I wanted this character to be, say, Jessica Alba in that outfit, how would you proceed?

[bobvan]

I am no expert, but here you can post pictures, your idea what you want to see and i will try to do it as a tutorial.
Good idea ?

Of course

[who did this ?]

This was xrattletrapx original picture.


I used the Masked Inpainting feature to cover the character :



This were my settings. You should read the three links in the first post to understand the importance of the keywords.
Most important thing is the Initialization Image Strength. In the thread were i posted some of my own pictures i kept it high. Higher means, stick more to the original (0.9 and upwards).lower values allow stable diffusion to make up more things that are not in the original material. This works to 0.5.
If you go lower than 0.5 (which can be useful), the original image is just a rough guide of what to create.
And you need to stay close to the image ratio of the original image, otherwise everything looks squeezed.



As you notice, some parts changed dramatic, some parts not at all.So you could go on and describe the boots and dress in detail, but i think you get an idea.

[who did this ?]

How about someone makes a photo of his/her desk and a photo of a celeb that should end up on this desk ?
I tried it with some crappy 2 min collages as a base and it worked.

[Granelex]

What about dynamic shots from Second Life? I was thinking of using my character to pose shots then use stable infusion to make it better. For instance how would you enhance this shot of an alien craft shrinking Vanessa down?

[who did this ?]

What about dynamic shots from Second Life? I was thinking of using my character to pose shots then use stable infusion to make it better. For instance how would you enhance this shot of an alien craft shrinking Vanessa down?

I failed ..



Maybe we should use some higher quality sources. The green light beam and the low resolution were too difficult for SD to work with. I could use some other programs to enhace your source, but we should try other source images first.

[Granelex]

No you did not fail I provided a difficult image. Here this one should be better.

likely second life doesnt make it easy. Perhaps it may be easier with other images like this.

[bobvan]

Pardon my ignorance what is NSFW?

[chocolatejr9]

Pardon my ignorance what is NSFW?

Not Safe For Work. Think stuff like nudes and sex.

[bobvan]

Pardon my ignorance what is NSFW?

Not Safe For Work. Think stuff like nudes and sex.

TY

[bobvan]

Im curious are you now able to build scenarios in DS basic like a beginner with no lighting textures ect to have a template you can instantly render then spruce it up with AI?

Edit Size related images is not the only wacky thing trying to use AI

https://www.youtube.com/watch?v=TVMYyOG-Cs8

[who did this ?]

Im curious are you now able to build scenarios in DS basic like a beginner with no lighting textures ect to have a template you can instantly render then spruce it up with AI?

Edit Size related images is not the only wacky thing trying to use AI

https://www.youtube.com/watch?v=TVMYyOG-Cs8

I just downloaded the latest version (1.8) and there are some major changes, including support for AMD. Its still Stable diffusion 1.5, but it seems to generate much better images by default.The mask feature can now understand a text prompt, what to mask and other things.
So i need to play to play with DAZ and NMKD Gui to what can be done with this version.




No mask, just the keywords

[bobvan]

Im curious are you now able to build scenarios in DS basic like a beginner with no lighting textures ect to have a template you can instantly render then spruce it up with AI?

Edit Size related images is not the only wacky thing trying to use AI

https://www.youtube.com/watch?v=TVMYyOG-Cs8

I just downloaded the latest version (1.8) and there are some major changes, including support for AMD. Its still Stable diffusion 1.5, but it seems to generate much better images by default.The mask feature can now understand a text prompt, what to mask and other things.
So i need to play to play with DAZ and NMKD Gui to what can be done with this version.




No mask, just the keywords

It looks more like Paul Stanley

Looking forward to what you come up with.

I'm sure it's going to keep evolving..

[Granelex]

I still like it. One foot seems deformed but it defintiely shows the potential when adding to images already created

[who did this ?]

Im curious are you now able to build scenarios in DS basic like a beginner with no lighting textures ect to have a template you can instantly render then spruce it up with AI?

Edit Size related images is not the only wacky thing trying to use AI

https://www.youtube.com/watch?v=TVMYyOG-Cs8

This wouldn't work. You would get clay people in a clay world. But you can NMKD as the best source image generator for Facegen. I used the base G8F to create new face by just describing the new face, based on the G8F.

[bobvan]

Im curious are you now able to build scenarios in DS basic like a beginner with no lighting textures ect to have a template you can instantly render then spruce it up with AI?
v=TVMYyOG-Cs8

This wouldn't work. You would get clay people in a clay world. But you can NMKD as the best source image generator for Facegen. I used the base G8F to create new face by just describing the new face, based on the G8F.

Its got be be KD
Seriously What's NMKD?

Its sounds like you more or less use G8 with AI VS someone's pic to create a custom face

Is that how you AI'd Sally?

I want to see if anyone else starts "getting" it..

[who did this ?]

I used an older Sally pic i posted last year and NMKD.

[bobvan]

Got it. You got decent at it.

[Prof Sai]

Does Diffusion Bee just not work or something? I can't get it to do anything useful. Or maybe the AI doesn't like me.

[bobvan]

Does Diffusion Bee just not work or something? I can't get it to do anything useful. Or maybe the AI doesn't like me.

Does not sound like it's only you. I think @foreverlurk is also having a time of it. Which why I am waiting to see if others have any luck.

[who did this ?]

Does Diffusion Bee just not work or something? I can't get it to do anything useful. Or maybe the AI doesn't like me.

You need to tell Diffusion Bee the things you don`t want to see as well :
Put the things you don't want in [square brackets] or "A bowl of apples:1 red:-1" = a bowl of apples, no red apples.

Sometimes you have to use more words to describe what you dont want to see to get something you wanted in the first place.
I have no idea how often Diffusion Bee gets updates, but as always, you should use the latest version and look at their forums for examples on prompts. If there is a text box for negative prompts, that is were you should spend the same attention to as in the box for the things you want to see in first place.

[Prof Sai]

Hmmm...

[who did this ?]

Hmmm...

As i said, you have to check the official forums to find out how use negative prompts.Have you tried,",no visible windows" ?

[Little Sally]

Hmmm...

As i said, you have to check the official forums to find out how use negative prompts.Have you tried,",no visible windows" ?

It looks like you can create some interesting backgrounds with it (to insert either rendered characters or real people).

I might look into that.