Trying out this AI thing

[Viridian]

It is basically my stupidity in all of this. I am using mage.space, and I can seed an image, but it doesn’t seem to do what I want. So yeah, it would be great to be able to put O-girl, or a Yvonne Craig Batgirl into a quicksand situation, but I’m at a loss as to how to go about it. Just like anything else, this is another example of a new technology that I just don’t understand at all.

I use a much more straightforward process.The tools are much more simple than you might think if you can do a few workarounds. Let's start with a (fairly low-res) O-Girl screen cap.

o.png

Mage.Space has a very useful inpainting function when you upload your reference image. You can manually paint a mask over the sections you want the AI to modify, or you can use the foreground/background buttons to automatically select the section you want to work with.

o_inpaint.png

Now I can add in my prompt. This is pretty complex in itself as I have my preferred blend of secret herbs and spices, including a few LoRAs, but throwing in something along the lines of "swamp" and various other background items, using the Tiki set, then followed with some AI Hi-Res fixes to up the resolution and fix the grainy low-res textures, I get something like this.

o_backgroundswamp.jpg

Alternatively, I could photo-edit the screen cap. Using Photoshop, I paint a crude background.

o_swampdraft.png

Then run it through the same prompt using the background inpainting tool, followed by the Hires Fix tool.

o_swamp.jpg

[Viridian]

However, the huge advantage with having a portfolio of quicksand photographs is that you can use the ControlNet function (also in Mage.Space). When selecting the Pose ControlNet, for example, the source image isn't used as a visual reference, but instead the AI uses it to build a wireframe model, allowing the user to build the image around the pose. Using the same O-Girl reference shot as before, you can swap out the subject matter.

[Jim_Weathers]

Jim will be the final judge... but i'm glad to see O Girl out in the wilds!

yes indeed, Jim will be the judge regarding these images. The thing with AI generated images is that, at the moment you can't copyright them unless substantial work was done to customise the image. So, as far as the O-Girl images here are concerned, the only person who has any claim to the copyright is Jim, because these images are based on work he did or owns the rights to. Therefore, as far as I'm concerned, these images belong to Jim and he can do with them as he wishes, I make no claims on them.

From what I've seen on the internet, AI images are not copyrightable. Because even if you have spent hours and hours working on an image, it is the software that is doing the work. It is all going to blow up eventually. With ChatGPT getting sued by the Times, this is all going to get crazy. It is the future regardless. Soon, people will be able to make whatever they want... it's going to be wacky.

[Jim_Weathers]

It is basically my stupidity in all of this. I am using mage.space, and I can seed an image, but it doesn’t seem to do what I want. So yeah, it would be great to be able to put O-girl, or a Yvonne Craig Batgirl into a quicksand situation, but I’m at a loss as to how to go about it. Just like anything else, this is another example of a new technology that I just don’t understand at all.

I use a much more straightforward process.The tools are much more simple than you might think if you can do a few workarounds. Let's start with a (fairly low-res) O-Girl screen cap.
o.png
Mage.Space has a very useful inpainting function when you upload your reference image. You can manually paint a mask over the sections you want the AI to modify, or you can use the foreground/background buttons to automatically select the section you want to work with.
o_inpaint.png
Now I can add in my prompt. This is pretty complex in itself as I have my preferred blend of secret herbs and spices, including a few LoRAs, but throwing in something along the lines of "swamp" and various other background items, using the Tiki set, then followed with some AI Hi-Res fixes to up the resolution and fix the grainy low-res textures, I get something like this.
o_backgroundswamp.jpg
Alternatively, I could photo-edit the screen cap. Using Photoshop, I paint a crude background.
o_swampdraft.png
Then run it through the same prompt using the background inpainting tool, followed by the Hires Fix tool.
o_swamp.jpg

Thanks for the info - I will give this a go and see what I can create

[cerberus]

From what I've seen on the internet, AI images are not copyrightable. Because even if you have spent hours and hours working on an image, it is the software that is doing the work. It is all going to blow up eventually. With ChatGPT getting sued by the Times, this is all going to get crazy. It is the future regardless. Soon, people will be able to make whatever they want... it's going to be wacky.

I think, if an AI images is simply generated from a prompt, it is not copyrightable; however, if a "substantial amount of work" has been done to the image then it is copyrightable. Of course, nobody has yet defined what is meant by a "substantial amount of work". If an image is generated using other images (e.g. for faces, poses or as a base image) that are copyrighted, I think a good argument could be made that the generated image shares the copyright of the used images. This has still to be tested in law, so at the moment it is the wild west out there.

[cerberus]

I use a much more straightforward process.The tools are much more simple than you might think if you can do a few workarounds. Let's start with a (fairly low-res) O-Girl screen cap.
o.png
Mage.Space has a very useful inpainting function when you upload your reference image. You can manually paint a mask over the sections you want the AI to modify, or you can use the foreground/background buttons to automatically select the section you want to work with.
o_inpaint.png
Now I can add in my prompt. This is pretty complex in itself as I have my preferred blend of secret herbs and spices, including a few LoRAs, but throwing in something along the lines of "swamp" and various other background items, using the Tiki set, then followed with some AI Hi-Res fixes to up the resolution and fix the grainy low-res textures, I get something like this.
o_backgroundswamp.jpg
Alternatively, I could photo-edit the screen cap. Using Photoshop, I paint a crude background.
o_swampdraft.png
Then run it through the same prompt using the background inpainting tool, followed by the Hires Fix tool.
o_swamp.jpg

I'd probably leave Inpaint until you are getting reasonable images and are happy with the basics, it can be frustrating watching it generate the wrong stuff again and again. I haven't had much luck with Inpaint yet, but haven't really tried that hard. The first thing is to select a decent checkpoint model (I like AbsoluteReality and DreamShaper, they should both be available in mage.space, I notice that Viridian seems to favour EpicRealism, that's good, I use that sometimes as well) . Then work on your prompt and negative prompt and select an appropriate style (the type of image, photo, anime, etc). Start with a square image (512X512 or 1024x1024 if using SDXL models), this is the size the models were trained at so you get less problems with generated images at these sizes.