Using AI in your "traditional" artworks

[parkdalegardener]

Generating a 3D object with a clean mesh would be great; but the tools aren't there yet. Soon, but not quite yet. It's the depiction of the unknown backside of the mesh. Generative AI can't "invent" unseen information like the back of a mesh with accuracy. However; there are models being refined ATM that will do such. If you are feeling energetic and willing to suffer disappointment, I can turn you on to such models. I don't recommend them for casual use.

Now; if you want to turn a sketch or drawing into a "realistic" character; not a problem. Something called Controlnet can do this easily using a number of different models. I'll show something later on. It's a little early in the AM to get my motor running, even with a pot of Fafnir's brew.

[Noshoba]

I think park they missed another use for AI that would help with their "normal"art.  Reverse engineering.  Not just using it for postwork. Use it to take a anime/style character over to the more realistic look that can be modeled in Daz or Poser.

[parkdalegardener]

You're welcome Hawk. Hope you can use it.

[FrahHawk]

This came across my desk this morning as part of my nVidia developer news. Free but you need RTX graphics. Well nVidia; what would one expect. Win 10 and any RTX enabled card will do it as far as I know. https://www.nvidia.com/en-us/studio/canvas/ Should integrate easily enough with whatever other software you use as it will output PSD documents. I used a limited release a year ago and it was pretty good then. This beta update came out for the Consumer Electronics Show in Vegas a week or two ago.

WHOA!!! I have been meaning to explore this. Wasn't sure what it was about. Glad you posted about this, PDG! Now I think I can do my AI artwork offline. Right now, I'm trying to work out an idea for this month's challenge (Egypt). Just saw a movie about the Egyptian gods and that gave me an idea for a scene. So, off I go! 

[parkdalegardener]

This came across my desk this morning as part of my nVidia developer news. Free but you need RTX graphics. Well nVidia; what would one expect. Win 10 and any RTX enabled card will do it as far as I know. https://www.nvidia.com/en-us/studio/canvas/ Should integrate easily enough with whatever other software you use as it will output PSD documents. I used a limited release a year ago and it was pretty good then. This beta update came out for the Consumer Electronics Show in Vegas a week or two ago.

[thelufias]

And they thank you PDG....

[sanbie]

 

[parkdalegardener]

Well I won't show you the video generated. When I got back from dinner all was done. Sadly it is in the state of causing one visual harm with all the jittering. In a hurry to post before I went out yesterday I failed to follow my own advice and check a frame or two near the centre of the video. The little robin took a jump and spin while eating it's seed on the branch. This created a minor freak out for the AI. It did what it was supposed to do; but AI isn't intelligent despite the name and there is actual visual harm and torture to the poor little robin in the machine's output. We can have that. I'd have to start posting those disclaimers about no digital animals being harmed in the creation of the work.

[sanbie]

Very interesting read pdg

[parkdalegardener]

If we take that seed value and plug it into the AI at a later frame, somewhere near the middle of our video; we can check to be sure all is good or to adjust our prompt if necessary.

We would do that yet again for a frame near the end to quality check. If all looks good just run img2img in batch mode specifying your input images folder and where you want the output images to be placed.

It is then a simple matter of reassembling the frames with ffmpeg, or your editor of choice; back into a mp4, avi, or what have you.

I'll post a link to the output animation when it's done. My lift is here and I'm off.