I used to draw portraits of people all of the time. So, I trained a checkpoint on those portraits. As well as some sketchbook drawings.
I had originally called it Art School Sketch, as it was meant to do sketchy portraits of people, but it turns out it does some pretty great other things. Lighting and contrast come out way better than expected.
Negative prompt Cartoon makes for a nice realistic lighting style.
Description
FAQ
Comments (6)
Thanks for sharing this. It's so fun to see people making models of their own work.
It'd be cool to see some of the training data.
It's mostly stuff from this gallery
https://www.deviantart.com/thepermman/gallery/84435345/handmade
I had drawn a bunch of portraits. A lot of them I photoshopped in different backgrounds though, from photos with an accented edges photoshop filter. That way the person sticks out.
JustMaierDec 18, 2022
Neat. How many steps did you do and what model did you train on top of?
I used openart.ai. I haven't figured out how to train models free myself, just how to upload the images. All the rest of the stuff is handled automatically for less colab savvy people like myself. I don't know how many steps it took. It took a little over an hour.
Here is a contact sheet of the training images
https://www.deviantart.com/thepermman/art/Training-Images-941484017
JustMaierDec 18, 2022
Have you tried to do a personal model on OpenArt? I did one the other day and it didn't turn out as good as I had hoped. I used the same training data previously and got results that were perfect.
I did one with my face a week ago. You actually asked me about it. It was that one with the Harley Quinn pictures. Mine came out pretty good despite not spending much time curating the photos. I've been able to do all sorts of things with it. I think I just had some photos from a decent camera was what helped the most. OpenAI did all the cropping, and it came out well. I cropped the sketch style ones myself though.
One thing you might want to try is uploading at least one photo touched up with an algorithm like GFPgan. The photos were over a 15 year span, so when it trained it was able to train on different versions of me. Also one of the photos had some heavy light shadow contrast. So, it was able to figure out the 3d of my face.
https://www.deviantart.com/thepermman/gallery/85548593/ai-self-portraits
Details
Downloads
456
Platform
CivitAI
Platform Status
Available
Created
12/18/2022
Updated
6/27/2026
Deleted
-
Trigger Words:
sks style
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.