KRЯRL DЯAWINGS: Photo Booth training by Open Art

Open Art trained on my style in their Photo Booth feature -- and then the AI program generated fake Krrrl figure drawings:

Playlist Best Krrrl Drawings 2022

Melobytes

Fake KRRRL Drawings

-- created by Open Art Photo Booth

Note: The two music selections for the video above were generated by AI from the first and the last image, in the image-to-music program of Melobytes.

Photo Booth cost $10 and will generate 400 images from my style model. After a month my CKPT model (about 2 gigs) goes away. However I can download the CKPT model file and use it in Stable Diffusion.

Download my CKPT model (2 GB)

YouTube tutorials:

I "trained" this model on the first 20 drawings from my book "Finish My Figure Drawings." Note: I trained on the web sized drawing, approximately 640 pixels in height and width:

Finish My Figure Drawings

The Photo Booth program automatically cropped my images because it only works with square images. While it did allow me to crop the images myself, I accepted their default cropping.

I trained on these first 20 images

from my book

I input "woman" as a class, since most of my figure drawings are of women:

It took about an hour to train on my style. Then it was ready to generate images:

GENERATIONS

of Fake Krrrl Drawings

Preset - Artistic Museum Portraits - Vintage Archie comic book

prompt:"anatomical drawing of a running woman"

prompt: "anatomical drawing of a woman swimming"

prompt:"anatomical drawing of a running woman"

prompt: "drawing of three woman dancing"

prompt: "drawing of a woman at a coffee shop"

Thin-Plate Spline Motion Model for Image Animation

This is not a Rhinoceros

(homage to Magritte)

The prompts do not always deliver. There is NO "rhinoceros" in the image below. But then again, I trained on 20 of my figure drawings, so perhaps my CKPT model does not even know what a "rhinoceros" is:

prompt: "rhinoceros"

Likewise, the generated image below may not be of a woman, but it is definitely not of a "woman riding a rhinoceros."

prompt:"woman riding a rhinoceros"

I like the abstract qualities

of this figure generation

Previous Failure

Colab Dream Booth

A month or two back I could not get the Colab for Dream Booth to work for me in this program --DreamBooth_Stable_Diffusion.ipynb. It stalled out by failing to upload folder. I believe that the program used a lot of resources, so I am not sure if my computer and graphics card were inadequate, or I needed to pay for a Colab Pro account to get enough muscle to run it.

There was a YouTube tutorial on all this -- Create Art From Your Face With AI For Free | Dreambooth Tutorial | Stable Diffusion Guide.

PREVIOUS FAKE KRRRL DRAWINGS

2020

Runway ML

In 2020, AI generated infinite Fake Krrrl Drawings after I trained the StyleGAN2 in Runway ML on 348 of my long figure drawings:

Runway ML -- 3rd "Model" (based on long poses)

One of the better

Fake Krrrl drawings of 2020

Runway ML -- 3rd "Model" (based on long poses)

500 Fake Krrrl Drawings

Download the PKL file here (364 MB)

This was my 3rd StyleGAN2 Model, and I wrote up a blog entry about it: Runway ML -- 3rd "Model" (based on long poses)

I summarized all three models that made the fake Krrrl Drawings in 2020 in this post: Artificial Intelligence creating Fake Bad Krrrl Drawings

2021

Playform

In 2021 on the Playform website, we trained Robert Atkinson's AI creations to alter my figure drawings:

Playform AI Experiments

I blogged about this too -- Playform AI Experiments

CONCLUSION

Photo Booth was easy and painless to use in Open Art, and was worth the $10 cost. It generated a lot better fake Krrrl drawings than the StyleGAN2 program in Runway ML two years ago.

Still I wish the program applied my drawing style to any images, not just human figures, but rhinoceroses and landscapes. I am sure that the AI now is capable of doing that.

Also I would like to make my style public domain, and allow anyone to use the model I trained on.

ChatGPT just came out (November 30, 2022), and isn't much of a help so far: