endgame dilemma

 

dreamboat excursion

 

wizard of oz convention

 

I used a DALL-E mini mega generative model to generate 200 256x256 images in a folder.  I then loaded them into a Transition Context in Studio Artist that loads the source area and subjected that virtual source image to a energy based model digital paint strategy into a higher resolution canvas

New Gallery Show Features

 

Finally added the gallery show preset memories to Studio Artist pre_V6.  Makes it way easier to try out custom art strategy ideas in Gallery Show while working on the fly.  Inaugural image here generated with the world's simplest GS cycle strategy using the paint synthesizer.

Working with a folder of mini DALL E generative ai synthesized source images for the gallery show grab below.

The gallery show paint art strategy being used here adds fine detail at the full canvas resolution (generative ai output was 256x256 pixels).

the End of Time

 

Besties

 

Dancing Propaganda Stabbing - 2

 

Baby Propaganda Poster Transition

 

Dancing Propaganda Stabbing

 

Collective consciousness

 

Working with sets of small mini DALL-E multi-modal generative synthesis thumbnails as source feed stock for Gallery Show in Studio Artist.  So a virtual source image created by a generative deep learning system is then rendered by a second generative ai system building new visual processing effects on the fly.  Keep in mind that the virtual source has been virtualized even more by Gallery Show, since any given output painting is a function of multiple source inputs that might be subjected to additional data augmentation by the system.

I'm still figuring out the best approaches to working with the extremely small thumbnail images generated by the mini DALL-E system.

baby Propaganda poster

 

All of these are using a mini DALL-E implementation on Hugging Face that uses a VQGAN rather than a VQVAE like the original DALL-E paper.  And the dataset of training images is 28X smaller than what OpenAI used in DALL-E.

But the last 'baby' example clues you into something fundamental associated with the mini implementation VQGAN.  You could probably improve the representation by adding additional layers to the model. But that kind of artifact is also associated with simple ReLU nets configured to represent images.  So a different activation function (think implicit neural representation like Siren) would be a better strategy i think for that part of the system.

the Brutality of War

 

2 People Walking on Beach

 

Paia Street scene