AI Generated Art - The Story Behind The Alien Biomech Portrait

I recently added a new artwork titled Alien Biomech Portrait to the portrait art gallery. Alien Biomech Portrait is a modified version of an image I created using Stable Diffusion. This portrait came about as a part of testing out some Python programs I was writing to access the Stable Diffusion libraries from Python programs running in the Google Colaboratory environment. This was all done as a part of developing and testing Python programs that were to serve as the basis for two workshops I was going to teach: AI Generated Art Using Google Colab and Stable Diffusion Workshop for the 2023 Capricon Science Fiction Convention and Using AI Tools to Generate Art and Text With Stable Diffusion, DALL-E, and GPT-3 Using Google CoLaboratory Workshop for students attending the 2023 International Space Development Conference.

Having completed my program testing regime, I set the images I had created aside and sort of forgot about them as I pursued other projects. That was in early 2023 and this image, along with a host of others, became candidates for deletion as I wanted to free up space on the drive they were on. In going through the images, it struck me that a number of them had artistic value. It was then that I decided that I would do something with some of them.

Unfortunately I no longer seem to have the various prompt parameters I used in the creation of the various images. At that time understanding how the various prompt parameters impacted the image being produced was important to me so they were all recorded in a spreadsheet. I needed to understand how each parameter, and combinations of parameter values, impacted the generated image so that I could convey that knowledge to the students of the two workshops I wound up teaching. These prompt settings included the text prompt (which is the central focus for most folks), the negative prompt, inference steps value, guidance value, size value (image size and aspect ratio impact the results), and seed value.

Post Processing of the Stable Diffusion Generated Image

To be clear, the modifications I made to Alien Biomech Portrait did not alter the structure of the image. Rather my processing involved cleaning, enlarging, coloring, sharpening, and contrast alterations. These are the types of image processing operations that a photographer would use in processing a photograph for publication.

The first thing I did was to use the DCC (Directional Cubic Convolution Interpolation) function of G'MIC, to upscale (enlarge) the image. DCC does a good job of enlargement, certainly better than what Adobe Photoshop does with its bicubic upscaling. I also played with using the Upscayl program to enlarge the image but in this case I felt that DCC did a better job at enlarging the original image.

Next, I opened the image in RawTherapee, a free, open source raw image processing program that is available for Linux, Windows, and Mac devices. I used RawTherapee to perform a number of global image enhancements. The options I used included exposure settings, tone mapping, local contrast, edges, haze removal, white balance, film simulation, and CIE Color Appearance Model 2002. G'MIC film simulation has a large collection of pre-defined CLUTs (Color Look Up Table) that are used to remap the colors in an image. I've created quite a few of my own custom CLUTs and used one of those to remap the image's colors.

The last step was to use Photoshop for additional coloring and manual repainting of some bits of the image to create a finished image that I was happy with. The recoloring I did in Photoshop was to set a paint brush to color mode and replace the existing color of some feature with a color more to my liking. This is most evident with respect to the 'eyes' of the portrait. The original image had no color differentiation for the eye - they were just part of the background. Coloring them red really makes them stand out. The manual repainting involved doing touch up work to remove what I considered to be blemishes - those being small features in the image that I felt created a visual distraction. Note that the work I performed using Photoshop could also have been done using GIMP.

On the Subject of Generative AI Created Art

One thing that saddens me is the fear and loathing displayed by a number of creatives towards art created by generative AI - an attitude which I don't consider to be all that different from the fear and loathing many traditional artists displayed towards computer art when it first arrived on the scene. This fear, as with similar fears displayed whenever a new technology emerges, is one of protecting one's own financial self-interest while ignoring the greater benefits that the new technology offers. This is an attitude that I find to be Luddite in nature and most unfortunate. I celebrate the arrival of generative AI as yet another accomplishment of the human intellect.

Parting Thoughts on Generative AI

I confess that since my initial exploration of using generative AI to create art and text responses, I have not pursued the new developments with those AI tools. Nor have I created anything (art or textual) since concluding my research for creating the two workshops I put together on the subject of generative AI. For my part I am mainly interested in the algorithms and technology behind generative AI and the associated Large Language Models (LLM) and how they can be used for our benefit in a wide variety of fields. I am far less interested in actually using these tools myself. As a creative coder, I enjoy the challenge of developing new generative art programs far more than the challenge of creating a mix of text prompts to create something using generative AI applications.

References:

Google Colaboratory: Commonly referred to as Colab, it is a virtual machine environment in the cloud that allows people to run Python programs with access to Google's GPUs. It is a wonderful tool and the environment is quite easy to learn. And it's free, with paid accounts having access to a greater level of memory and GPU resources. For more, see Welcome To Colaboratory: an introduction to the Google Colab environment and A TensorFlow Youtube video on Get started with Google Colaboratory (Coding TensorFlow) (TensorFlow is an open-source library for machine learning and artificial intelligence).

Rawtherapee: Rawtherapee is a free, open source raw image processing application for Linux, Windows, and Mac. It is a wonderful alternative to Adobe Lightroom and is my image processing application of choice when working with photographs.

G'MIC (GREYC's Magic for Image Computing): G'MIC is a great image processing framework that functions as either a command line tool (very handy) or as a filters plug-in for GIMP, Krita, Photoshop, Affinity Photo, PaintShop Pro, PhotoLine and Paint.NET. For my part, I regularly use the command line option (which is great for batch processing), the Photoshop plugin, and the GIMP plugin options.

Upscayl: Upscayl is a free, open source AI Image Upscaler that is a cross-platform application built with the Linux-first philosophy. I have minimal experience using this software so am unable to make any definitive comments on the quality of the image enlargements (aka upscaling) that it performs.

| Return to the Blog Index | This entry was posted on DAYOFWEEK, MONTH 99, YYYY and is filed under Computer Art, Creative Coding, Digital Art, and Generative Art.