Impressions - May/June 2024 - 45

the programs, you get that tool's version of a blue cartoon horse.
Contrary to popular belief, these kinds of AI tools do not simply
copy and paste a mishmash of related images to come up with
a new one. Instead, with each new image request, an entirely
new image is created via a multi-step process known as diffusion.
Because Adobe Firefly, Ideogram and Midjourney are all trained
using different sets of images, you will not only inevitably obtain
different results, you will often obtain very different results using
the exact same prompts.
In addition, because these platforms' diffusion models all begin
with random noise, every image they create will be unique. Results
cannot be repeated.
Embrace The Awkwardness
Creative professionals have long mastered various commercial
design tools. Photoshop, Illustrator, Corel Draw and other programs
have formed the backbone of this industry since their emergence
in the early 1990s. Every trained designer knows how to use layers,
channels, the pen tool, blur commands and the other graphic
toolsets that are part and parcel of these programs.
AI image creation is different. Suddenly, your success has as
much to do with your vocabulary as anything else. Professional
designers who have mastered these skills are suddenly being out
designed by that accountant in the corner. It feels unnatural. No
wonder there is some friction within parts of the design community.
The good news is an untrained AI creation tool user doesn't
have the training, knowledge or creative mind to utilize these tools
fully. A professional designer using these new tools will invariably
create and iterate better images than someone without that kind
of creative eye-it's only natural. They simply need to get used
to how these platforms operate.
Which is the weirdest thing you can imagine. Words.
As they say, words matter, and with AI design tools even more
so. The fun part is the results stemming from different words that
mean the same thing can produce different results. For example, the
word " minimalistic " can be interchanged with " spartan, " " simple, "
" sparse, " " austere, " " basic " or " unique. " Each word will affect the
results differently when inserted into the same prompt sequence.
Words and word sequences are the alchemy employed to create
images as opposed to the tools the profession is used to. New
images can be rendered in under a minute. On Midjourney, you
can simultaneously create several dozen images with the correct
prompt commands in little more than the blink of an eye.
Choosing the Right AI Platform
Everyone has different tastes, skills, work habits and budgets, and
each of the aforementioned AI tools works a little differently.
Here's a breakdown of what you need to know:
DALL-E 3 - This platform is owned by OpenAI, the same outfit
that brought you ChatGPT. Since you will be using images created
for print, an HD image measuring 1024-by-1024 pixels will cost
you $0.08, while a 1024-by-1792 or 1792-by-1024-pixel image
will run you $0.12. DALL-E 3 is trained on 400 million image
pairs. You can use DALL-E 3 for free on the Bing Image Creator.
Adobe Firefly - This is baked into the Adobe suite, which many
graphic professionals already use. If that's you, you're already paying
for it. There is a free plan, but users who upgrade to the premium
version ($4.99 per month or $49.99 per year) can remove the
Adobe Firefly watermark. Adobe Firefly is trained on the images
in the Adobe stock image universe, which occasionally makes the
results look somewhat limited. Adobe Firefly is the backbone of
the generative fill engine for the Photoshop upgrade.
A closeup of two of the dozens of results shown in the image on the
facing page. Note how this kind of variation makes it possible to finetune
the end result. Images courtesy of Marshall Atkinson
May/June 2024
P.45
Impressions - May/June 2024

Table of Contents for the Digital Edition of Impressions - May/June 2024
