AI Image Creation With Grok-2
Instant picture making with a sentence.
Musk announced Grok, the company’s answer to ChatGPT, in November, while Grok 1.5 debuted in April. The chatbot is available exclusively to Premium subscribers on X. Musk's xAI startup is also preparing to build what might be the world’s largest supercomputer in Memphis, Tennessee, which will handle Grok training. Emily Price, PCmag.com.
Long before I became a photographer I had studied art. Fine art to begin with and later graphic design and advertising. During this time I’ve done some of everything for designing logos, airbrush illustrations, fine art painting, portrait painting, etc. Later, when home computers became the next big thing I was interested in using them for design and art projects. I recall my first experience in the mid 1980’s with 3D rendering using Sculpt 3D on the Commodore Amiga. Just rending a few objects took hours, sometimes in the evening running the computer all night to render simple images. But, it looked fantastic! From there on I was on my way with computer graphics and added them to my stock portfolio.
One day I received an email from one of my photo stock agencies that AI images were acceptable for submission and sale. I knew nothing about it and was unsure where to start. Then recently, Grok-2 was announced by Elon Musk for August. Up until then I had only used Grok to answer questions for research purposes. I didn’t know it could render images as well. Once I learned that Grok could create images I decided to explore what it could do. What follows are my experiences with this new imaging technology.
Creating An Image
It’s as simple as writing a descriptive sentence. For the image above for the fruit I simply wrote “generate image still life.” And in a few seconds it appeared. You again do variations by using the Regenerate icon below the picture or typing try again or again and a different image will appear. (Note that In clicking Regenerate the image will be made over the original; use again to see the new image in the thread.) Sometimes it’s exactly what you want and at other times slightly off or it can be way off. See my examples below when things don’t go as envisioned.
With images of people, or random groups, I noticed that Grok-2 has a tendency to generate people mainly as Caucasian. In this example I had to be specific if I wanted to see a doctor of another race.
Often commands at this beta stage are not conveyed to the code that generates the image. A case in point is trying to render scenes on Mars. The AI constantly puts a large moon in the sky, often resembling our own moon (this occurs with moonscapes as well). After many attempts my workaround was to say “…the Martian desert clear sky.” It worked. Simply adding the words Martian and desert worked. It is beta after all, and the parsing of commands should be improved over time.
Photographic Realism
No doubt, the images being created by Grok-2 certainly look real enough to the untrained eye. For those of us who have been working in the imaging field for years, these AI images look slightly off. Many have that clean and pristine look wherein CG pictures reveal their origin. With certain genres of imaging, such as a stand of forest it will have no dead or fallen trees among them.
At this stage of development defects occur as well. At present Grok is not good at signage, either on buildings or insignia on aircraft. The letters are often mangled and unreadable. One rendering of Bill Gates gave him an extra left hand.
Accurate depictions of famous people depend on how well known they are. Some such as Donald Trump are mostly spot on. But the notorious Jeffery Epstein, despite being well known, never resembles himself. Since I am not familiar with the complexities of how these images are generated it is to be expected in this level of beta.
A Variety Of Styles
Besides photographic realism, Grok-2 can produce a great variety of image styles and artistic techniques. After telling the AI what I want it to render I will add the following: watercolor style, anime style, etc. Most of the time I get what I want and other times I have to regenerate the request.
Here are some of the examples it rendered.
Watercolor
Oil Painting
Cartoon
Anime
Pen and Ink
Pencil
Line drawing
Steampunk
Who Owns The Copyright To This Images?
The next frontier is Copyright laws and how they are addressed when using AI generated images, or rather, Generative AI. A lot has not been decided in case law and it’s an ongoing murky area. So far, Congress has not crafted any new amendments to the Copyright Act regarding this issue. Who really owns what? The user is only typing a sentence which sends a signal to the underlying databases working on numerous servers which are doing the rest of the image crafting process.
As stated at crsreports.congress.gov:
A final question is who is (or should be) liable if generative AI outputs do infringe copyrights in existing works. Under current doctrines, both the AI user and the AI company could potentially be liable. For instance, even if a user were directly liable for infringement, the AI company could potentially face liability under the doctrine of “vicarious infringement,” which applies to defendants who have “the right and ability to supervise the infringing activity” and “a direct financial interest in such activities.” The lawsuit against Stable Diffusion, for instance, claims that the defendant AI companies are vicariously liable for copyright infringement. One complication of AI programs is that the user might not be aware of—or have access to—a work that was copied in response to the user’s prompt. Under current law, this may make it challenging to analyze whether the user is liable for copyright infringement.
So we’ll have to wait this one out.
Endnotes
I have noticed that in some of the renders what appears to be an artist signature. It’s distorted making it unreadable. I take it this is the result of the processes used to craft the image. One can assume in later versions of Grok this will be ironed out.
Sources
Grok-2
https://www.pcmag.com/news/elon-musk-grok-2-ai-arrives-in-august
Copyright Issues
Generative Artificial Intelligence and Copyright Law
https://crsreports.congress.gov/product/pdf/LSB/LSB10922
Generative AI and Copyright Law
Will copyright law enable or inhibit generative AI?
https://www.weforum.org/agenda/2024/01/cracking-the-code-generative-ai-and-intellectual-property/