In the last few months, there has been a massive spike in interest surrounding the rapid development and wider availability of artificial intelligence (AI). This rise in popularity is likely because of claims that newer software can produce creative assets with minimal effort whilst retaining a similar quality as if a human had created it.
Another reason why attention has shifted to AI imagery is because of the recent controversy surrounding the winning picture for Sony’s World Photography contest. When this entry from artist Boris Eldagsen was chosen, he declined the prize admitting it was the product of AI.
He applied to test if “competitions are prepared for AI images. They are not.”1 It is clear these tools are becoming more sophisticated and it’s proving challenging to distinguish between human and machine-created outputs.
Ref: Pseudomnesia / The Electrician. Sony World Photography contest2
While much has been said about these tools recently, AI image creation has been around for quite some time. In the early 1970s, Harold Cohan was one of the first creators to make a drawing machine that took commands to produce images.3 He acknowledged the results were primitive, only using ink to create an outline, but it achieved the goal of standing out and drawing more attention to this unconventional technique. The same can be said for Eldagsen’s winning photograph. Had the image been a regular photo it wouldn’t have caused such controversy.
Ref: Turtle. Collection of the Computer History Museum4
With a tool that creates results almost impossible to distinguish from ‘real artists’, why doesn’t everyone try to enter and win such prestigious awards? At a glance, the process sounds straightforward enough. Just add some keywords to an algorithm that makes amazing photos, right? However, there’s more to it than just spamming commands. On top of learning these processes, there is also a level of understanding needed for what makes a strong image.
Some tools allow you to go into further detail, choosing where the focus of the image should be, what lighting to use and even the option to match a particular photographer’s style (see below for an example of aesthetic presets). But all require an input, a starting point whether that be keywords or an existing image to manipulate.
There are a range of platforms that offer different levels of treatments. AI that produces more detailed effects takes time, money and plenty of practice to learn. In some instances, as with NightCafe, DreamStudio and DALL-E, they require bulk payment for credits (using one credit for each image created).
Ref: Example of image creation stage with Nightcafe
The most popular tool at the moment is Midjourney which uses text prompts in Discord to make images. Adding a keyword creates a particular element to include in the image but adding more text doesn’t mean more detail. It is the application of a combination of shortcodes, style and cool effects alongside your image description that can generate some staggering results.
At a basic level, say you wanted an image of a ginger cat on a street you would enter into the application (ginger cat) and (on a road) as these are the key pieces of information. You could then enrich the image by adding (a city) for the background and (black and white) for contrast. The idea is that these are desired elements but if not all of them can be incorporated then you would still be satisfied with the results (as shown in the example below). Four renditions of the image are then created for you to choose from. After selecting one you can upscale and ask for further results and refinements.
Ref: Four renditions from prompt 'ginger cat on a road in a city, black and white'
Adobe is also jumping on the AI bandwagon. Currently, in its beta form, Firefly is taking a much broader approach generating not only images but interactive 3D models, highly customisable templates, and detailed logos. Firefly only requires a simple sketch to work from. This is great for businesses with a small team or a lower capacity for creative work.
At the same time, it is also appealing to users with less experience with the likes of Progressive Controllable Image Synthesis which uses simple marks and brushstrokes to easily add desired elements to an existing image. For example, Firefly can recognise a drawing of two circles and a line in between as a pair of glasses which can be easily added to a portrait.
Ref: Meet Adobe Firefly5
With so many possibilities, and with technology pushing the boundaries of what is real or artificial, there also come risks and the opportunity for exploitation.
Adobe has commented that:
‘With the volume of digital content increasing, people want to know that what they’re seeing online is authentic.'4
They are starting to introduce industry initiatives to discourage malicious intent with AI such as the Content Authenticity Initiative. This helps to declare whether an image is made with AI to develop an industry standard for digital content attribution.
Just as AI can be used for good, it can also be used for malicious purposes. These new industry initiatives highlight the need for a standard to curb disinformation. The internet is already full of misinformation so pairing a piece of fake news with an image could further cement the credibility of false stories.
We also tend to overshare personal information. It doesn’t take much for someone to take something in the public domain, say your profile picture, and use it to create a new image showing you in a negative light or even steal your identity.
Speaking of oversharing, we don’t always know what happens to the images we manipulate. While a team photo where employees are turned into elves for a Christmas card might sound fun, you need to consider what happens to the pictures you submit.
There are so many new tools being made that it’s easy to accept and skip past the terms and conditions and put your images into cyberspace without a second thought about who else gains access to them.
Unlike Mr Eldagsen, not everyone has an eye for great photography. A majority of these software (primarily the inexpensive or free ones) are marketed for the average person to use. Without hours of tweaking and editing code, the results aren’t always accurate and can easily slip into the uncanny valley. Although the technology uses millions of images of people for reference this doesn’t mean the results are completely accurate. From lighting that doesn’t quite fall right on a face to hands that gain an extra finger (like below - yikes), there is still more to be done to perfect image likeness and accuracy.
Ref: Image generated from prompt 'man holding head in hands'
With so many issues and no guarantee for a picture-perfect image (without lots of refinement), why don’t we go back to searching for and licensing images? After all, by relying on AI, aren’t the creators, photographers and artists being cheated out of a job? Maybe these tools have become too convenient, and we are getting more complaisant or maybe even lazy.
Although some may have concerns, there’s a reason AI has become such a hot topic recently. Whether it’s for a blog, social post or website, all businesses need images. Scrolling through thousands of results only to choose a picture that is ‘close enough’ to what you want can be frustrating and a waste of time.
Then there is navigating the appropriate license type, its expiration period, and the requirements for artist attribution, all of which can be time-consuming. Additionally, even with a paid subscription, premium images may entail unexpected additional expenses. It can be disheartening to invest significant time in finding an image only to encounter further charges, leading many to reluctantly spend more. At least with AI, the cost for a credit is the same each time so there are no nasty surprises!
These tools also mean you can cut right to the chase and specify exactly what you need in an image. Say you wanted a picture of a blue frog but could only find images of green ones, instead of having to change the colour with the likes of Photoshop, you can simply specify those extra features and get exactly what you want by entering your keywords.
Having so much control and creative freedom also means the images you create are unique to your company and helps your content to stand out against the other generic and uninspiring stock images.
While AI is becoming more integral to the content we create today, there are many misconceptions about its capability. Some people see it as a solution to all their problems, creating instant content that will hit the mark each time, while others fear it could replace their roles. Neither of these is completely true. Yes, the technology has advanced and is far more capable of performing more complex tasks than it could fifty years ago but self-learning still requires information and code to be input from a human. Rather than see it as the be-all or end-all of image creation, we should use it as a building block to keep refining our images and acknowledge that it is our vision that shapes the final outcome. There is no true algorithm for creativity.
Thankfully the Essential team has yet to be replaced by AI as great images still benefit from a human touch. If you are looking for engaging creative content that stands out from the crowd why not get in touch and see how our creative team can help?
References:
[1] Grierson, Jamie. Photographer admits prize-winning image was AI-generated, April 17, 2023: https://www.theguardian.com/technology/2023/apr/17/photographer-admits-prize-winning-image-was-ai-generated
[2] Eldagsen, Boris. Pseudomnesia / The Electrician. Sony World Photography contest. 2023
[3]Garcia, Chris. HAROLD COHEN AND AARON—A 40-YEAR COLLABORATION. Computer History Museum. August 23, 2016: https://computerhistory.org/blog/harold-cohen-and-aaron-a-40-year-collaboration/
[4] BBC. 6 epic translation fails: https://www.bbc.co.uk/programmes/articles/2YYBmQsxxB9TFLbd9gKRwpn/six-epic-translation-fails
[5] Adobe, Meet Adobe Firefly: https://www.adobe.com/sensei/generative-ai/firefly.html