From disturbing Deepfakes to hyperrealistic images, AI has revolutionised the creative landscape as we know it. Machine learning tools like Midjourney have changed the content we consume and produce, be it in the realm of marketing, social media or fashion. But what about food – is AI food photography the future? And if so, are traditional creatives ready to embrace that radical change?

In collaboration with our client, SA Pork, we prompted Midjourney to reimagine food images, determining whether AI can outplay humans in the creative department.

Is there control with Discord?

AI has filtered into our everyday lives. Since its inception in 2022, the popular language model tool, ChatGPT, has allowed professionals to craft compelling copy, and can even help home cooks decide what’s for dinner by suggesting recipes based on what’s in their fridge.

When it comes to the visual aspect, text-to-image generators like Midjourney, DALL-E, Stable Diffusion and Adobe Firefly are equally impressive. Midjouney has captivated creators and social media followers with its ability to produce hyperrealistic images using a singular prompt. All users need to do is join the Discord server and communicate with the chatbot to unleash limitless creativity – and when we say limitless, we mean limitless!

“So much of the amazing AI-generated image content around is highly conceptual and artistic – for example, brief it to create an image depicting a dancing unicorn in a Dali-esque style and you’ll get a fantastic image that combines all of those things in a way that you probably couldn’t dream. In fact, it will be able to give you a million versions of that same image that will all look pretty amazing, because you had no preconceived notion of exactly how you wanted the image to look. When it comes to a super-specific brief, like generating a food image, it gets more challenging,” says Crush Food Editor and Stylist, Julie Velosa.

While the capabilities are astonishing, AI food photography presents newfangled challenges for content creation agencies with clients to impress. Don’t even get us started on the question of originality – we’d ask ChatGPT to write an article, but it “is at capacity right now”. Nonetheless, we were hungry for answers…

---

Trusting the creative process

In collaboration with our client, SA Pork, we prompted Midjourney to reimagine food images, determining whether AI can outplay humans in the creative department. Naturally, we pitted our trailblazing production team against the skilled text-to-image generator.

The brief was simple, create a digital image for a recipe using Midjouney; the creative process, on the other hand, proved frustrating for experienced creatives, yet novel users.

With the ‘/imagine’ command and a detailed prompt, we asked the Midjourney bot to generate a “photorealistic image of a pork schnitzel double patty burger with a side of onion rings”. In a matter of minutes, it churned out multiple images with variations. Though impressed with its capabilities, the images looked commercial and didn’t match our unique style. Our team struggled with the inability to change surfaces and backgrounds with ease, getting exactly what was wanted, which is what would be expected when delivering on a client brief. Overall, we felt that the images had ‘yum factor’ but looked unrealistic, with elements that weren’t briefed originally appearing in the shots. We felt they lacked texture and believability.

The one obvious pro, of course, is the speed at which the images are produced, and the fact that you need very little equipment. “AI is like suddenly realising you have a team of a thousand expert assistants that you never knew existed, your old HD camera instantly upgraded to 8K resolution, or having every camera lens your heart desires, a never-ending room full of props or access to global locations. But don’t presume you will immediately be able to create with AI what you have done in the past with a traditional process that you have been practising for years,” said Crush Creative Director & Photographer, Matthew Ibbotson.

“It’s definitely a frustrating yet exciting process, learning how to meld your traditional skills with AI,” he added.

---

AI food photography – better or just more convenient?

Traditionally, food content production requires sound strategy and a thorough understanding of the creative brief before even setting foot in a studio. Establishing mood boards to source creative inspiration, and finding props and ingredients to help bring concepts to life can be laborious, but necessary.

Sure, harnessing AI food photography could mean significantly reducing your campaign budget and sparing resources like time, but it doesn’t compensate for the ‘human touch’ – especially when it comes to something as personal as food.

Despite providing detailed prompts and specifying the camera model, lens type and aperture ratio, there were a few inconsistencies among the generated images. While high resolution and good quality, they often deviated from the prompt and took creative liberties like featuring random ingredients. Even after streamlining our initial prompt, the revised version lacked character and was aesthetically bland, to put it mildly.

AI ain’t going anywhere – whether it will replace food photographers and stylists overnight is doubtful…

The true advantage that AI definitely has is efficiency – buying ingredients, hauling props, setting up lighting, reshooting, restyling styling sets, prepping and cooking food on the day, and editing shots take up time and resources. Ultimately, the vision might not match the outcome when using Midjourney, especially if you don’t have a world of experience.

When compared, the untrained eye might have trouble spotting the difference and knowing the human-captured shot from the generated image. Certain elements like the texture of the meat, the lighting, sprinkled salt flakes and other details might hint at the fact that the left image was produced by our creative team. The image on the right has enhanced lighting and shadow placements; the texture of the lettuce and the sauce at the base of the bun all look rather realistic, but the schnitzel looks polished and over-edited. It also has sesame seeds in the coating, which were not in the brief or the recipe.

Accuracy is also crucial when capturing content for clients. The image on the right includes random ingredients that were not part of the prompt like the addition of pickles; it also omits the caramelised onions listed in the recipe. In the end, we weren’t able to get the exact image we wanted using Midjouney, but our results improved with modified prompts.

When it comes to delivering scalable and high-quality content at record speeds, AI is more than capable of the task – but just because you can leverage it, does it mean you should?

“The lack of control around changing minor elements within the shot was frustrating – Midjourney has a habit of adding elements and ingredients that were not in the brief each time it recreates the image. So where you might have been happy with a certain part of the image, that element could be replaced with something else when reimagined,” says Velosa.

Did they resemble something straight out of a fast-food commercial? Yes, but as a food publication, we want to inspire people to cook through captivating content. Don’t get us wrong, AI can produce complex imagery; even in its experimental era, it remains a powerful tool that creatives can benefit from. But as far as producing compelling, unique and relatable content, it still has a ways to go…

“In terms of the capability of AI, that is obviously improving at a staggering pace. What it can do now versus what it will be able to do in 6 months’ time, and 6 months after that, is going to be very different. The things I found frustrating could very well evolve with updates to the software, and with our skills in using it improving as well,” added Velosa.

---

A burger with a side of artificial fries, please

Machine learning models generate photorealistic images using large-scale datasets. “To learn how to assess the aesthetics of food images, i.e., whether they are aesthetically positive or negative, we aim for high variety in the collected samples: food categories, viewpoint, lighting condition, and layout,” says Research Gate.

Most datasets are trained to prevent overfitting, a concept in data science where machine learning models produce accurate but over-optimised outcomes, thus failing to predict new datasets. In our case, we created a recipe for a burger – arguably one of the most popular foods, thanks to American food culture – which could very well have been overfitted.

Fast-food establishments often have similar visual traits displayed in online burger ads and social media posts – think saucy shots and flawless food images achieved through hours of photo retouching. So, while the images fit the bill, they weren’t unique – an important distinction for food brands competing within the digital and social media food space.

One thing is for sure, though, AI ain’t going nowhere – whether it will replace food photographers and stylists overnight is doubtful. All things considered, it has massive potential as a stock image generator, but it’s certainly not without its flaws, especially when it comes to generating realistic lifestyle images. Distorted objects, disfigured facial features or one too few or too many fingers on human hands prove that AI doesn’t always have the upper hand in the creative department.

---

Human creativity versus Artificial Intelligence

For brands and agencies alike, exercising creative control is everything and more. When it comes to delivering scalable and high-quality content at record speeds, AI is more than capable of the task – but just because you can leverage it, does it mean you should? Brands need to ask whether the content will resonate with audiences and garner engagement.

We’ll stick with the traditional creative process for now.

Still, it’s worth testing the waters and experimenting with different versions currently available. Specialised AI apps like Lunchbox and Swipeby have helped restaurants create images for online menus to drive sales, though, it could disappoint customers if their burger doesn’t even vaguely resemble the one from the menu.

As a content creation agency, we’ll stick with the traditional creative process for now. Crush Food Editor and Stylist, Julie Velosa, said it best: “While I don’t doubt the amazing opportunities that lie ahead for AI in the food space, there is an element of it that does feel quite soulless. The preparation of food and the enjoyment of eating it are actions of the heart, as much as they are of the hands. A picture-perfect dish created by a computer with not an ounce of love does feel quite sad. Perhaps I just prefer knowing that the sourdough I’m looking at has been lovingly kneaded and coaxed by human hands versus a string of code. Sentimental? Maybe. The day AI starts 3D printing my morning croissant will be when I know it’s officially my time to step down!”

Curious about the recipe? Try this Southern Fried Pork Schnitzel Burger with Avo Mayo, and tell us what you think.

For more tasty inspiration, explore these SA Pork recipes.

sapork.co.za | Facebook | Instagram