Among webmasters, marketers, designers and traffic buyers, AI has become the mainstream of 2023. Initially, it was the ordinary users who amused themselves with neural networks by creating psychedelic pictures and unusual avatars. Now neuroart is a powerful tool for performing routine tasks in a variety of areas where time = money.
Can AI replace the designer?
How much would you have to pay a designer to make a realistic image out of a sketch? AI products will perform this task instantly and free of charge. The value of a masterpiece is entirely another matter. Of course, it’s unlikely that the work of artists, designers, and illustrators will completely depreciate in the coming few years. However, with the emergence of AI products, it’s quite possible that novice graphic designers may lose a small “slice of the cake.”
For example, you can use the Logojoy service to create logos. In a matter of seconds, the tool will produce several hundred layouts and logos in different colours, and all you have to do is to choose the most viable ideas.
Although neural networks have come a long way, the images they generate often don’t come close to the real masterpieces. But we’ll talk about such sad things later.
How does Artificial Intelligence create images
AI services are based on mathematical models of neurons in the human brain. Although still far from perfect, these programs are capable of self-learning. The more tasks you entrust to artificial intelligence, the more new information it “remembers”. It can remember how elements look and what they are called, and can learn how to find identical and different fragments.
At the heart of Google’s image search system is a neural network “trained” to analyze images and offer similar ones. It was actually after Google revealed their approach and they published the source code of the algorithm that photo conversion tools began to appear en masse.
The most popular services are those that come up with a picture based on a text query. All users need to do is to specify the names of objects with a detailed description. The algorithm uses all the elements in its database to ensure that the results reflect the text query as accurately as possible.
What tasks can be resolved by AI image generators
AI image products today can copy artistic styles, bring models and portraits to life, and create new images. The different tasks are carried out by means of different tools.
The NST method based on ultra-precise neural networks can generate images and simulate the style of a given sample. Generative adversarial neural networks exist in order to create new works of art in the style of other artists. GAN-programs simultaneously use two models: a content generator and a discriminator which evaluates it. By using such services, you can draw the face of a person or the face of a dog, or even bring the interior of a room or landscape to life.
However, the most popular AI products today use language models. The tool generates images from a text query which can consist of a huge number of words. You can significantly change the result just by deleting or adding a single word. Today there are even platforms where you can buy query-prompts, in order to get an image in a certain style.
Developers “teach” AI image generators to find connections by training their algorithms from a huge number of pictures and text descriptions. They also practice creating an image from a set of random points with gradual improvement, noise removal and the maximum approximation of the result to the given keywords.
Unfortunately, neural networks can’t yet claim to be capable of creating promotional materials, but attempts are just beginning. For example, Rehab managed to generate a two-minute promotional video about Deer Bob using three AI tools. Chat GPT was used for the text, Midjiourney for the image, and Soundraw for the sound. The designers set AI the task of creating a video on a given topic in 5 days. All they needed to do, except for setting the theme, was to upload the results from one platform and transfer them to another for the next task. We think the video was quite spirited (in terms of product that the algorithm worked on, not the person himself).
Can AI create “forbidden” fruit?
When speak about AI potential, we have to broach the element of censorship. As in most tools, the developers took this matter into account. OpenAI DALL-E 2, Google Imagen and Midjourney and many other media content generators will not help you create scenes of nudity, violence, or realistic faces of famous politicians. Traffic purchasing specialists who often channel onto forbidden verticals must have heaved a sad sigh. But not so fast….there’s still hope. There are tools without such restrictions. Stability AI – the developers of the Stable Diffusion, argue that this model can do everything without any filters. We can assure you that if you do global research, you will find many such programs and even be able to choose the one that’s most easy to use.
What are the disadvantages of AI creatives for advertising?
- First, it is already obvious that over time, without training and improvements, AI products will become repetitive and eventually lose their unique quality. In order to prevent this from happening, artificial intelligence databases must be regularly updated.
- Secondly, algorithms are rarely able to come up with creatives that convert. We all know that an advertising creative is not just a picture related to the subject of the offer. It is the result of the analysis of the target audience, part of the strategy that needs to be tested and optimized.
- Thirdly, one noticeable problem of AI tools in the absence of emotions on people’s faces. If we’re talking about creating advertising materials, then emotional faces are often required (the joy of a large win at the casino, the pleasure of using 18+ products, the shock of a miracle cure, etc.)
- Fourthly, for the time being neural networks seem to cope well only with the generation of futuristic landscapes and objects, fictional creatures and space subjects. When it comes to the human body, they often have problems with fingers, and the face of a model. Free-of-charge tools are especially bad at this, and the problem is relevant for both realistic images and vector graphics.
This is the sort of result you can count on, if you decide to generate a similar image in dream.ai or deepdreamgenerator.com.
Which verticals are suitable for neurocreatives?
In their attempts to keep up with trends, media buyers and marketers have already tested neural networks to create advertising materials. They came to the conclusion that neurocreatives are best found in astrology offers, psychology, cooking, gambling, online dating, games (including adult). In general, for adult channelers, image generators are a real gift of destiny.
The following images are the fruit of the “imagination” of dreamstudio.ai.
Unfortunately, futuristic landscapes and abstract visuals, which artificial intelligence has so far best coped with, are not suited for the promotion of nutra products or e-commerce, where you need to demonstrate the item to be sold.
Finally, some interesting information for reflection: on March 22, an open letter was published on the Future of Life Institute website, in which Elon Musk, Evan Sharp, Yoshua Bengio, Stuart Russell, Berkeley, Steve Wozniak, Yuval Noah Harari, Andrew Young and a number of scientists, experts and businessmen demanded that all processes for teaching artificial intelligence be stopped. Everyone who signed the letter (the list of signatures continues to be supplemented) is extremely concerned that AI may get out of control and launch irreversible processes that might change the history of mankind. It is unclear what lies behind this: a concern for the future of civilization or something else?
We have to admit that image generators have become more than just a toy. Will the AI-trend lost its relevance or will become a rival to humans? Time will tell. However, we feel that this is just the beginning of something very big. AI product developers clearly still have a lot of work to do. There are many gaps to be filled before they can start replacing specialists. For example, neural networks need to be taught the more subtle points – such as context, quality and consistency of elements. High hopes have already been set on AI algorithms. The latest release of Midjourney proves that the speed of progress is simply astonishing.
The Near Future
It is most likely that neural networks will soon “learn” to analyze texts and visuals of advertising creatives and predict CTR in accordance with the specified parameters of the target audience. The next step will be AI’s ability to make recommendations for improving visuals and texts. Finally, it will be able to generate great converting creatives, analyze and optimize them. Then designers and marketers will only have to set the tasks. Well, we’re waiting. In the meantime, we return to the harsh reality: we have to study the target audience, accumulate experience, switch on our intuition and get on with the work 😉