Banner Creation in 2023: Is It Possible to Create High-Quality Ads Only Using AI?

If I were given a dollar for every mention of AI in my feed for the last several months, let’s just say our marketing team would be flamed up with a happy new Porshe 911 owner.?

For the last decade, AI-based technologies have gathered their might to fully knock us out with just how accessible they’ve become. Whether it’s text, images, music, or even video production, it seems like for every media form; there is a handy algorithm ready to help you create it.

So, since this is a newsletter about advertising, we won’t obsess over “Machines are gonna take our jobs!1”, or “AI lacks the human touch!!” kind of claims today. As a hammer to a blacksmith, we’ll observe AI tools in free disposal to automate an ad creation process.

We’ll see how far AI has come in this regard, learn a bit more about generative models, and just have fun with some crappy results.?

Without further ado, let’s rock!

How Does Generative AI Work?

To start off, let’s take a look at how generative AI works so there’s a better understanding of what beeps in that boop machines. Don’t worry; we’ll try to make it as clear as day. (Skip this to the next heading if you came for the lulz specifically)

Firstly, artificial intelligence is a discipline that creates automated technology to solve tasks like humans. Think of it as chemistry or math.??

Machine learning is a subfield of artificial intelligence that focuses on training systems with data. Its main goal is to help machines learn without explicit programming, sort of like humans do.?

For that, it uses two of the most common types of ML models: supervised and unsupervised. The supervised model comes with labels – certain characteristics that are important for learning. This model is “supervised” with information during the training process to learn from past mistakes and produce future values.?

For instance, in this case, the supervised model predicts the future tip based on if it’s a pick-up or a delivery.

On the contrary, unsupervised models don’t have any labels. Those give the machine raw data expecting it to create its own patterns, thus separating data into groups.?

Like “Does the number of years in the company correlate with the income?”

Generative AI is a subset of yet another subset called Deep Learning. The point of deep learning is to teach a machine with a small amount of supervised and a large amount of unsupervised data, a.k.a. semi-supervised data.

This process resembles how the human brain works. We initially understand basic concepts through examples and later utilize this knowledge to categorize and comprehend new information.?

And it’s not a coincidence.?

The architecture of artificial neural networks is inspired by our brains. Just like there are thousands of interconnected neurons in our heads, there are thousands of nodes in a network. When visualized, their structure is similarly intricate and complex.

Generative AI, utilizing these artificial neural networks, leverages the aforementioned methods to learn. For now, the whole point of generative AI is that with the help of generative models like GPT, LaMDA, and PaLM, the machine can help us create new content based on the existing one.?

5 Ad Generation AI Services: Our Experience

Ok, now that you know how generative models work, the operational aspect of AI-driven tools below won’t be that alien to you. Now it’s time to jump to the fun part.

We’ll try using some ad-generating services to create banners to advertise Epom white-label DSP. We’ll use our site as the landing page and use our YouTube channel for video/audio assets.?

The human intervention (mine, specifically) will be minimal; let’s just provide AI with information and see how far it gets us.

AdCreative.Ai

AdCreative.Ai is one of the most well-known platforms for creating ad banners, so we decided to check if it’s worth the hype.?

The service we are using promises to generate ads with minimal user input. After setting up an account and entering the brand's name, logo, and other pertinent details, users can select from a variety of ad formats to experiment with.

Our first attempt to create something decent was the large leaderboard ad.

I copypasted the project description from our DSP page, and as for the other text… You don’t need any, tbh. The neural network handles CTA, description, and slogan just great.

For the background image, I put some random screenshots of our product’s dashboard, and as a result, we get about a hundred AI-generated results.

Don’t mind the transparent “Ad Creative” on each result; once you download it, it’s gone. Btw, here’s my personal favorite.

There’s also an option to combine two background images into one creative, so I happily decided to test it out. This time “Social Post” was chosen as a format, and the results were mostly OK.

Yeah, I messed up a bit with choosing the background picture that had text on it, and the AI doesn’t consider that.?

Still, just thinking about the fact that it took me like 5 minutes to create all of these, AdCreative.AI gets a positive mark on our test.

Pencil

Pencil is the next ad creation machine on our list. At first glance, it seems to offer the same functionality as AdCreative.Ai, producing ads in batches.?

One notable feature of Pencil is its ability to directly import brand or product descriptions from a URL. This didn’t work for our brand for some reason, but it got the product info anyway.

Then the AI learns about the brand based on the assets you provide. We supplied it with several product pictures, videos, and logos. Let’s hope that’s enough.

Once the AI completes this phase, we define the brand goals and all the other necessary details. The generated results are presented in the form of a short video (with a choice of 5/10/15 second duration) and a static image.?

So the results were… well, look for yourself.

The video didn’t render, however. Maybe, it’s for the better since the AI marked the final result as “Loser” with a poop emoji, which is heartbreaking.

Welp, the results are somewhat mixed. The rendering took a long time, technical issues arose, and despite the ample assets provided, the model didn't offer anything outstanding.?

BHuman

Remember deep-fake? BHuman works similarly. You upload a video of a person, and the machine learns from their speech and movements, generating audio or video content based on this data. Not quite clear? Keep reading, and prepare to “have your mind blown.”

The site immediately welcomes you with an AI-generated response from the company’s founder, a handy touch.

To test BHuman, we used a video of our CMO, Lina Lugova (cringe alert: apologies in advance, Lina). What’s peculiar is that the video is already edited; you can watch the original here.

So, what’s happening here? The system provides an automatic transcription of Lina’s speech, and I could modify most of the variables to make Lina say something different, creating a personalized experience. In this example, I assign the Name tag to “Lina,” the Company tag to “Epom,” and the Tech tag to “DSP.”

Then I upload a CSV document containing 3 variables for each of those tags. In theory, this should result in three videos, each reflecting these variable changes. In practice, however, your humble servant completely lost it upon seeing the results.?

The network understands what and how to say but does it so artificially that there’s no trace of human speech left (which is very funny, tbh).

Words are mostly irrelevant; just check the results out for yourself here.

The AI performance would probably improve with time once the service gets more data to work with. Right now, at least, BHuman can only improve your mood.?

BrandMark

The logo is a simple yet very complex challenge for the designer of an ad creative, and it’s absolutely crucial for any ad campaign. Luckily, we have one for our DSP, but what if we wouldn’t?

In this case, BrandMark offers its services. The platform asks you to input the brand info and keywords and choose the color palette.

The logos are generated very quickly, and there’s like A LOT of them. IMHO, the results are pretty decent, but only a designer has enough expertise to judge. Here are some of my favorites:

The logo generation is fast, the options are nearly infinite, and the service is pretty easy to use. Go check it out, at least, out of curiosity.

Taplio

Finally, we decided to test out a GPT-4-powered tool. However, we chose not to use Chat-GPT itself, as it has become somewhat commonplace.

Taplio uses GPT-4 model to generate LinkedIn content, schedule posts, and grow your network. That’s a lot to handle for a machine; let’s see how it goes.?

Once the trial is activated, we need to configure a personal AI bot. Then, we navigate to the dashboard, which has a lot to explore.?

Bear in mind we won't be conducting any performance measurements; the primary goal of this article is to demonstrate how well these AI tools generate content.?

So, I press the button, and the service creates a load of posts in just a minute or so.

Out of all these, I like this one in particular:

“””

?? In-house vs. agency marketing: Which side do you stand on??

?? Many companies are opting for in-house marketing teams, while others still rely on the expertise of agencies. Each approach has its pros and cons - in-house brings more control and customization, whereas agencies offer diverse skills and experience.?

?? So, how do you decide which route is best for your business? Assess your goals, resources, and the complexity of your campaigns to make an informed decision.

Let's chat: Share your thoughts on in-house vs. agency marketing, and let us learn from each other's experiences!?

“””

The GPT-4 proves its power once again. The results may seem a bit generic, but it’s a good basis to work with. Taplio really packs a lot on board in terms of content (uncovered features like “Industry News” and “Inspiration”), and there’s more regarding outreach. Overall, Taplio gets a good recommendation from us; if your focus is LinkedIn – try it out.

Can AI Create Decent Ad Content?

The answer to this question is 50/50. Some of the above services deserve their praise, while others are incomplete and clunky. Either way, despite having a lot of space to improve, AI's ease and usability are impressive.

The market is booming, and in the upcoming months/years, the developers will probably sharpen their products to perfection. We hope in the future, every content creator, such as ourselves, can use these tools to create some banger ads in just a few clicks.

Even then, however, you’ll still need a platform to place all these creatives with. Epom White-label DSP has a ton of ad placement options, making it a future-proof technology for your growth. Access innovation with Epom WL DSP!