Making AI image generators work for you: apps and approaches

In short

AI image generators are a little different from AI content generators – but they work on the same principles: custom visuals created by “learning” from millions of images. For marketers, it helps attract and engage people into your content – but not all image generators are created equal. This article looks at the different apps and approaches.

Key Takeaways

• AI content and AI image generators make a great team
• Being concise and precise leads to fewer surprises
• They can also be huge timewasters if you don’t know what to do
• True creativity comes from mixing different concepts
• Dall·E, Midjourney, and Stable Diffusion are the big names …
• … but it takes human talent to get the most out of them

Introduction

If you think natural-language content generation looks like magic, AI image generation looks like voodoo. How can a machine – a machine without eyes, without any sense of art or beauty – produce incredible photo-realistic images of almost anything you can describe, often in seconds?

Fortunately, there are no dark arts involved in this form of creativity. There is, however, a huge volume of data and clever algorithms in this exciting new sector. With billions of dollars in venture capital and some of the world’s top talent going into the sector, AI image generators are now big business, able to produce art capable of winning competitions and “photographing” people who never existed. Yes, on some level it’s “copying” – but all great artists got started by copying the work of others.

Today, thousands of organisations are using AI image generation technology to add eye candy to their websites, blogs, and articles – often so on-message it’s unclear if no human was involved. And producing it often needs no more than a descriptive sentence or two.

While Contentoo is a content marketing platform, writing isn’t all we do – let’s remember “content” means more than text”. It’s everything the reader sees, including visuals. So, in this article, let’s see what AI image generators can bring to your marketing strategy, with a few fun examples.

How does an AI image generator work?

Natural-language AI content generators – like ChatGPT – work by spitting out a series of “tokens” (word bits) that are statistically likely to be patterns that make sense to a reader, aka “a piece of text.” An image generator doesn’t work like that, because images are nonlinear. (You can’t predict what else will appear in an image by looking at the top left hand corner!)

But machine learning algorithms can extract information from a set of training images, like colour, shading, tone, and “feel,” based on similarities and differences between one image and another. And that’s what happens. Train an AI on pictures of cats, and it’ll learn that most cats have a fuzzy face, green eyes, and whiskers. Add pictures of dogs, and at first it’ll mistake them for cats – but with thousands of examples, it can learn these animals are different. And when you go from thousands of images to millions, it’ll start telling apart siamese and sphynxes, poodles, and pomeranians.

While the coding is genius-level, the principle is that simple: look for the characteristics that define an image. And when you identify those traits as plain-language labels – ”a green field,” “a purple dinosaur,” “a large cheeseburger,” – you can ask the AI to generate an image containing them.

You’ll notice an issue here: simple “prompts” – the sentence you put into an AI image generation app – lead to simple (and strange) images. Especially when there are living things in the image, it’s common to feel a sense of the “Uncanny Valley,” where things look realistic but small details look odd.

As with AI content: Garbage In, Garbage Out. The more precise and detailed you can make your AI image prompt, the better the image will look. And with today’s tools trained on millions upon millions of images, those images can be very precise. Let’s look at some of their features.

Features of an AI image generator

Just as there’s an infinite number of art subjects, AI picture generators can offer a near-infinite variety of styles. If you want an astronaut in the style of Salvador Dali, go ahead. Or an Instagram influencer reimagined by Van Gogh (preferably with both ears intact). The more information you give in your prompt, the more defined your image will be.

Features extend across both technology and styles. A futuristic cityscape that looks like Blade Runner, or a hatchback car as it might exist on Mars, or a desktop PC from the Art Deco period. And you can specify a photorealistic image, a cartoon for kids, or a Japanese Manga vibe. All are possible.

In fact, the biggest feature of AI graphics generation to create AI images is that it’s not a limited feature set; it’s an infinite one, limited only by your imagination.

That said, there are some commonalities. All are web-based, to take advantage of cloud computing capacity far beyond your laptop’s. Many use the same back-end dataset, such as OpenAI, and are paid for with “credits” you can top up each month.

Most also offer the chance to iterate your image, with a range of variations on the same theme, and to edit your initial prompt to add further detail. (In case, you thought your dinosaur needs some fries with his cheeseburger.)

And the general syntax of entering a prompt tends to be the same: natural language, rather than selecting from forms and menus. Perfection is hard, but output is easy. This means playing with AI-generated pictures carries risks: hours can go by in what seems like a moment.

Types of AI image generators

Software apps for AI-generated images fall into two groups: text-to-image and image-to-image. (A “text-to-text,” of course, is AI content generation – another interest of Contentoo’s.)

Text-to-image generators

The dinosaurs above are examples from text-to-image generators: you type in a prompt and the AI comes back with a picture. (Usually, a set of similar pictures you can iterate). It’s the simplest and easiest way to get images from AI fast – almost any prompt you can think of (even complete nonsense) will produce an image of some sort.

Of course, this leads to overkill. If you remember the 80s, when Apple first launched its Mac, every office newsletter used about 50 fonts, because that’s the possibility space the Mac enabled. AI picture generators from text carry the same risk of “artistic diarrhoea,” making every image look like an explosion in a paint factory.

Image-to-image generators

Image-to-image algorithms work differently, and are often used for a different purpose. You show the AI an image of your own, and it uses it as the basis for visual generation – ”recognising” the elements of the image and producing variants to your taste.

Image-to-image generators have broader applications. Connected to other datasets, they can track down the location a photo was taken, or guess whether people are related. Graphic designers can use them to produce a set of photographs with the same lighting or colour palette, and animators can use them for storyboarding a movie. They’re not as advanced as text-to-image picture generators yet – but they may prove to have greater business benefits.

Benefits of using an AI image generator

At first, AI-generated pictures may seem gimmicky, useful only for illustrating your blog or surprising your friends. But with a little imagination, these image generator AIs become valuable tools in your marketing strategy. Here are a few:

• Ideation. A picture may or may not tell a thousand words, but it’s great for sparking a conversation – and if you’re in a meeting room desperate for fresh ideas, AI drawings can light the fire. Put a few random phrases suggested by your co-workers into the prompt line, and you’ll soon be on a roll.
• Mocking up. Industrial and product design takes time, but 99% of that time is often below the surface. If you can show what a product looks like before the design is complete, you can win hearts and minds within the company – and win budget, too.
• Exploring the possibility space. Solutions come in many forms. If you’re looking for new ways to arrange chairs in a theatre, or office colour schemes that make people happier, you can see what they look like quickly and cheaply.
• Index unstructured data. These tools are developing so fast, they’re becoming databases; “give me a picture of P.56 of the Shakespeare First Folio” is already possible, even if no such page exists as a photograph anywhere. Information is being stored in new ways – and AI is enabling them.

However, despite the infinite possibility space, look around the web and you’ll see many AI images look surprisingly “samey”: cats, purple dinosaurs, and scary women called Loeb popping up again and again. True creativity needs uniqueness. So, how do you add it?

Creating unique images with an AI image generator

Ask most artists, and they’ll say creativity is the result of different ideas coming together in a new way. It’s why some of the most engaging AI drawings connect two worlds: think of Star Trek as a 1920s Fritz Lang film, or the current rash of non-Wes Anderson films if Wes had directed them.

So, if you want your images to be unique, rather than variations, think of the most unique and disparate things your business does – and put them into the prompt in pairs. You’ll find the images have a certain originality.

Alternatively, mix your business USP with a description of something totally different. “Europe’s leading supplier of office chairs relocating its factory to a tropical beach,” perhaps, or “The most innovative SaaS company setting up a data centre in space”? You may not use the first images you see – but they’ll give you a base to iterate from.

Tips for getting the best results from an AI image generator

AI today is creating pictures of things that never existed, bringing old photographs to life, producing commercials, and jazzing up your PowerPoints. But at Contentoo, we believe the best results aren’t from AI alone, but come from mixing human and machine talent.

Accordingly, we’d suggest the best people to judge your AI pictures are people you already know – your graphic designers, art directors, and visual storytellers. People with art in their soul. Anyone can tell if an image contains blue or red – but only someone with training can say if the image is the best it can be.

There’s a place for other talents too. Your copywriter or editor (we work with the world’s top 3% at Contentoo) can help you refine and rewrite your prompts, making sure they include all information and communicate it to the application in the most effective way.

Also, swing in your brand owner or marketing director. If they’re seen by customers, a quarter can be missed or slammed by having the wrong characters in a photograph, or a strange setting that doesn’t resonate in a particular culture – so as always, build an approval process and check, check, check.

Safety considerations with AI image generators

As with AI content generation, a bad or malicious prompt can produce very bad results – think hate speech, racism, sexism, and more. In addition to making factual errors, this gets even worse when it’s in an image rather than text.

Imagine the PR issues if your AI-generated image produced a character who looked like a known criminal, or depicted crime scenes as comedy. Because they use a database of millions of images from across the web, there’s bad content in that data. Again, build your internal approval processes to catch them early.

Another area is copyright infringement. Creators are already taking legal action against the users of some AI-created images, claiming their content was used in ways they never approved. And if your image looks too much like a “real” artist’s style or substance, they may have grounds for legal action. This area of law is fast developing – don’t ignore it.

Popular AI image generators

Finally, let’s look at a few applications. The best-known are DALL·E, Midjourney, and DreamStudio. Here are their strengths and differences.

1. DALL·E 2: The visual twin of ChatGPT

DALL·E comes from OpenAI, the makers of ChatGPT, and uses the same mass of unstructured data scraped from the internet. Of these three, it’s probably the easiest to use – a straightforward natural-language prompt, like its ChatGPT sibling.

This flexibility, however, creates a few issues. Images need a lot of curating and iterating; its photorealistic images aren’t really very photorealistic. And its free version limits you to 15 or so images a month: not enough for business use.

2. Midjourney: serious power for serious realism

Midjourney takes it up a notch. Its images, particularly those with an applied “style” like movie set or photo studio, seem better-looking that other AI pictures; often the first variant is usable without further iteration. Living creatures, including humans, also seem expertly rendered.

At the moment, though, the application is only accessible via a Discord server – in effect, putting every image you create in the public domain. While you’re free to use your images, Midjourney keeps a copy of every one … creating potential duplicative and copyright issues ahead if you use them in your branding.

3. DreamStudio / Stable Diffusion

Our third choice works in a different way. The actual AI is called Stable Diffusion – ”diffusion” being the process of element-mixing that generates the image in the first place – and DreamStudio is its official API: the means you use to access the image.

DreamStudio is a professional tool, with far more choices than a simple prompt. Intuitive sliders let you blend images, soften edges, change colour palettes; upscale, reshape, and resize without losing the image elements (in effect, giving a fresh image each time). It’s a powerful product –but needs some practice to use.

The AI image generator with the best results

For our money, Midjourney offers the best option for now. While every image becomes instantly public, you still have the right to use the image yourself, and as an AI-generated picture you don’t have to worry about stealing someone else’s imagery – even the same prompt used twice will give different results. So, we’d suggest: play around with Midjourney first.

Conclusion: Use AI for your images – but use human talent too

Two years ago, AI image generation was a novelty. Today it’s racing ahead, producing realistic visuals that look as good as photographs – and advanced applications, like video, aren’t far behind. Used together with AI content generation for your text, these tools can turbocharge your content creation – letting you launch more content, in greater variety, on a faster schedule than ever before.

But in all this excitement, don’t forget the human. Your audience is human, and engaging them needs human judgement. People aren’t being pushed out of this picture; they remain the most important part of it.