What benefits actually come with image generative AI, well it does not necessarily automate the whole process of image creation. In fact, it does not automate anything, yes it can generate an image within 5 seconds, but it will be as good as typing the same thing into Google and picking a random image. About a year after the release there should be new models that will not only speed up the process, but also have much better interpreters and classifiers, making them universally accessible and usable. Since everyone will have access to it, it’s going to be terrible, so get ready for a flood of low quality images all over the internet.

What it gives

A lot of people have a misconception about this, the single image creation won’t get as much improvement as it seems, but after developing Prompt it’s almost no effort to create 1000 similar images with slight variations. So the benefits really come from scalability, meaning that whatever business you are going into should be centred around mass-producing flexible looking images rather than a specifinc one tailored for the story behind it.

Simple Business

1. Concept art – Probably the first to be affected, artists will be able to generate and rather vague representation of their will to review the concept and collect some examples. Concept art is rather deliberate, so there will be some time before it is fully implemented, but only with significant manual user input.
2. Clip art – Usually digital stock assets, sold to other graphic designers to use in their product. Etsy is filled with it, as it is a digital product it scales indefinitely, and cost $0 to store it. The byproduct of that there will be oversaturation in prtint-on-demand utilizing it.
3. Posters – This one require about 300DPI, with doesn’t mean much as DPI is relative unit, but in translates for ~2500×3500px for an A4 paper size and 5000×3500 for A3, with is really out of scope for those 512×512 models. With November 2022 came hig.res-fix and advanced title upscaling meaning resolution is no longer the bound.
4. Stock images – It’s always a pain to search for them, being able to generate them from a short text description seems like heaven-sent, but the quality is still far from being able to use them in any non-professional production. YouTubers will probably be the first to use it, even if it’s not yet ready, alongside filler images for SEO blogs.
5. Art design – It will be used for products around art with a lot of creative freedom. For example, tarot card decks, playing cards are a great product, you’re basically printing images on paper and charging the buyer $60 for it. There’s 78 different cards in a deck, so as you can imagine, it was quite expensive to create the art for that by hand. It’s obviously not a golden opportunity, the AI art is quite good for it, but the market is quite saturated, so you need to do some targeted marketing campaigns.
6. Art services – First, Fiverr will be flooded with people offering low-level concept art, and then image-to-image services, i.e. translating a picture of your cat into a Van Goh style, or some such silly thing.

Advanced business

AI “Instagram” model

There seems to be a lot of fuzz about this, especially in the space of YouTube business gurus, but as always with these people, they will throw out all their ingenious strategies without actually knowing anything. After all, that’s their business model, selling glimmers of hope to the not-so-bright.

1. Ultra-realism – It’s hard, achieving photorealistic quality with an AI is not an easy task, much harder than any abstract art style.
2. Personal connection – As bad as it sounds, the reality is that an average customer for such content, develops an emotional bond, well, we have somehow evolved into that, who could predict internet back then… Even if in regular case there is another person responsible for connecting with the audience, the person it’s all about is a real person, so it’s a bit easier to start assimilating with them, than with someone fully artificially created for pure cash profits.
3. Legality – Bottom funnel sites are for generating money, as for that you need to verify, not only with legal ID, but also with bank. These sites even if they are aware then they don’t endorse externally managed accounts, so the actual model needs to verify with their real identity, as far as I’m aware there is not really an option yet for “digital figure”. This means that ‘personas’ that only exist digitally will not pass on the identity of the account manager. Although I’m rather positive this will change in the future, looking at the current virality of such businesses, there will be plenty of demand and with those site draconian fees that’s an obvious thing to do.
4. Video centrism – This is an area where video dominates by a huge margin and you are not going to do it with AI. The complexity of video control is increasing exponentially, so it is hard to say how long it will be before something good enough to fool others emerges.
5. Face cohesiveness – Purely a technical issue, most likely to be solved in the future. Getting the face of the same person is much more complicated than it sounds, especially for the purpose of fooling people who are exceptionally good at recognising them.

So what’s the conclusion? The benefits of AI are almost non-existent in this space, as it actually creates more problems than it solves. AI influencers will be the thing in the future, as it will be paired with business strategy for such agencies, whose biggest struggle at the moment is centred around the models/creators themselves. Not only are they problematic to work with, but they are also quite expensive and demanding. Having an almost free digital persona with no real morals or second thoughts seems too lucrative for them not to invest in this technology.

The gurus who recommend it often forget to mention that it’s an extremely competitive space, and you’re bound to fail if you expect SM growth to come organically. It requires advanced campaigns and strategies, even for the best looking real model. Expecting this to happen on its own for an AI generated mediocre looking model is as likely as winning the lottery.

Visual story telling

Comics, VN games, manga, slideshow videos… You name it, there are plenty of media that use images to tell a story.

Information engineering

You may not know it, but spreading false information on the internet is also a job, not a badly paid one, as for the morality of such a thing, I leave it to you.

1. How difficult is it to detect an AI-generated image?? – Extremely difficult, in fact most people have no chance against it. “But I’ve seen a lot of AI images, and they’re so obvious to spot!”, yeah, and you’ve only spotted the bad ones…

   How do you read the news? On your 6" smartphone? On the move? Scrolling through the headlines? You see, not many people are doing pixel inspection on their calibrated 30" 4k monitors, in pitch dark rooms, using digital image forensics tools. A small, dirty screen with reflections all over it while standing on a shaky train with the sun shining in your eyes.

   It’s not a perfect forgery, of course, but it’s getting harder and harder to spot, and even I, who consider myself an expert, can’t see that it’s been created using diffusion models.

2. Spreading the information – Before social media, someone was responsible for the information they shared, now they’re not necessarily. Conscious and not easily fooled users will share what they want, and are 5x more likely to do so if the information correlates with what they believe. So, yes, you can distort reality to suit yourself without any consequences.