Why AI images have a reputation problem

Everyone's seen the six-fingered hand. The face that's almost right until you notice the eyes don't quite track. The skin texture that reads as real for half a second and then looks like wax. Most of that was a 2023 problem. Current models are far better at it, though they still fall apart on crowds and on long stretches of text, and hands in a busy frame are still a coin flip. The reputation outlived most of the flaw. It still walks into every conversation about using AI images for business, and it's the first thing you end up arguing against.

The bad results you do still see mostly come from lazy prompts. Someone types "professional business team meeting" into Midjourney or ChatGPT's image generator (DALL-E was the old name for the same thing) and gets back exactly what you'd expect from three words of direction: a generic, slightly melted approximation of a stock photo. The model isn't broken. The input was thin.

Specific, constrained prompts produce better output. Vague prompts produce vague, uncanny nonsense. Most of the difference is right there in the prompt.

Commercial use rights for AI generated images

Settle this before you generate anything you intend to publish, because the rights come from your plan, not from the image. Start with the files you already have. Output generated under Midjourney's old free trial carried a noncommercial Creative Commons license, so anything from that era may not be licensed to go on your service page even though the file is sitting on your desktop.

Then check your own revenue against your plan. As of this writing, Midjourney ties commercial ownership of your images to company size: above a million dollars in gross annual revenue, you need Pro or Mega, and the entry-level plan no longer covers you. That threshold applies to the whole company, not to the person holding the login, so a landscaping company doing north of a million a year is exactly who gets this wrong. Midjourney has moved that number and those plan names before. Read the live terms page rather than any figure you saw in a tutorial, including this one.

Neither of these tools is expensive. Entry plans for Midjourney and ChatGPT both sit in the ten to twenty dollars a month range, with the higher commercial tiers well above that. Check current pricing before you budget for it, and count the time as well as the subscription: the subscription is rarely where the cost of an AI tool actually lands.

There's a second legal wrinkle nobody mentions in the tutorials. In the US, an image generated entirely by AI generally can't be registered for copyright; the Copyright Office wants human authorship, and the courts have backed it up so far. So the image is yours to use, but it isn't yours to stop somebody else from using something nearly identical. For a blog header, who cares. For a logo, or anything you'd want to defend, that's the whole ballgame.

Where this actually helps a small business

With that settled, skip the fantasy use cases. Here's where AI image generation earns its keep for an HVAC company or a law office:

  • Hero images for blog posts and service pages, where the alternative is a stock photo that six of your competitors are also using
  • Social media backgrounds and post graphics that need to look intentional without a design budget
  • Concept images for proposals and pitch decks, showing a client what a finished patio or renovated lobby could look like before anyone breaks ground
  • Placeholder imagery for a new service page while a real photo shoot is still on the calendar, with a reminder on that same calendar to actually swap it out, because "placeholder" is a lie you tell yourself

That third one needs a guardrail. A render you hand a homeowner during a sales conversation functions as a representation about the work, and deceptive-advertising rules generally don't care that you meant it as inspiration. Label every render "Concept illustration only. Not a depiction of your property or of the work to be performed." Put the same line in the written proposal, and never set a render next to real job photos without an obvious visual difference between the two. If you're a licensed trade, have a local attorney read your proposal template once before you start sending renders out with it.

What these have in common is that "close enough and on-brand" beats an expensive photo shoot or a free stock photo everyone recognizes. Nobody is scrutinizing your blog's header image for anatomical accuracy. They're scanning past it in half a second, and half a second is all a decent AI image needs to survive.

One surprise worth knowing about before it happens to you: most generators write some kind of AI marker into the file, either IPTC tags or a full C2PA manifest, and the social platforms read it and attach an "AI info" label to your post. What each tool writes is changing quickly, so assume the label can show up. Harmless on a blog header. Startling on a post you didn't think of as an AI post.

How to prompt for professional-looking results

This is the part people skip. A good prompt is doing several jobs at once, and the same discipline that makes a text prompt work applies here. If you want the general version first, start with the three-step prompt framework.

Specify the style, using terms that map to something visible. "Professional photography" is thin. "35mm, f/2.8, window light from camera left, editorial photography, shallow depth of field" gives the model a look it can actually render. Camera-jargon stacking matters less than it did on older models, and focal length and the direction of the light move the output a great deal more than sensor talk like "full-frame" does. Spend your words accordingly.

Describe the setting with real specifics. "A backyard" gets you a generic yard from nowhere. "A landscaped backyard in an Arizona suburb, desert-adapted plants, afternoon light, stone pavers" gets you something that actually looks like it could be your job site. The model rewards detail. Give it geography, time of day, materials, weather, anything a real photographer would have to think about.

Handle the stuff you don't want with the right mechanic, because plain prose like "no text, no logos" is unreliable, and naming a thing can still raise the odds it gets drawn. That happens less on current models than it used to, but it's still the way to bet. Midjourney has a parameter built for this: put --no text, logos, watermarks at the end of the prompt. ChatGPT's generator has no equivalent, so describe what you do want instead. "No signage in frame" becomes "open sky, unbroken." People work the same way: rather than telling the model anything about hands, frame the shot so it never has to solve them, with the subject turned away, seen from behind, or cropped at the shoulder. That sidesteps the uncanny-face problem instead of fighting it.

Then look at four of them together. Midjourney hands you a grid of four by default; in ChatGPT you have to run the prompt four times, and it's worth doing. Judge them side by side rather than one at a time, because in isolation every render looks fine.

When to still use real photography

Real photography still wins in a few places, and two of them aren't negotiable.

Headshots and team photos need to be real people. An AI face on a bio page gets clocked. It's the wrong kind of memorable. If a prospective client googles your team and something feels off about the photos, that's a trust problem you created for no good reason.

Before-and-after job photos are non-negotiable. If you're a landscaper showing a backyard transformation or a contractor showing a kitchen remodel, that image has to be a real photo of a real job. This is the entire point of a before-and-after: proof. An AI-generated "after" shot is a lie dressed up as evidence, and if a customer ever finds out, you don't get that trust back.

The line is whether the image is illustrative or evidentiary. A blog header, a concept mockup, a social background: AI is fair game there, and usually the better economic call.

Let's talk about your site's images

Your site's images should look like your business. Right now most of them probably look like everyone else's. We build sites with imagery that actually looks like you, AI-assisted where it makes sense and real photography where it has to be real. Send us the page you're least proud of. If you're not sure what the page is supposed to accomplish in the first place, start with what a website should actually do for a service business.