AI Generated Images vs. Stock Photography: A 2025 Reality Check

In Short

• AI isn't the enemy of photography; it's a different beast entirely. While traditional stock photography is dying a slow, beige death, real photography of your people and products is more valuable than ever.
• Trust is the new currency in 2025. With the Australian Government's latest guidance on AI transparency, using fake images for real products isn't just lazy—it's potentially misleading conduct under the ACL.
• The tools have finally grown up. We've moved past the six-fingered nightmares. Models like gpt-image-1 and the new aggregated Adobe Firefly are producing photorealistic results that actually look human.
• Context is everything. AI images are perfect for attention-grabbing blog headers and abstract concepts (like my denture site example), but terrible for health sites where authenticity builds trust.
• The workflow has shifted. I'm currently bypassing the big platforms to generate custom stock via API, giving me control that generic stock libraries never could.

---

I am a professional photographer. I am also a web developer. By all rights, I should hate AI generated images.They are, technically, stealing the bread off my table. Every time I generate a prompt, a stock photographer somewhere loses a royalty check. It represents a massive conflict of interest for Dygiphy. But I’m not going to lie to you—or myself—about the reality of website design in late 2025.The game hasn't just changed; the board has been flipped over.For years, AI images were a parlour trick. A bit of fun. You’d ask for a "business meeting" and get five people with melted candles for faces and a hand with seven fingers clutching a coffee cup that defied physics. It was noise. Useless for professional work.Then OpenAI dropped the gpt-image-1 range a few months back.That was the moment. The threshold. Suddenly, we weren't looking at "digital art" anymore. We were looking at photographs of people who don't exist, standing in offices that aren't real, lit by sunlight that never shone. And it was terrifyingly good.

The Trust Paradox

Here is the problem: AI generated images look too perfect now. They are hyper-real. And from a visitor's perspective, this creates a weird friction.On one hand, they provide incredible visual interest. I can conjure a specific scenario—say, "a golden retriever wearing a safety vest on a construction site in Perth at sunset"—in seconds. Try finding that on Shutterstock. You can't. But on the other hand, they impact trust.If you visit a law firm's website and the lawyers look like movie stars with poreless skin, your brain triggers a warning. Fake. And if the lawyers are fake, is the advice fake? Authenticity is the casualty of perfection.This isn't just a vibe anymore; it's a legal minefield. The Australian Government's recent guidance on AI content is pushing heavily for transparency. They are explicitly tying the lack of disclosure back to misleading conduct under the Australian Consumer Law (ACL). If I put an AI generated doctor on a medical site, am I misleading the patient? In late 2025, the answer is leaning towards a hard "yes".

The Tool Landscape: A Messy brilliance

There is so much noise out there. You have likely heard of Midjourney. In the early days, you had to use Discord to access it, which was a user experience nightmare—scrolling through thousands of other people's anime waifus to find your own image. Now they have a web app, and it's… fine. But in my experience, Midjourney still leans too heavily into the "fantastical". It wants to make art. It wants drama. I don't need drama; I need a receptionist answering a phone.The real shift has been with Adobe.Early on, Adobe Firefly was—let's be honest—bad. Safe, but bad. The people looked like plastic dolls. But Adobe did something smart. They pivoted Firefly from a proprietary model into an aggregator. Now, Firefly works as a hub. You can run a prompt and see results from their own Image Model 5, but also side-by-side with models from Google, OpenAI, Flux, and Runway.It’s brutal for the competition. You can see the winners instantly. Firefly might actually emerge as the standard tool for creatives simply because it stopped trying to be the only model and started being the best interface for all of them. It’s making a real dent in traditional stock photography.

My Current Workflow (And Why It Matters)

Despite Adobe's comeback, I’ve gone rogue.As of right now, I am almost exclusively generating web stock using gpt-image-1-mini. I don't use the public chat interface; I use an in-house tool we built that hits the OpenAI API directly. Why? Control. The public models are too "safe", too filtered. Via the API, I have the flexibility to tweak parameters that simply aren't available on the consumer web versions.It’s the only way to get images that are usable for business rather than just "cool to look at".That said, the Gemini "nano banana" models (ridiculous name, serious tech) are creeping up. Their rendering of human skin texture is arguably better. Ask me again next week, and I might have switched.

The Medical Case Study

Let me give you a concrete example of where this wins. I was building a site for a medical clinic recently. Do you know how hard it is to find stock photography of elderly people smiling that doesn't look depressing or clinically terrifying?Impossible.I used AI. I generated images of older Australians, outdoors, active, smiling naturally. The key was to make them look representative rather than real. We weren't claiming "this is Bob, a patient". We were communicating a feeling: vitality. The creative flexibility saved the project. We got images that simply did not exist in any library.

The Verdict: Co-existence, not Replacement

The mutation issues—the six fingers, the melting eyes—are largely solved. Hallucinations still happen, but prompt engineering fixes 90% of them. So, is photography dead?No.If you are selling a product, you need a photo of that product. If you are a team of accountants, you need photos of your team. AI images are filler. They are the garnish, not the steak. They supplement the real content to create greater beauty and context, but they cannot replace the truth.We are in a weird transition period. AI images will better suit some sites than others. Blogging sites? Go nuts. They need clicks, not dollars. But for high-stakes businesses, the human element is the premium product.

Decision Tool: AI or Real?

Struggling to decide what to put on your page? Here is my mental checklist:

• Is it a specific product for sale? Use a Real Photo. (Never trick a customer on the goods).
• Is it a staff profile? Use a Real Photo. (People buy from people).
• Is it a blog post header? Use AI. (It’s cheaper, faster, and more engaging).
• Is it an abstract concept (e.g., "Cybersecurity")? Use AI. (Stock photos of hackers in hoodies are cringe).
• Is it for a health or medical service? Use Real Photos for procedures/results; AI is okay for generic "lifestyle" context if clearly labelled.
• Does the budget allow for a photoshoot? If yes, Real Photos always win on authenticity. If no, AI is now a better option than cheap, low-res stock.