The honest case for AI-generated imagery on small business websites

What AI imagery is good at now

Mood and atmosphere. Hero backgrounds, section dividers, abstract concept illustrations. Generated to your brand colour palette in seconds, unique to your site.

Industry-specific scenes. Generic situations (a plumber installing a hot water system, a chef plating food, an accountant at a desk). The AI versions are better than stock because they’re tailored to your brief.

Article illustrations. Blog post hero images, abstract concept art, decorative graphics. AI handles these brilliantly and uniquely.

Iterating on briefs. Need 10 variations of the same concept to pick the best one? Stock photographers won’t do that. AI does it in 30 seconds.

What AI imagery is bad at

Real people. AI faces are getting better but still have artefacts. The uncanny valley is real. For team pages, About sections, anywhere visitors will scrutinise faces, use real photos.

Specific products. AI can’t generate accurate images of YOUR specific products. It can generate generic versions of similar products. For a product catalogue, use real photography.

Specific places. “Newcastle harbour” from AI is a generic Australian harbour, not the actual one. For location-specific imagery, use real photos.

Hands and fingers. Year-old AI models still struggle with hands. Image of someone holding a tool, three fingers. Improving fast but still inconsistent.

The economics

Stock photography: $20-50/month subscription for limited downloads, or $10-50 per image. For a 20-image site, $200-1000+ over the project.

Custom photography: $1,500-$5,000 for a half-day shoot with a professional. Worth it for hero shots and specific deliverables. Not worth it for every section.

AI imagery: Pennies per image. For a 20-image site, the AI imagery budget is often under $5. The savings go into the time spent picking the right ones.

The ethics question

AI image models were trained on existing imagery, much of it without explicit consent from photographers. This is genuinely contested ground.

Where we draw lines: licensed AI tools only (Imagen, DALL-E, Midjourney with commercial licences). Never imitate specific living artists’ styles. Never generate content meant to deceive the viewer about authenticity.

Where reasonable people disagree: whether the training-data question makes AI imagery off-limits entirely. We think the practical alternative (stock photography of stock models in stock situations) is also problematic but in different ways. Use judgement.

How we handle it on client builds

Default mix: 60% real photography (the trust-critical stuff), 35% AI-generated (the fillers and illustrations), 5% stock (when AI doesn’t fit and we don’t have a real photo).

For new clients without existing imagery: AI carries more weight. We’ll generate a coherent set of 20-30 images tailored to the brand brief. Then we encourage them to invest in real photography over time and gradually replace.

For clients with strong brand photography: we lean heavily on theirs and use AI sparingly for filler.

Most clients are fine with this. A few aren’t, and we respect their choice and use stock instead. Both approaches are defensible.