Confidently Incorrect: The Risk of AI Images for Industrial Markets

Scriptorium Team
Apr 9
3 min read

AI images make a striking first impression. Generative models like Midjourney or Stable Diffusion excel at lighting, composition, and mood, which are the characteristics that make an image feel professional. When your content needs supporting visuals, it’s tempting to render a campaign-worthy scene in seconds.

Unfortunately, AI fails at something your audience cares deeply about: technical accuracy.

AI can render a perfect sunset but struggles with machines. For example, a single-serve coffee maker with a gigantic power button and no other options:

A single-serve coffee machine with a power on button, but nothing else.

Sometimes the errors are fun and harmless. But AI’s machine-rendering weakness presents a real business risk, especially for those serving industrial and technical markets. The people seeing your images might include project managers, engineers, health and safety leaders, and technicians: people who are great at noticing things that don’t fit. A beautiful but mechanically incorrect image signals that you don’t understand their work at all. And if the image depicts safety-critical components, you risk your reputation.

Why AI Struggles With Machines

AI models don’t understand objects as engineered systems. Instead, they learn patterns based on being fed tons of examples. As a result, AI knows that all coffee makers have buttons, but it doesn’t know why.

Organic imagery like skies, landscapes, or human faces is forgiving because we tend to accept inconsistencies as natural variation, and it’s harder to spot geometric errors.¹

However, machines are constraint heavy. Parts must connect, seals must be flush, joints must align. Yet AI consistently struggles with maintaining structural integrity, as well as creating rare or novel concepts.² It can produce images that seem plausible but are mechanically impossible, especially with specialized industrial equipment.

AI Tries Safety Gear

For example, safety equipment is a subject that demands precision. AI will render great-looking pictures, but they’ll be full of subtle and even dangerous errors. Have a look at these AI-generated images of personal protective equipment:

Fall Harness

The strap configuration is reasonable, but the rubber coil is more suited to bungee jumping, not fall prevention.

Emergency Eye Wash Station

Based on where the spouts are sitting, it’ll be tough for water to reach your eyes (not to mention the drain is halfway up the sink).

Manual Resuscitator

These emergency devices are used for providing air to a person who isn’t breathing. It helps if the mask is the right way around.

Even if this imagery was meant for decoration, not instruction, the inaccuracies are still alarming. Realistic-looking but wrong AI outputs are partly what fuel debates about AI needing built-in safeguards.

A computer chip background with the quote, "he people seeing your images might include project managers, engineers, health and safety leaders, and technicians: people who are great at noticing things that don’t fit. A beautiful but mechanically incorrect image signals that you don’t understand their work at all. And if the image depicts safety-critical components, you risk your reputation."

When AI is Fine, and When It’s a Risk

If you need a sunset, a background texture, or a conceptual “mood” image, AI can be a cost-effective option. If you need accurate depictions of machines, tools, controls, or equipment, AI is a shortcut to avoid. Risks to your business include:

Loss of credibility – Technical audiences notice errors, and you’ll look either inattentive or amateur.
Brand damage – AI inaccuracies signal carelessness to markets that value precision.
Liability – Misrepresenting equipment with safety or emergency features can create reputational or legal problems.

The rule of thumb is to use AI for atmosphere only. If the image needs to be right, use real-world sources.

Scriptorium specializes in technical writing and graphic design for health and safety procedures – exactly where you need human oversight.

#BusinessCommunication #Documentation #GraphicDesign #ProjectManagement #TechnicalWriting #IndustrialImages #IncorrectAI

Sources:

Schoonderwoerd, T. A., & Kondrak, G. (2025). A mixed-methods approach on human perception of AI-generated images. Frontiers in Artificial Intelligence, 8, Article 1707336. https://doi.org/10.3389/frai.2025.1707336
Zhang, T., Wang, Z., Huang, J., Tasnim, M. M., & Shi, W. (2023). A survey of diffusion based image generation models: Issues and their solutions. arXiv. https://doi.org/10.48550/arXiv.2308.13142