AI image generators add a smile by default

Even when you ask for tired.

It's not a prompt bug. It's the training data. Stock photo libraries are full of professional headshots and lifestyle images where every face is grinning. When you train a model on millions of those, "person" stops being a neutral category. It becomes "person who looks pleasant in a photo".

So the model averages every face toward that default. Tired chef at 11pm? Smiling tired chef. Worried real estate buyer? Hopeful worried buyer. Annoyed driver in traffic? Mildly amused driver.

The longer-prompt workaround doesn't work for long. You add "serious expression, no smile, weary, eyes downcast" and the model gives you a closed mouth but the eyes still have that subtle warmth. Because the training won the war before your prompt even started.

What actually works is a different workflow. You generate the base image, then you run a second pass with a model that specifically handles expression isolation. Or you use a smaller, more controllable open-source model where you can override default biases at the weights level instead of fighting them in plain English.

Most agencies don't do this. They run one prompt, ship the smiling tired chef, and call it a day. That's why every AI restaurant photo looks like the staff just won a lottery in the middle of service.

If your AI photos all look weirdly happy, that's why.

The fix exists. It just takes a workflow, not a prompt.

If your team uses AI for visual content and the results look uniformly cheerful in a way that doesn't match your brand, that's the workflow gap. Get in touch and I'll show you the override path.