How to Get Professional Results with Gemini AI
Don’t worry about writing prompts for the Geminis, here are some tips to help you create results that are professional-looking.
My First Attempt Didn’t Look Professional
To be honest, when I was first trying to get Gemini to make me a “professional headshot”, I just typed “professional headshot, please”. The result I received was a wax figure in front of a grey gradient. Flat light, weird plastic skin, dead eyes. Not quite Linked In quality.
I had to ponder upon it for a long time to understand the logic behind it. It wasn’t the tool, it was the way the tool was used. Of course, I was talking to it like a search engine, rather than a photographer.
But Gemini’s image models really can create images that would look like photographs of a studio shoot, as long as you provide them with the same information that a real photographer would provide to set up a shoot.
Rudimentary asks yield rudimentary response. You’ll find what you ask for when you ask for it. That’s truly the secret.
So this is what I have discovered for myself to get Gemini to stop playing guessing games and actually create a professional looking image.
1. Do Not Describe the Subject. Begin to Talk About the Shoot.
People’s natural inclination is to say what they want: a person, a product, a scene. It’s essential but it’s not enough. It’s not about what you are taking a picture of, it’s about what they’re taking the picture of.
Rather than using the term “a picture of a woman in a business suit”, use the more descriptive terms that a photographer would use to describe a job to a helper:
- What is the subject?
- What is the action?
- Where is the action taking place?
- How is the subject lit?
- What kind of camera and lens did the photographer imply?
The more the prompt looks like a shot list, the less time Gemini has to rely on default, cookie-cutter responses.
“corporate headshot, three-quarter angle, soft window light from the left, shallow depth of field, neutral gray backdrop, shot on an 85mm lens, subtle catchlight in the eyes”
is going to outperform
“professional headshot”
every time.
Not just a longer it’s for a longer — each of the phrases in the it is working a job.
2. The Magic Happens When You Illuminate Your Language.
When it comes to levers that affect the needle more than anything else, lighting vocabulary is the number one lever.
It’s clear that AI image models have been trained on a vast amount of captioned photographs, and photographers are very precise about their description of light, so the model is attuned to that precision.
Terms like:
- “soft diffused lighting”
- “golden hour side light”
- “Rembrandt lighting”
- “harsh dramatic shadows”
aren’t just decorative flourishes.
In fact, they are the ones who guide the render.
Most models take a default stance of flat, even, slightly overexposed lighting when left vague, the equivalent of the fluorescent lights in an office.
Often the single greatest perceived “professionalism” is the naming of the lighting setup you desire.
3. You Don’t Need Just a Scene, You Need a Camera.
This one I found most interesting as far as it’s output changes simply by saying the camera and lens parameters (no actual camera here obviously).
All phrases suggesting that the image was shot and not created are pushing towards the image being shot, not created.
It’s a quick jump into the visual code of real photography that has been acquired by the model from millions of captioned photographs!
4. Explain What You Don’t Want to Change.
If you are uploading your photo and asking Gemini to restyle or enhance it (not create a new one), the following part of this post is most important:
Clearly communicate what you want to keep the same.
If your editing prompt is:
“keep facial features, skin tone, hairstyle, and identity exactly the same while applying the requested edits”
then that should become a reflex line in every editing prompt you write.
If it doesn’t, the model can drift, a slight change in a face, its expression, and so on, or an “improvement” of something you never asked it to do.
Preserving identity is not automatic – it is a specific request that must be made.
5. Create the Prompt in the Form of a Formula, Rather Than a Sentence.
After a little bit of writing, a pattern will begin to develop.
Don’t overcomplicate the sequence of the wording; a strong photo prompt is likely to contain one or more of the following elements:
- Subject and action
- Style
- Lighting
- Background
- Camera details
- Quality descriptors
In the context of product photos, it becomes even more literal.
This is like a checklist:
- Remove the background
- Correct the lighting
- Improve color accuracy
- Increase sharpness
- Reduce reflections
- Place the product on a clean white background for a professional commercial quality image
This itemised approach is more beneficial than a single flowing description for e-commerce prompts.
6. Balance Your Requests for Resources and Support. Avoid Asking for All at Once.
A habit that needs to be broken is to place five or six prompts and hope the model performs them all well.
It usually doesn’t.
Get the lighting right first.
Look at the result.
Then request the colour grade.
Then sharpen.
It’s better to make many small, specific changes over time, than to make one big “do everything” change, because it’ll catch a trend you are going down before it gets three times as far.
7. Use a Good Source Image.
In fact, when editing, the quality of the input is important.
A high-resolution image with a fairly simple background provides Gemini with much more to work with, and it reflects in the output.
It’s like giving someone a dirty photo and asking them to clean it; that is the kind of work that should be the responsibility of the model, not the photographer.
I test every AI prompt before publishing — no guesswork, just results that work. When I’m not prompting Gemini, I’m hunting for the next AI trend worth sharing.
