Prompting Like a Creative Director

Getting better results from generative AI isn’t about luck, it’s about precision. Prompting is a creative skill in its own right. The more fluently you can direct a model, the more consistently you’ll get output that actually works.

But good prompting isn’t about stuffing a sentence with keywords. It’s about knowing what to say, what to leave out, and how to guide the model in the right direction. These tips go beyond the basics and reflect what I’ve learned from countless hours spent crafting visuals with tools like Midjourney.

1. Treat It Like a Collaboration, Not a Command
You’re not dictating. You’re co-directing. Let the model bring something to the table. If you over-describe every single element, your result might feel rigid or overcooked. Leave just enough room for the system to surprise you, then iterate from there.

Example Prompt: Dreamlike image of fox in a foggy forest, inspired by magical realism, soft light, whimsical.

2. Be Precise About What You Don’t Want
Negative prompting is underrated. Want to avoid text in your image? Mention it. Sometimes saying what you don’t want is just as important as what you do (but don’t overdo it).

Example Prompt: Portrait of a woman in soft window light, no text, no logos, neutral background.

Pro tip: In Midjourney use the --no parameter.

3. Use Real-World Camera Language
“medium format camera,” “35mm lens,” “f1.8,” “cinematic depth of field” —the models respond well to real-world technical language.

Example Prompt: portrait photo of a jazz musician under a streetlamp, shallow depth of field, 85mm, f1.4, cinematic bokeh, warm tones.

4. Stack Descriptors Strategically
Instead of cramming every idea into one sentence, think in layers. Describe the subject. Then the style. Then the setting. Then the mood. This creates structure and makes your prompt easier to parse (for both you and AI). Start simple, build, and iterate.

You can also add details such as the camera type and lens (Hasselblad medium format, 50mm), but generally I recommend starting from a more simplified baseline and building on it.

[Subject][Style][Location][Lighting/Mood][Tone/Emotion][Framing/Composition].

Example Prompt: A young Scandinavian man wearing athletic clothes, running through a sunny park, cinematic style, soft lighting, warm tones, wide angle.

Pro tip: The order of the tokens (descriptors) can have an impact on the final output (emphasis from first to last). For example, if you want the location to be the focus, place it before the subject.

5. Re-Prompt From Your Own Outputs
When you get something that’s close, don’t start over. Copy the good parts of the prompt, remove the weak ones, and regenerate. Sometimes one small change makes all the difference. Learning to prompt is really learning to edit.

Example Workflow: Original prompt too busy? “A woman running through a field of tulips, sunrise, painterly style, wide-angle shot.” Remove “wide-angle shot” and see how the framing changes.

6. Build a Personal Prompt Library
When you find language that works, save it. Build your own library of prompt snippets, style anchors, and lighting setups. Don’t rely on presets or trending prompts. Build from your own taste and experience. You can also utilize Moodboards using your own work to reference in Midjourney.

Example Entry: “Studio portrait, Rembrandt lighting, Canon 50mm look, neutral color grade — use this as a base for portraits and tweak per project.

7. Understand the Model’s Biases
AI models are trained on massive datasets that come with defaults. Often Eurocentric, glamorized, commercial. If you want something more real, diverse, or stylized, you have to ask for it clearly. Otherwise, the output will default to what’s been seen the most.

Example Prompt: Portrait of an elderly Southeast Asian woman in traditional clothing, photographed in natural light, Fujifilm look, no western facial features.

8. Let the Image Lead the Next Prompt
Once you generate something interesting, ask yourself: what if the lighting changed? What if it was zoomed in? What if it felt more surreal or more raw? Prompting isn’t a one-shot process, it’s a conversation and requires iteration.

Example Iteration: First version: “Vintage car on a foggy road.” Next: “Rear 3/4 view of a 1960s Jaguar driving into morning fog, golden hour, filmic texture, lonely atmosphere.” Add detail with each pass.

Prompting well means you’re thinking like a creative director, not just a user. The more experience you bring about photography, design, storytelling, the more you can guide the model toward something that feels like you made it.

Don’t just describe what you want. Describe why you want it. The more intention and detail you bring to the process, the better the results.