MJ Basics: How I Create Images Using Text Alone (Patreon)
Content
This Midjourney user still creates their image the old-fashioned way. (This is a guest post from our friend Gus, I hope you guys enjoy!)
I genuinely enjoy working with --sref and --p. They’re powerful, fast, and they solve real problems. Yet, over time, I started experimenting (again) with doing more of the steering directly inside the prompt itself, as a complementary habit. What came out of that is a way of writing prompts that I find pretty enjoyable. I’m sharing it here in case it’s useful to you too.
How I Usually Start
When I’m writing prompts, I tend to follow a simple structure. I start with the subject, then put more information gradually: what it’s doing, where it is, how it’s lit, and only then how it should look stylistically.
In practice, this usually means something like: subject first, details second, environment next, lighting after that, and style or technical cues at the end. I believe that Midjourney pays more attention to the beginning of the prompt, so I try to put the most important decision up front and let everything else refine it.
A rough mental template looks like this:
[Subject in action], in [setting], [lighting], [style or technical cues]
Here’s a simple example:
A rugged mountain knight standing on a cliff at dawn, soft early-morning light, shot on 85mm lens, high detail, natural color tones
And here’s a more stylized variation using the same structure:
A lone samurai walking through a quiet street at night, rain-soaked environment, neon backlighting, anime-style illustration, crisp line art, cel shading
What I like about working this way is that each layer does one job. The subject defines what the image is. The environment and lighting define how it feels. Style cues and technical details narrow the visual language without overwhelming the prompt.
I’ve also found that being specific helps more than being dramatic. Instead of saying “cinematic” or “realistic” and hoping for the best, I get better results by naming things Midjourney already understands: a camera, a film stock, an era, or a visual medium. Small choices like that tend to make the image feel more intentional, even before I start refining it.
Realistic Images: Describing the Photograph, Not the Vibe
When I aim for realism, I focus less on adjectives and more on photographic language. Cameras, lenses, film stock, and lighting feel like strong anchors, even when I’m also using references elsewhere.
An example of the kind of prompt I often write:
A cinematic portrait of a middle-aged man with weathered skin, shot on Canon EOS R5, 85mm lens, f1.8, natural window light, soft shadows, Kodak Portra 400 color science, high detail, subtle film grain
What I like here is that each line limits ambiguity. The image doesn’t have to guess what “cinematic” means: the constraints already define it.
Anime: Letting Illustration Rules Lead
For anime-style images, I intentionally step away from photographic thinking. I treat them as illustrations with their own internal logic: lines, color blocks, and controlled shading.
This is a structure I return to often:
A young swordswoman standing in the rain, anime-style illustration, crisp line art, bold outlines, cel shading, vibrant but controlled color palette, dramatic pose, detailed character design
If the image starts drifting toward realism, I reinforce illustration terms. If it feels too soft or painterly, I clarify line behavior. Small adjustments usually go a long way.
Retro: Being Explicit About the Era
With retro aesthetics, I’ve found that being very explicit helps more than being subtle. Era, color language, and media artifacts all matter.
For example:
A neon-lit city street at night, 1980s retro-futuristic aesthetic, electric blue and hot pink neon lights, VHS film grain, scanlines, high-contrast lighting, light fog, shot on 35mm film
The goal here isn’t accuracy so much as coherence. When the signals agree with each other, the result tends to feel intentional.
How I Debug Prompts
One habit that made the biggest difference for me was changing one thing at a time. When a result is off, I don’t rewrite the whole prompt. I tweak a single cue, regenerate, and compare. Patterns show up quickly when you do this consistently.
For example, if a realistic image starts looking painterly, I usually remove anything that hints at illustration and add technical constraints instead (lens, aperture, or lighting behavior...). Swapping “cinematic lighting” for “studio flash with hard shadows” often works better than piling on more adjectives.
I also keep a simple prompt notebook with phrases that worked, phrases that didn’t, and notes about order. Order matters more than it looks. And I avoid stacking conflicting style cues. When a prompt tries to do everything, the result usually feels averaged and vague.
Exploring Without a Target
Sometimes I use prompts to explore rather than control. In those cases, I leave space on purpose.
A mysterious photograph from the 1980s, something strange happening, moody atmosphere
I use this approach especially at the edges of a style. Broad phrases like “nostalgic neon memory” or “alien carnival at midnight” are useful starting points. Once something interesting shows up, I switch modes and become more precise.
For me, the balance matters. Guided prompts give reliability. Open prompts give discovery. Going back and forth between the two is how I map what the model is capable of — and where I actually want to be.
What's your favorite approach?
Files