Prompting like an art director — one anatomy, four dialects — step 1 of 8

Tell the model what to make

You're in the graphic design studio. PromptDojo is a practice workspace, not an image generator — you will not make pictures in this browser. What you're doing this lesson: write image prompts the way you'd write a brief for a junior, then see how four tools read that brief. The code box, if it appears later, is optional.

You do not need to be an art director, and you do not need Midjourney flags memorized. You need a brief the model cannot fill in for you: what's in the picture, how it should look, and what must never appear. That translation is the job. The only new part is who's on the other end — a model that will invent the missing sections from the average of the internet.

If you are starting here, this lesson is that skill, written down. If you already turn "make it pop" into a spec a junior can execute, you already have this skill. Either way, start here — not with tool dialects.

This chapter's production prompt is a studio kit brief — five sections that compile a brand artifact (illustration, poster, mark, the campaign still that has to match the kit):

1. Subject — what's in the frame. Not "coffee stuff." A rooftop espresso bar at dusk.
2. Medium — what kind of artifact this is. Flat vector illustration. Risograph print. Photograph. This single choice moves the output more than any other.
3. Style — the execution spec. Stroke weight, texture, print process. Camera language lives here only when the medium is a photograph. 2px uniform stroke, isometric.
4. Palette — named colors, ideally your brand's. Terracotta and deep teal on off-white.
5. Negatives — the do-not-appear list. No gradients, no photorealism, no watermark.

This is not the image chapter's eight-knob camera brief (subject, composition, lighting, style, camera, lens, palette, negatives). That shape is for a shot. Composition, lighting, camera, and lens are first-class there because the job is a photograph. They are not first-class here because the job is a designed object — medium is the load-bearing slot. When a studio deliverable is a photograph, stack them: kit first (medium, palette, banned looks), then the eight camera knobs on top. Don't flatten one shape into the other, and don't grade a kit prompt as "missing lens."

Skip a section and the model doesn't leave it blank — it fills it with the statistical average of everything it's seen. The average of all posters is the poster you'd be embarrassed to present.

Same anatomy, four dialects — later, not first

Once the five sections are solid, the tools you'll meet in client work all consume this anatomy. Each speaks its own dialect. Learn the anatomy first; glance at vendor docs before you bill anyone for a flag name.

• Midjourney (an AI image tool you prompt with prose plus flags) — the taste engine. Parameters: --ar 4:5 for aspect, --no for negatives, --sref for style references, --seed for repeatability. Wins on aesthetic instinct; Midjourney's current quoted-text rendering made lettering more legible, but it still trails the text specialist for letter-perfect copy.
• FLUX (another AI image model, more literal than Midjourney) — the literalist. Full descriptive prose, no flag syntax; it follows compositional instructions ("subject in the lower-left third") more faithfully than the taste engines. Photorealism is its home turf.
• Ideogram (the AI image tool built for text-in-image) — the typographer. Put your exact copy in quotes and it usually sets it.
• Nano Banana — the production intern. Google's speed-optimized image family: conversational, strong at instruction-based edits, and cheap enough to generate fifty candidates and keep three. (Which tier is cheapest rotates; the batch-workhorse role doesn't.)

Version numbers on all four rotate every few months; the anatomy doesn't. Treat every tool name, version, flag, and price in this chapter as an illustrative mid-2026 snapshot, not a curriculum-verified fact. The skill this chapter teaches is the anatomy, which outlives whichever model wins next quarter.

The rest of this lesson drills the five sections until they're reflex, then the checkpoint has you compile three briefs into their tools' dialects — which is exactly what a busy Tuesday looks like.