The Anatomy of an Effective Image Prompt
Generative AI Prompting for Image Models
Chapter 3 · The Anatomy of an Effective Image Prompt
prompt1-2 built its own anatomy of a good prompt around four instruction-oriented pillars: Clarity, Context, Constraints, Output Format. That anatomy makes sense for a system built to comply with a request (imgai1-1). An image model needs a genuinely different anatomy — one built around description, not instruction — because, per imgai1-2's own mechanism, every word in the prompt is just another statistical anchor pulling the denoising process toward a region of image space, over and over, at every step. This chapter builds that anatomy: six descriptor categories every tool-specific chapter from here on (imgai1-4 through imgai1-6) assumes without re-teaching.
Six Categories, One Shared Vocabulary
Regardless of which of the three tools this course covers, a strong prompt tends to touch some combination of the same six things. Syntax differs wildly by tool (covered next, in imgai1-4–imgai1-6) — this vocabulary doesn't.
Subject
Who or what, doing what, with what specific details. The single biggest lever — vague subjects produce generic results (see this chapter's own warn-box).
a scruffy terrier mid-jump catching a red frisbee weathered hands
Style
The overall aesthetic or art-historical reference point. Named living artists raise real, direct questions — previewed here, covered fully in imgai1-9.
cyberpunk art nouveau photorealistic watercolor
Composition
Framing and layout — how the subject sits inside the frame.
close-up wide shot rule of thirds symmetrical bird's-eye view
Lighting
Where light comes from and how it feels — one of the most reliable levers for mood.
golden hour dramatic rim lighting soft diffused light neon glow
Camera / Lens
Real photography terminology — works precisely because it's what real photographs are captioned with (imgai1-2's own CLIP training data).
35mm lens shallow depth of field bokeh long exposure
Medium
What kind of object the output should look like it is — the single strongest lever for realism vs. illustration.
oil painting 3D render pencil sketch digital art photograph
imgai1-2, CLIP was trained on real (image, caption) pairs scraped at massive scale — a huge fraction of which are real photographs with real photography metadata and captions attached ("shot on 35mm," "shallow depth of field"). The embedding space genuinely encodes what those terms visually correlate with, because real photographers' own captions taught it to. This is also exactly why the same terms don't work identically across all six categories equally well — camera language is unusually well-represented in the training data specifically because photography captioning is common online.
A Worked Before/After Example
a lighthouse
a weathered stone lighthouse on a rocky cliff [subject], oil painting [medium], impressionist style [style], wide shot with the lighthouse off-center [composition], dramatic golden-hour lighting with long shadows [lighting], shallow depth of field, 85mm lens [camera/lens]
Notice the "after" version doesn't read as an instruction to a listener — it reads as a dense list of descriptors, exactly as imgai1-1 predicted. Nothing in it asks the model to do anything; every phrase describes what the finished image should look like.
imgai1-2's own mechanism, the text embedding for a bare, vague prompt sits in a huge, poorly-differentiated region of the model's learned space — it's consistent with an enormous number of very different training images (photos, paintings, cartoons, day, night, close-up, distant). The denoising process gets pulled toward whatever's statistically most common across that whole broad region — which is exactly what "generic" means here: not a flaw in the output, but the predictable result of an under-specified target region. Every one of the six categories above exists specifically to narrow that region down.
Ordering Carries Real Weight
Across most tools, terms earlier in a prompt tend to exert more influence on the final image than terms later in it — not a hard universal rule, but a consistent enough pattern to plan around. This is one reason subject conventionally comes first: it's the single most important thing to get right, so it goes where it has the most leverage. imgai1-7 covers the more precise, tool-specific mechanisms for controlling relative influence (term weighting) directly.
Hands-On Exercises
Contrast this chapter's own six-category anatomy against prompt1-2's Clarity/Context/Constraints/Output Format anatomy, and explain — using imgai1-1's own distinction — why an image model needs a genuinely different anatomy rather than a relabeled version of the same one.
📄 View solutionUsing this chapter's own warn-box and imgai1-2's own mechanism, explain mechanically why a vague, single-word prompt like "a lighthouse" produces a generic result, rather than describing it as simply "the AI being lazy."
📄 View solutionExplain, using this chapter's own tip-box, why real camera/lens terminology ("35mm lens," "shallow depth of field") is a genuinely effective descriptor category rather than an arbitrary one, tying the explanation back to imgai1-2's own description of CLIP's training data.
📄 View solutionChapter 3 Quick Reference
- Six shared descriptor categories: Subject · Style · Composition · Lighting · Camera/Lens · Medium
- Vague prompts produce generic results because their text embedding sits in a broad, poorly-differentiated region of the model's learned space (imgai1-2)
- Camera/lens language works because real photo captions are heavily represented in CLIP's own training data
- Earlier terms generally carry more influence — subject conventionally leads for this reason; imgai1-7 covers precise weighting mechanisms
- This vocabulary is shared across tools — syntax differs, starting with Next chapter: Midjourney — Parameters & Its Own Artistic Bias