Image Prompting

Overview

A reusable method for turning a content brief into a precise, repeatable prompt for a modern instruction-following image generator (the gpt-image family and equivalents). Load it whenever a task needs a visual asset: a blog cover, a video thumbnail, a social-post image, an inline illustration, a diagram/infographic, a logo mark, or an edit of an existing image.

The skill produces a prompt and a verification pass, not the image itself. It calls no rendering API — it gives the caller the text to feed one, the size/quality settings to request, and a checklist to judge the result.

Core principle

A generator renders what it can resolve from the words it is given. Two failure modes dominate — under-specification (it invents the parts you left blank) and drift (across edits or iterations, the parts you wanted kept silently change). The whole playbook defeats both:

1. Name every load-bearing decision — subject, medium, framing, light, palette, and any literal text. State them or the model decides for you, differently each run.
2. Separate what must change from what must stay — state invariants explicitly and repeat them every iteration. The single highest-leverage habit.
3. Change one variable per pass — isolated edits are debuggable; rewrites are not.

The loop

intake → spec → prompt → render → verify → (refine one variable) → ship

Stay in spec until the brief is unambiguous. Most wasted renders trace to a thin spec, not a weak model. Intake resolves purpose, subject, placement, text-in-image, brand/style anchor, mood, and hard-noes — asking only when the answer changes the image.

Prompt anatomy

Order the prompt from the outside in: [purpose/medium] → [scene/background] → [subject + pose] → [key details] → [composition/framing] → [light] → [palette/mood] → [literal text, quoted] → [negative constraints / invariants]. For anything reused or handed to another agent, prefer the line-per-slot form — maintainable, diffable, easy to vary one slot at a time.

The levers

• Composition — framing/crop, camera angle, subject placement, depth, and deliberate negative space for later text/logo overlay (the move that makes covers and thumbnails usable).
• Style & medium — state the visual medium first (photo, 3D, flat vector, watercolor, isometric…), then layer qualifiers only as needed. Name a concrete reference register ("editorial magazine illustration") rather than an artist.
• Camera & lens — only for photoreal: the explicit word "photorealistic", focal length, aperture/DoF, film/sensor cues, authenticity markers.
• Light — direction, quality (soft/hard), time/temperature, special (volumetric, rim, high/low-key). Sets realism and mood more than any other single factor.
• Mood & palette — one or two adjectives carried through; a concretely named palette (or hex anchors for a brand).

Text in image, constraints, and size

Default to keeping text OUT of the generated image and compositing it in a layout tool afterward — more legible, more on-brand, trivially editable. When words must be baked in, quote the exact string, specify typography as constraints, and add "render this text once".

Distinguish exclusions ("no watermark", "no extra text") from invariants ("keep the same face, pose, background") — and repeat the full invariant list on every iteration, since they do not persist across turns. Request the native aspect ratio the placement needs (16:9 covers/OG, 9:16 stories, 1:1 posts) and let the tool render it natively rather than upscaling a square.