Brand · AI-generated media

Rules and copy-ready prompts for media made with generative tools — for the site, blogs, decks, and social. One principle governs everything: AI renders atmosphere; the design system renders information. Anything a reader must trust — numbers, labels, diagrams, type — is built in HTML/SVG with the tokens. AI supplies the celestial-enamel world around it.

0The house style block

Append this verbatim to every image or video prompt. It is the portable form of the Celestial Enamel language — edit the subject, never the block.

STYLE — warm cloisonné-enamel illustration on a deep espresso ground (#1C140D, near-black umber, never pure black). Flat fields of color separated by thin dark-umber gaps; every major shape rimmed in a fine muted-gold line. Celestial motifs: suns, eclipsing moons, stars, orbit arcs. Palette: cream white #FFFFE6, sun-gold #EAAA00, camel #C19A6B, chocolate #6F4A2E, plus sky #03CCD9 for luminous pulses. Matte, jewel-like, hand-crafted, golden-hour warmth. NEVER: gloss, photorealism, neon, purple-pink gradients, corporate-memphis figures, lens flare, text or lettering of any kind.

Whites in generated art must be cream #FFFFE6 — if a model returns pure white, grade it down. No text in generated images, ever: type is always set live in Goudy/Avenir over the art.

1Informational graphics

Data is never generated. Charts, diagrams, timelines, and stat callouts are drawn in HTML/SVG using the system (Goudy numbers, Avenir labels, one sky accent). AI contributes only a quiet decorative backing behind or beside the graphic.

PROMPT — abstract celestial-enamel backdrop: a sparse constellation of small gold-rimmed shapes and faint orbit arcs drifting across deep espresso, very dark and quiet, 80% empty ground, no focal subject, corners darkest. + STYLE block.

Contrast test

Cream-white 16px Avenir must pass 7:1 anywhere type will sit. If the backing fights the chart, darken the backing — never outline the type.

Spot glyphs

Small explanatory pictograms come from the icon set (Lucide, 1.75 stroke) — not from a model. Generate only scene-scale art.

Sizes

Blog inline 1600×900. Full-bleed deck backing 2560×1440. Export 2×, JPEG q85.

2Illustrations

Hero art, blog headers, spot illustrations. Compose for type: reserve a clean third of the frame on the side the headline will occupy.

HERO — [subject — e.g. "a circle of people and their glowing allies gathered on a night field beneath an eclipsed sun"], wide cinematic composition, subject held to the right half, left third fading to plain espresso for headline type. + STYLE block. → 2560×1440 (16:9)
SPOT — [single subject — e.g. "one gold-rimmed ally medallion among drifting stars"], centered, generous dark margin on all sides, quiet and emblematic. + STYLE block. → 1600×1600 (1:1)

Subjects that work

Allies as luminous enamel figures/medallions, fields and realms as landscape, orbits, thresholds, gatherings. Abstract over literal; never render real people or UI screenshots.

Lead accent

Gold and camel usually carry the image; sky and mint bring the luminous moments. Give each frame a clear lead accent so the eye knows where to land.

Fade to ground

Edges resolve into flat #1C140D so art sits seamlessly on any page — same rule as the OG cards.

3Video

Short ambient loops — site heroes, social clips, deck backings. Kiduna motion breathes; it never cuts, zooms, or strobes.

PROMPT — [subject or an existing still to animate], camera locked or drifting almost imperceptibly, enamel shapes breathe and glimmer, gold rims catch slow passing light, stars drift; gentle, continuous, seamlessly looping. + STYLE block. → 4–8s loop

Specs

Site hero 1920×1080; reels/stories 1080×1920; square feed 1080×1080. 24fps, H.264/WebM, muted autoplay, under 4 MB for web.

Motion budget

Under ~20% of the frame moves at once, so overlaid type stays readable. Start from an approved still (image-to-video) to lock the palette.

Sound & voice

Web loops are silent. No AI voiceover or AI music without explicit approval — captions carry the words.

4Memes

Memes ride recognition — never re-skin a meme into brand style. Use the native template at native fidelity, standard text treatment and all. The brand shows in the wit, not the pixels.

Format

If generating a template-style image, prompt for the original's look ("candid photo, early-2010s digital camera quality") — do not append the STYLE block. Skip the style block here and only here.

Branding

No logo, no watermark, no brand colors. The posting handle is the attribution. A meme that looks designed is dead on arrival.

Voice

Wry, warm, in on the joke about agents/AI/organizing. Never punching down, never dunking on named people or products.

5Social imagery

Posts follow the OG-card anatomy: espresso ground, art held to one side and faded in, type in the clear half, gold as the single accent (see Meta / Open Graph cards).

Sizes

X / LinkedIn 1600×900 · IG feed 1080×1350 (or 1080×1080) · stories/reels 1080×1920 · link previews 1200×630 per the OG spec.

Type over art

Set live or in the export layer — never prompted. Goudy headline 54px+ equivalent, Avenir sub, nothing below 28px at 1× post size.

Series rhythm

A carousel or series keeps one art style block and varies only the subject — palette drift across a series reads as sloppiness.

Before anything ships

Grade

Blacks to #120C07, whites to cream #FFFFE6, a clear lead accent per frame. Crop out any hallucinated text or extra limbs.

Export

2× resolution, JPEG q85 when art dominates / PNG for flat enamel, under 1 MB for web, real alt text describing the scene.

Disclose

AI-generated art is fine to use and fine to say so; never present generated imagery as photography of real people, places, or events.

neverPhotorealistic humans · fake screenshots or fake data · logos or wordmarks rendered by a model · purple-pink gradient slop · glossy 3D chrome · generated text of any kind.