Brand Identity Sheet

A brand identity sheet is a single generated image that shows a brand's whole visual system at once — not one mark, but a UI card carrying the wordmark, a grid of textured/gradient app icons, pill-shaped buttons in the brand palette, a headline-scale wordmark treatment, and a secondary lockup with a small brand glyph. It's the format you reach for when the ask is "help me see the whole system," not "give me a logo."

This is a sibling to logo-prompting, not a replacement. Use logo-prompting when the deliverable is one mark. Use this skill when the deliverable is a board that puts the mark in context — alongside icons, buttons, and type at scale — the way a designer's Dribbble-style brand-sheet shot looks.

It's also downstream of moodboard-generator, not a competitor to it. A moodboard explores direction before anything is locked (loose adjectives, an open palette, rough logo sketches); this skill renders the locked system once real colors, typography, and a real mark exist. If step 1 below turns up no brand.md and the user can't state real hex/typography values, that's the signal to run moodboard-generator first rather than inventing values here.

The two artifacts this skill produces

1. A generation prompt — a tight descriptive paragraph, Ideogram-legible, that produces the sheet in one shot.
2. A compositional deconstruction — a structured JSON breakdown of the resulting image (or of a reference image the user supplies) with a high_level_description and an elements array of {type, bbox, desc} / {type: "text", bbox, text, desc} entries. This is not decoration — it's what makes the sheet reusable: another agent (or a future you) can read the bbox spec and know exactly which panel is the wordmark card, which is the icon grid, which is the CTA button, without having to re-look at the image. See references/composition-spec-format.md for the full schema and a worked example.

Always produce both. The paragraph prompt is what you feed the image model; the compositional deconstruction is what you save back to the vault so the sheet's structure survives past the single image (feeding hallmark/website work later, the same way a logo-prompting output can feed a site rebrand).

Workflow

1. Pull the brand truth

If the project has a settled position or a token lock, read them first. The position comes from position-recommender and tokens from kindform-tokens (DESIGN.md); treat both as settled. When they exist, the direction is already made and your job is to execute it, not to re-open it. Re-deciding by eye is how a derived token set quietly becomes taste again.

What to take, and from where:

A project with an older position.yaml and no DESIGN.md: read the same five things from it (tokens.colour, tokens.typography, diagnosis, intent, selection.fails_at).

Leave confidence labels and source appendices alone; they belong to the position document, not to this artifact. And if a settled file disagrees with something the user just told you, say so out loud rather than silently picking one. A contradiction between a standing position and a fresh instruction is a decision for the user, and it usually means a re-run (position-recommender for the position, kindform-tokens for tokens) rather than a workaround here.

Where neither exists, carry on below as normal.

Read the project's brand.md (Strategy + Voice + Visual layers) or design.md if the project has moved past brand.md into a locked design system. You need, specifically:

• Colors — exact hex/OKLCH values and their usage rule (which is the dominant paper/ink, which is the one accent, which is reserved for a single state).
• Typography — display face, weight, case (roman vs. italic).
• Style keywords + reference brands — these tell you the texture/material vocabulary (e.g. "metallic," "carbon-fiber mesh," "brushed aluminum" for a technical brand; "warm paper grain" for an analog one).
• The brand's own avoid-list, if it has one — many brand docs carry an explicit "anti-slop" / "avoid" / "don't do this" section naming specific visual clichés this brand must never look like. Pull the exact nouns from that list now, before you start drafting — you'll need to quote them verbatim in the prompt later (step 3), and a guardrail you only half-remember from this read is a guardrail that gets paraphrased into something too vague for the image model to act on.

If there's no brand.md/design.md yet, ask for the equivalent before drafting — a sheet built on invented colors is decoration, not identity, and will need to be redone the moment a real brand.md exists.

2. Choose the panel set

A full sheet doesn't need every panel from references/sheet-anatomy.md every time — pick the ones that earn their place for this brand. The canonical set, in rough reading order:

• Wordmark card — a white or paper-toned UI card with a small brand-name label at top, the wordmark large in the middle, and a tiny supporting line (edition/price/tag) at the bottom. This is the panel that anchors "what does the logo look like on a clean surface."
• App-icon grid — 4-8 square (or one tall) rounded-corner icons, each a distinct texture/motif pulled from the brand's material vocabulary: a gradient/glow icon, a mesh/weave texture icon, a metallic or material-sample icon, one mechanical/utility icon (gear, tool) if the brand has a technical register, one character/mascot icon if the brand has one. Each icon is a material sample, not a redundant logo restate — the point is range, not repetition.
• Pill buttons — 2-3 stacked pill-shaped buttons cycling through the primary, accent, and neutral/secondary colors, showing how CTA hierarchy reads in the palette.
• Giant central wordmark — the same wordmark at headline/hero scale, tightly spaced, anchoring the composition the way a type specimen sheet would.
• Secondary lockup — wordmark + a small geometric brand glyph (a star, a dot, a simple mark distinct from the primary logo) — this is the "what goes in a footer or a stamp" variant, not a duplicate of the main logo.
• Character/mascot panel (only if the brand has one) — a single stylized character icon, rendered in the same material vocabulary as the rest of the sheet.

Naming the panels explicitly in the prompt (not just "a moodboard of stuff") is what keeps Ideogram from defaulting to a generic swatch-and-Pinterest-board layout.

3. Write the paragraph prompt

Structure, in order — same discipline as logo-prompting, applied to a whole sheet instead of one mark:

1. Frame it as a brand identity guide / UI style sheet layout, not a "moodboard" — the word "moodboard" alone nudges the model toward loose Pinterest-collage energy; "identity sheet" / "style guide" nudges toward the structured grid you actually want.
2. Name the exact panel set from step 2, in the order they should read.
3. Color: exact hex/OKLCH, stated as the dominant paper + one accent + one secondary — never as an unweighted list.
4. Typography: face + weight + case for the wordmark specifically, since it appears at three different scales across the sheet (small card label, giant hero, secondary lockup) and needs to read as the same face at all three.
5. Material/texture vocabulary for the icon grid, pulled straight from the brand's style keywords — this is the part that most differentiates one brand's sheet from another's, so don't leave it generic ("some icons") — name the actual materials.
6. Negative space / anti-slop: this is not a generic "keep it tasteful" line — see references/anti-slop-discipline.md for the full universal-ban list (no neural-network nodes, no glowing-orb-as-hero-icon, no headset/padlock/lightbulb clichés, no purple-to-pink gradient wash across the whole sheet, no stock-photo textures, no invented product screenshots unless the brand has a real product to show) and name the brand doc's own avoid-list items verbatim if it has one — quoting the brand's specific nouns, not paraphrasing them into your own words, is what makes this clause actually bind. A "glow" icon is still a legitimate single panel (per sheet-anatomy.md); a gradient bleeding across the whole composition or background is not — see the anti-slop file's "glow containment" check.
7. Background/composition note: state it's a clean, structured grid layout (blueprint grid lines, light or dark ground per the brand) — this is what keeps the panels legible as distinct cards rather than bleeding into a collage.

Keep it one paragraph. If you want to explore two structurally different sheets (say, a light-paper version and a dark-mode version), write two separate prompts — same "structural variety over palette-swaps" rule as logo-prompting: two sheets that only differ by which hex is plugged into the same slots aren't two directions.

3.5. Run the anti-slop gate before generating

Before calling generate_image, score the drafted prompt against the pre-generation gate table in references/anti-slop-discipline.md (brand-specificity, guardrail transcription, universal-ban sweep, glow containment, restraint). Anything scoring under 3 on any axis means revise the prompt now — regenerating a shipped image is expensive; revising a paragraph before you generate is not. This step exists because palette and typography accuracy (step 1) and slop-freedom (this step) are different failure modes — a prompt can get the named-palette hex values exactly right while still defaulting to a generic AI-icon vocabulary, and only checking for the second catches it.

4. Generate it — don't just hand back a prompt

Call mcp__ideogram__generate_image with the paragraph prompt. Sheets like this benefit from style_type: "DESIGN" and a squarish-to-portrait aspect ratio ("1x1" or "4x5") since they're laid out as a grid of cards, not a single wide scene. Use rendering_speed: "QUALITY" — this is a composition-heavy image with a lot of small text-bearing panels, and QUALITY holds up legibility better than FAST/FLASH at that density.

If the user supplies a reference image they like (their own upload, or a prior generation) and wants variations, use mcp__ideogram__remix_image against it instead of a fresh generate_image call — pass image_weight around 40-60 for "same structure, different brand" remixes, higher for closer variations.

4.5. Check what actually rendered

The anti-slop gate scores the prompt; it cannot see the image. Before treating the sheet as usable, run references/post-generation-checklist.md against what came back. On a locked system this matters more than on an exploratory board, because the sheet's whole job is to show decisions that were already made: a drifted ground colour or an invented typeface here does not read as exploration, it reads as the system.

Retrieval, including the WebP conversion needed before the file can be viewed at all: references/ideogram-asset-retrieval.md.

5. Reverse-engineer into a compositional deconstruction

Once the sheet is generated (or if the user hands you an existing reference image to work from), produce the structured JSON breakdown described in references/composition-spec-format.md. Two ways to get there:

• From a generated image: you already know the exact panel layout you asked for in step 3 — write the JSON directly from that, since you control the composition.
• From a reference image the user supplies: call mcp__ideogram__describe_image first to get the model's own read of the image, then structure that description into the high_level_description + elements schema — add bbox estimates based on visual inspection of the image, not fabricated precision. If you can't confidently place a bbox, use a coarse quadrant estimate and say so rather than inventing exact pixels.

Save both artifacts (prompt + JSON) to the project's brand folder — see step 6.

6. Save what you made

Write the paragraph prompt, the JSON compositional deconstruction, and the resulting image URL(s) to 04-projects/<project>/logo-explorations/identity-sheet-<label>.md (or wherever that project keeps its branding work — check for an existing logo-explorations/ or branding/ folder first and match it). Follow the "No Context Lost" rule: nothing generated in this skill should live only in the conversation.

If a panel or a color usage turns out to be a pattern worth reusing across brands (not just this one project's specific palette), add it to references/sheet-anatomy.md as a named panel type, the same way logo-prompting promotes reusable techniques into its own reference files.

Reference files

• references/composition-spec-format.md — the full JSON schema for the compositional deconstruction (high_level_description, elements[] with type/bbox/desc, text elements with text), plus the worked Fizzwright example this skill was seeded from.
• references/sheet-anatomy.md — the named panel types (wordmark card, app-icon grid, pill buttons, giant wordmark, secondary lockup, character panel) with what each one is for and when to include or drop it.
• references/anti-slop-discipline.md — the universal AI-image-cliché bans (glowing orbs, neural nodes, gradient washes, stock-photo textures, and the rest), why a brand's own avoid-list has to be quoted verbatim rather than paraphrased, and the pre-generation gate to run before every generate_image call. Read this every time, not just when a brand doc happens to list its own guardrails — the universal bans apply regardless.
• examples/fizzwright-identity-sheet.md — the real paragraph prompt + full JSON breakdown this skill was built from, as a worked reference for format and level of detail.