Generate Graphic

Generates an on-brand marketing graphic through an image model — the one opt-in feature that reaches the network — optionally stamped with the brand logo and a caption.

Overview

The generate-graphic skill is the opt-in raster engine. Everything else in brand-forge

(logos, social, docs) stays vector and zero-permission; this skill is the only path

that calls an external image provider, so it runs only when a user explicitly wants a

photographic or illustrative graphic. The generator builds a brand-aware prompt with

lib/raster.mjs, calls the provider with lib/genimage.mjs, and can composite a crisp

vector logo and caption over the result with lib/composite.mjs. Text is deliberately

kept out of the image and added as vector on top.

Prerequisites

• An active brand profile created with /brand-new. Read loads its color-system.json, typography.json, and visual-identity.md.
• Two opt-in signals, both required: BRANDFORGERASTER=1 and a provider key (GEMINIAPIKEY by default, or OPENAIAPIKEY). The 30-preflight emitter surfaces this at session start.
• Authentication: the provider api key is read from that environment variable and sent as a request header, never placed in the URL. No credentials are written to disk.
• Node.js on the PATH and Write access to an output/ directory.

Instructions

1. Verify both opt-in signals are set. If either is missing, stop, explain how to enable raster, and offer the vector path (generate-social or a template) instead, which needs neither.
2. Use Glob to locate the active brand directory, then Read and validate the profile with validateProfile.
3. Build the prompt with lib/raster.mjs (no network). It encodes the palette in words, the imagery style, and the tone, and turns the brand's don't-rules into an "Avoid:" clause. Warn the user when literal text is requested and route text-heavy work to the vector engine.

   import { loadProfile } from '../../lib/brand.mjs';
   import { buildRasterPrompt } from '../../lib/raster.mjs';
   const profile = loadProfile(activeDir);
   const prompt = buildRasterPrompt(profile, { subject: 'a mountain trail at dawn' });

1. Generate the image (the network call). The provider is swappable and failures retry with backoff, never leaving a half-written file:

   import { generateImage } from '../../lib/genimage.mjs';
   await generateImage({ prompt, outPath: 'output/northwind-hero.png', provider: 'gemini' });

1. Optionally composite the brand logo and a caption over the raster with lib/composite.mjs, which writes an SVG overlay referencing the PNG.
2. Hand the result to the visual-guardian subagent for a palette and legibility pass. Write the final files and report the saved paths.

Output

A raster PNG under output/, plus an optional vector overlay:

output/northwind-hero.png
output/northwind-hero-overlay.svg

• The PNG is the model's image, steered toward the brand palette, style, and tone.
• The optional overlay is a small SVG that references the PNG and adds a crisp logo and caption. Flatten it to a single PNG with /brand-export.
• Output is local files only — the skill never auto-posts or uploads.

Error Handling

To troubleshoot a failed generation, verify the two opt-in signals with /brand-status, then re-run; check references/brand-raster-prompting.md to debug an off-brand result.

Examples

Example — hero background

> /brand-make a hero image of a mountain trail at dawn

Builds a brand-aware prompt, generates output/northwind-hero.png, and offers to composite the logo over it.

Example — opt-in not enabled

> /brand-make a photo background

With BRANDFORGERASTER unset, the skill stops, explains how to enable raster, and offers a vector template instead.

Example — text routed to vector

> /brand-make a graphic that says "Now hiring"

Warns that baked-in text renders poorly, generates the background here, and composes the words as a vector overlay on top.

Resources

• references/brand-raster-prompting.md — how the prompt is built from the profile, why text stays out of the image, provider notes, and steering tips.
• Sibling skills: generate-social and generate-doc-template for text-heavy, zero-permission assets — route between them with the /brand-make command.
• Gemini image generation docs — the default provider's API.
• Agent Skills documentation — how skills are authored and invoked.