Brand context

Every generation includes a project-scoped brand context that's injected into the vision prompt. Two different projects pointed at the same source image produce different motion — the brand context tells the model what your product is, who you sell to, and what tone it should hit.

What lives in the brand context

ts
type ProjectBrand = {  brandName: string;            // "Acme"  colors: {    primary: '#0d9373';    secondary: '#1f2937';    accent: '#f59e0b';    background: '#0a0a0a';    foreground: '#fafafa';  };  fonts: { heading: 'Inter'; body: 'Inter' };  logoUrl?: string;  // Populated by deep-crawl:  contentInventory?: {    tagline: string;    description: string;    features: { title; description; icon? }[];    pricing: { tier; price; features }[];    testimonials: { quote; author; role? }[];    stats: { label; value }[];    announcements: { title; date?; body }[];    codeExamples: { language; code; label? }[];  };  messagingContext?: {    tone: string;               // "technical, confident"    targetAudience: string;     // "platform engineers"    valueProps: string[];    ctas: string[];  };  designContext?: {    vibe: string;    borderRadius: 'sharp' | 'rounded' | 'pill';    backgroundPattern?: 'dot-grid' | 'line-grid' | 'gradient-only' | 'solid';    hasGlassmorphism: boolean;    hasGradientBorders: boolean;    hasRadialGlows: boolean;    monoFont?: string;    sectionLabelStyle?: string;    colorScheme: 'dark' | 'light';  };};

How it's injected

The vision model receives the brand context as a system-prompt block before it inspects the source image:

text
Brand: Acme — fast SaaS observability platformTone: technical, confidentTarget: platform engineersColors: primary #0d9373, accent #f59e0bVibe: dark, sharp edges, dot-grid backgrounds, no glassmorphism

Then it generates a motion prompt that fits both the source image and the brand. Without brand context, the model defaults to a generic "make this image move tastefully" prompt that ignores your product identity.

Three ways to populate it

Per-asset overrides

Brand context applies to every generation in the project. To bypass it for a single asset, pass ?prompt=… and your override replaces the brand-aware prompt:

text
…/hero.png.mp4?prompt=cinematic+slow+pan+left

Capped at 500 chars. Use surgically — you lose the brand-specific guidance.

When brand context is empty

Brand-config-empty projects still generate fine; the prompt is just less opinionated. New projects start with empty brand_config; the dashboard nudges you to run brand-detect on first sign-in.

Inspecting the prompt

Every asset row stores vision_prompt in D1. To see what the vision model wrote for a given asset:

bash
curl -b cookie.txt "https://motioness.com/api/assets/$ASSET_ID" | jq .asset.vision_prompt

Useful when an output looks off — read the prompt, decide whether to improve the brand context or pass ?prompt= for that one URL.

Updating brand context

Whenever you update the brand on the project, the next generation picks it up. Already-generated assets are content-addressed and won't re-run unless you regenerate them (POST /api/assets/:id/regenerate) or vary idem.