Brand Resolver

When a user mentions a URL in an authoring prompt — "make a launch video for stripe.com" — Workshop's brand resolver automatically extracts brand colors from the referenced website and injects them as context for the AI authoring process. The video arrives on-brand without manual color configuration.

How It Works

When a URL is provided, Workshop runs an intelligent heuristic extraction engine against the target website to discover primary, accent, and background brand colors. These are ranked, deduplicated, and injected into the LLM's context to generate brand-accurate .ws scripts.

The result is a compact brand palette that can inform the authoring prompt or seed a ThemePack.

Integration with AI Authoring

When POST /api/author receives a prompt containing a URL, the brand resolver runs automatically. Extracted colors are added to the LLM's context. The LLM uses these to inform mood, color token, and emphasis choices in the generated .ws script.

Extraction is best-effort and non-fatal. If the URL is unreachable or yields no usable colors, authoring continues without brand hints.

From Extraction to ThemePack

The brand resolver produces a one-time extraction. For persistent brand enforcement, register the discovered palette as a ThemePack. This converts a dynamic extraction into a permanent, compiler-enforced brand specification.

Limitations

Works best with well-structured marketing pages that prominently expose brand colors in their markup and stylesheets. Less accurate for pages that load visual identity dynamically via JavaScript, use image-based logos without vector equivalents, or require authentication. The resolver does not take screenshots or execute client-side scripts.

Next Steps

• AI Authoring — using the brand resolver in practice
• Author API — endpoint that triggers brand resolution
• ThemePacks — registering extracted colors as a permanent brand