Derive a usable seed from brand color evidence

Visual overview

Use the workflow to follow the task, and the architecture map to separate responsibilities. These are conceptual maps; the guide below defines implementation details and verification limits.

1. 1Sample locally↓
2. 2Remove background noise↓
3. 3Rank candidates↓
4. 4Confirm brand intent

1. Boundary 1Opaque pixel samples
2. Boundary 2Candidate seed
3. Boundary 3Semantic variants

Connected responsibilities, not a required class hierarchy or an execution trace.

Context

An extracted logo color is a candidate, not proof of an official brand standard. Prefer a user-provided brand seed when available. generate_color_system accepts locally sampled opaque sRGB colors; the MCP adapter also accepts imagePath for a local PNG. It never uploads images or fetches URLs.

Local workflow

1. Supply an explicit local imagePath (PNG, at most 1024 × 1024 and 4 MiB), or use a trusted local decoder to produce opaque sRGB samples. The PNG adapter skips transparent pixels and rejects color profiles requiring conversion. Exclude known background regions before manual sampling.
2. Pass a bounded list of colors and positive frequency weights as imageSamples.
3. The generator rejects very light/dark and near-neutral background-like samples, combines identical colors and ranks by frequency, with deterministic tie-breaking.
4. Inspect the chosen seed and candidates. The report includes black/white text contrast, showing which foreground is unsuitable without adjustment.
5. Confirm the intended identity with the designer. Regenerate supporting semantic variants and inspect them in real screens.

{"imageSamples":[{"color":"#FFFFFF","weight":900},{"color":"#145AC8","weight":160},{"color":"#E34D7A","weight":30}]}

This heuristic favors chromatic logo colors. A monochrome logo can legitimately produce no extracted seed; provide black/gray explicitly rather than pretending background rejection identifies brand intent. Existing project colors and design tokens can also supply seeds, but neither reveals the designer's semantic roles automatically.

Limits

The existing bundled PNG decoder is reused with size/dimension bounds. JPEG/HEIC/SVG, color-profile conversion, alpha compositing, spatial segmentation, near-color clustering and trademark lookup are unsupported. JPEG noise may create many small buckets; pre-quantize with the local decoder or select an explicit seed. Do not claim this is a full image-understanding system. Sample weights must be finite positive values; inputs are bounded for predictable local operation.