1. Raw model output
- Format
- 458 KB raster PNG
- Palette compliance
- Visual drift — gradients, shadows
- Ship as logo?
- No — raster, no scalability
AI image models drift off your hex palette. brandkit fixes it downstream — generate, vectorize, snap-to-palette, verify. Ships as a CLI and an MCP server.
Prompt: "minimalist geometric symbol for a dashboard analytics app, abstract chart-bracket motif" · Palette: #FAFAF9#5B21B6#10B981#1C1917
verify: 48 offenders
verify: 0 offenders ✓Same source raster across columns 2 and 3. The only difference is brandkit's snap-to-palette + verify-or-drop step. See the recolor logic.
npm install -g brandkit
brandkit --help
git clone https://github.com/gent8/brandkit
cd brandkit
docker build -t brandkit-mcp:latest -f mcp/Dockerfile .
Then add to your MCP client config:
{
"mcpServers": {
"brandkit": {
"command": "docker",
"args": ["run", "-i", "--rm",
"-e", "FAL_KEY",
"-e", "RECRAFT_API_KEY",
"brandkit-mcp:latest"]
}
}
}
Provider keys: fal.ai (Ideogram v3) · Recraft (vectorize). Bring your own.
brandkit_genFull pipeline. Prompt + palette → palette-locked survivor SVGs returned inline. No filesystem handoff.
brandkit_recolorPure SVG-in / SVG-out. Snaps every hex and rgb() to the nearest palette color.
brandkit_verifySVG-in / JSON-out gate. Returns offenders with normalized hex + suggested palette match.
Solo-maintained, AGPL-3.0. Read the status doc before opening issues.