# #logo-maker-machine

**Kitt** (2026-03-05 17:21)
**Logo Maker Machine**

Current tools produce generic output that ignores style references. We need a proper pipeline.

**Problem:** Recraft generates logos with zero connection to the visual DNA of reference brands.

**Goal:** Build a system that can extract style from case studies (like center.design/project/ayoh/) and generate logos that actually inherit that DNA.

**Open task from #phat-foods:** Olevia & Aria logo exploration — paused until this pipeline works.

Ready to work on this tomorrow.
**Assaf Dagan** (2026-03-06 07:49)
What do you propose in this front - built a super clean process to get stuff from the strategy board through the taste board and create a clean brief for Gemini / recraft ?
**Kitt** (2026-03-06 07:50)
**Yes chef** — I'll design a clean pipeline from strategy board → taste board → generation brief.

Looking at the problem: Recraft generates generic logos because it's getting generic prompts. We need **style DNA extraction** + **generation bridging**.

## Proposed Logo Maker Pipeline

**Stage 1: Style DNA Extraction**
- Input: Reference case study URL (like center.design/project/ayoh/)
- Extract: Typography choices, color palettes, layout principles, visual hierarchy
- Method: Gemini Vision analysis on captured screenshots + structured prompts
- Output: Style DNA profile (structured data)

**Stage 2: Brief Generation Engine**
- Input: Style DNA + brand positioning/name
- Process: Translate visual DNA into generation-ready prompts
- Method: Structured prompt templates that map DNA to Recraft parameters
- Output: Optimized generation brief
**Kitt** (2026-03-06 07:50)
**Stage 3: Iterative Generation**
- Input: Brief + style constraints
- Generate: Multiple concepts via Recraft
- Filter: Auto-evaluate against style DNA criteria
- Output: Logo concepts that inherit the reference DNA

**Implementation Plan:**
1. **Tatiana** — Build style DNA extraction system (visual analysis framework)
2. **Thibault** — Create the bridging engine (DNA → generation prompts)
3. **Anton** — Quality validation framework (does output match DNA?)

This solves the core problem: generic output becomes **DNA-informed output**.

Ready to brief the team and build this system. Should I start with Tatiana on the style extraction framework?