Every time I ship a side project, I spend 3 hours on brand assets I don't care about. Logo. Favicon 16 and 32. Apple touch icon. OG image. Twitter card. Discord embed. GitHub banner. iOS icon. Android icon. Different sizes, different crops, same logo.
Each one is a browser tab, a prompt I'll throw away, a manual copy-paste, a resize in Figma.
So I built img-pilot — the first Claude Code plugin that generates AI images from your terminal. And because I refuse to pay for eight API calls when I only need the first one, I built a cost optimizer that derives every variant locally.
One API call. Eight assets. Nine providers.
What it actually does
From inside Claude Code:
/img-pilot logo
That triggers four phases:
-
Claude reads your context — if you have ux-pilot or brand-pilot installed, it parses their brief files. Otherwise it runs a 6-question discovery (ABCD + free text) and writes
img-pilot/brief.mdfor future runs. - Claude builds the prompt — from the brief, your brand vocabulary, your target provider's strengths. Not a generic "make me a logo" — a real production prompt.
- The cost optimizer picks the cheapest path — this is the interesting bit. More below.
- Dry-run confirmation — you see the plan (provider, exact prompt, cost estimate in USD) and type yes before a single byte goes out.
The cost-optimizer trick
Before every asset request, the optimizer walks a decision tree:
