generate-mockups

name	generate-mockups
description	Generate visual mockups of the finished hardware project using fal.ai's nano-banana (Gemini 2.5 image) model. Useful for early-stage visualisation, communicating intent to collaborators, or scoping enclosure design before any parts arrive.

Generate Mockups

Produce AI-generated mockup images of a project from its written spec.

Prerequisites

• The fal.ai MCP server is available (mcp__*__fal__run_model or equivalent), or the user has a FAL_KEY and we can call the HTTP API directly.
• A project with spec.md exists.

Flow

1. Build the prompt from spec.md. Include:
   • Form factor (handheld / desk-mounted / wall-mounted / outdoor enclosure / wearable).
   • Key visible elements (display, buttons, sensors poking out, antennas, LEDs, connectors).
   • Material / finish hint (3D-printed PLA, machined aluminium, laser-cut acrylic, custom PCB visible).
   • Style hint (clean product render, exploded view, lifestyle shot, technical schematic-style).
   • Negative prompt — no text overlays, no logos, no people unless requested.
2. Pick the model — default to fal-ai/nano-banana (text-to-image). For variations of an existing image, use fal-ai/nano-banana/edit. Both are fast and cheap, suitable for rapid iteration.
3. Generate 2–4 candidates in parallel with slight prompt variations (different angles or styles) so the user has options.
4. Save the images to <projects_dir>/<slug>/mockups/ with names like mockup-<timestamp>-<variant>.png. Save the prompt used alongside in mockups/prompts.md so the user can iterate.

Output

• Files written to <projects_dir>/<slug>/mockups/.
• A summary line in the conversation showing each saved file and the variant description.

Iteration

If the user wants tweaks ("make it smaller", "show it mounted on a wall", "battery should be visible"), use fal-ai/nano-banana/edit with the previous best image as input rather than starting from scratch — preserves the bits that worked.

Caveats

• Mockups are illustrative, not engineering drawings. Don't claim dimensional accuracy. They're for communication / inspiration, not fabrication.
• The model will sometimes invent components that aren't in the BOM. Note any discrepancies for the user when presenting candidates.