The prompt was the product.
I built the same mobile app idea twice. First from a rough looksmaxing prompt. Then from a verified v-1 Library prompt. The difference was not that one agent typed faster. The difference was the design source material it started from.
The setup
The first app was Lookline, a rough looksmaxing app I generated without a v-1 Library prompt. It had routes for home, grooming, fit, and proof. It was functional. It looked like something an agent can make quickly when the brief is mostly vibes.
The second app was Lustre, built from the v-1 Library prompt you supplied. Same broad category. Very different input. Lustre came with a product thesis, exact routes, token rules, data locks, motion notes, image rules, and a clear hero interaction: the before and after glow timeline.
What changed
The rough prompt produced a usable app. That is not the insult. The problem is that usable is the floor now. Anyone using Codex, Claude Code, Cursor, Bolt, or Lovable can get a screen on the page.
The hard part is getting every screen to feel like one product. That is where the Library prompt changed the build. It gave the agent a design system before the agent started inventing components.
Looksmaxing tracker with a clean dark UI.
A premium glow-up tracker with a before/after light seam, monthly re-scans, deltas, and a kind voice.
Four screens existed, but the system was invented as the build went.
Timeline, Today, Insights, and Plan were specified as real routes before code started.
Good enough palette, but easier to stretch or drift under pressure.
Porcelain ground, espresso ink, bronze UI, rose deltas, fixed type roles, and shared chrome.
The agent decided most constraints itself.
The prompt named native constraints, asset rules, motion, data locks, and exact copy.
The catch
v-1 does not remove implementation work. The app still needs a good coding agent, a real repo, route wiring, state, responsive checks, and visual verification. The Library prompt gives the agent better source material. It does not magically make verification optional.
Your agent does not need another screenshot to copy. It needs a source of truth.
Builders already have code agents. What they are missing is the design layer those agents can trust: one prompt, one system, screen contracts, copy, data, and constraints.
This post uses the real local demo reel and screenshots as proof. For a future cover or social thumbnail, the queued image direction is Higgsfield `nano_banana_pro`: two premium mobile app screens side by side, rough prompt on the left, Library prompt on the right, clean desk, high contrast, no watermark.
I built the same mobile app twice. Rough prompt first. v-1 Library prompt second. The second one did not win because the agent was smarter. It won because the prompt had a design system inside it.
Full disclosure, I built v-1.design. I tested the same looksmaxing app idea two ways: a rough prompt, then a verified Library prompt. The lesson was painfully obvious: the agent needed design source material, not another screenshot to copy.
I built the same app twice to test something simple: does a stronger design prompt change the final product, or just the copy around it? The answer showed up in the routes, tokens, motion notes, and data locks.
I turned a local demo into a field note: one rough prompt app vs one v-1 Library prompt app. The rough version proved speed. The Library version proved why design source material matters.