♪ Take notes ♪The rules of honesty at this station.
Let AI do front-end UI, interactive experience, product imitation, not just words describing aesthetic and body. The descriptional bandwidth such as “More smooth” “No AI” “Reference mature products” is too low and leads to repeated error.
Core principles: The front-end demand prioritizes the use of screenshots, videos, mature product pages as a common context. First, AI dismantles mature products and makes them imitated and adapted to the current project, rather than directly using language.
Recommended process
- Collection of reference materials: key pageshots; scrolling, input, loating, error screens; multiple competitions.
- Let’s start with the millimeter dismantling.: page area structure, list of components, font-spaced corner colour, information density, empty operational failure, dynamic rhythm. Let AI take it down first, don’t write the code.
- Refit the project.• Replicate the logic of experience, what can only be learned, and what needs to be adapted to its own users and operations.
- Post-achievement Chassis: allow AI to capture its own screenshot for the time being, with only one or three of the most affected issues per round as indicated by the reference figure.
Counter Mode
- Just say “a little higher,” “a little more like a product.”
- Without a screenshot, let AI guess the mature product UI
- Write the code directly, not decode it first.
- I’m gonna do the check and check.
One sentence: The aesthetic can not speak, but can produce evidence.