Sharper analyses: style direction, reference imagery, and copy voice
A meaningful expansion of what an analysis tells you. The new fields move the output from "describe the design" toward "give me what I need to build it." If you analyzed a site before this release, run it again to see the new sections.
- New
Style direction with a spider chart
Every analysis now scores the design across 12 specific style directions (Swiss editorial, cinematic product, calm luxury wellness, grid-heavy SaaS, and more) instead of one generic label. The result lands as a spider chart on the entry page, with the dominant directions surfaced as tags on thumbnails so you can describe a site as "calm luxury wellness with notes of opulent minimalism" rather than just "minimal."
- New
Reference imagery, sourced automatically
Analyses now include a build-actionable description of the photography or graphics that drive the site, plus three matching reference photos pulled live from Unsplash. The photos and a search prompt land in the entry page and the downloaded design.md, so a build agent gets working image URLs out of the box.
- New
Copy voice and drop-in copy
Reads the visible copy on a site and infers the brand voice as a system: register, rhythm, vocabulary, and how CTAs phrase. Then produces in-voice placeholder strings (nav labels, hero headline, body, CTA labels, promo banner) you can drop straight into a build.
- Improvement
design.md downloads carry every new field
The exported design.md now includes the style scores, imagery direction, reference image URLs with attribution, copy voice rules, and drop-in copy. Hand it to any AI agent and the site builds with real photos and brand-correct language, not lorem ipsum.
- Improvement
More resilient screenshot capture
Sites with aggressive bot detection (some luxury, corporate, and editorial brands) used to fail before the analysis even started. Capture now retries with a relaxed mode so most of those sites succeed, with a slightly imperfect screenshot rather than a hard failure.