Svara · design exploration · 2026 · round 5
Eighteen logged-out aesthetics for the same model card.
Same product. Same content. Eighteen fundamentally different page architectures. The product underneath is consistent: svara-global-v1, a 780M-parameter open-weights TTS model that holds a single voice across 50+ languages in one API call, with inline emotion / prosody tags and Apache 2.0 licensing. V1–V5 were round 1, V6–V10 were round 2 grounded in real provider playbooks, V11–V15 are round 3 — variations on the V6 shape pitched at content teams, creators, and business buyers (Murf, Hume, Lovo, ElevenLabs verticals, Cartesia realtime). V16 is the Mistral / Voxtral-style model launch. V17 and V18 sit on a shared subtle-grid foundation (Simplismart / Baseten / Vercel family — 1px grid, hairline dividers, plus-marks, sharp-cornered cards, playground as the hero, blind A/B test, two-path open-source CTA) — V17 in mono cobalt, V18 in a six-hue rotation on the same skeleton.