docs: add the demo video link (README + the app's Watch-the-demo button)

perf: also warm STT during the popup so the first hold-to-ask is fast (Whisper is CPU, no quota)

perf: warm the model during the welcome popup (once per session) so the first describe is fast

popup: teach how to use (gestures + live-mode voice command) in text and narration

fix: popup empty on cold load — lazy-load gradio client so UI never waits on the CDN

feat: welcome popup on every open (reads pitch aloud, voice/tap close, video link)

ui: brand mark = minimal line eye (option 2); remove preview page

preview: brand-mark candidates page (temporary, /static/marks.html)

ui: taste-skill design pass (Outfit display font, color/radius system, softened glow)

ui: add the iris brand mark (idle identity)

ui: polish pass (depth scrims, live dot, speaking/idle motion, answer entrance)

live mode: stop self-hearing, prioritize user questions, kill repetition/hallucination

live mode: in-browser object-detection gate + browser-language onboarding

tests, inference trace, English-only repo, testability refactor

Iris: hands-free live mode, money/bill reading, accessible UI, Qwen3-VL-2B

Iris: voice-first assistant for the blind (gr.Server + Qwen2.5-VL + Whisper + Piper)