Decos: castle slightly higher (90 → 70) + phoenix slightly lower (660 → 690)

Decos: drop sorting hat down to top: 670 for better balance with phoenix

Decos: push bunny down to balance with castle / phoenix on the left column, nudge cat for vertical centering

Decos: bigger gap (80px) from frame + fixed-pixel vertical positions (no more drift on resize)

Decos: more frame-edge gap (50px) + clamp vertical positions to stable ranges to limit drift on resize

Decos: anchor to the frame's edges (not viewport) so they keep a constant gap on resize + center bunny between castle and phoenix

Decorations: nudge inward (5% → 8%) + grow ~10% + drop the bunny tableau down to top: 30vw

Decorations: drop the max-height media query — let them stay anchored and clip if needed, like the frame

Decorations: vertical positions in vw (not %) so they stay fixed when viewport height shrinks, matching the central frame

Decorations: hide corner decos when viewport height < 800px so they don't pile up when the frame can't fit either

Decorations: explicit deco-frame / deco-overlay / deco-portrait roles fix multi-image stacking and zero-height single slots

Decorations responsive: width-only sizing in vw/px + height auto from image aspect ratio (no more box distortion)

Revert shutter sound feature — keep silent capture for now

Shutter click via robot speaker: render WAV with OfflineAudioContext, pre-upload at boot, play via reachy on flash

Add synthesised shutter click via WebAudio, fired in sync with the gold-star flash

Post-capture celebration: await pre-shot end → 1s transition → success2 with synced audio → return to base

Play enthusiastic1 emotion after capture (both modes) — Reachy celebrates while photo is shown

Animation: restore gold star flash between countdown end and capture (~180ms beat)

Animation: 1s goto-first-pose, then move + 3s countdown in sync, capture at t=3 (no flash)

Animation: switch to move-2 for the pre-shot choreography

Animation: start countdown at the beginning of the move, capture lands ~2.25s in

Use custom cdeplanne/photobooth/move-1 for pre-shot anim instead of generic emotion

Magic Start: re-enable the button on success path so it's clickable again after returning to menu

Key validation: clear inline error on every retry/edit so stale messages don't linger

Validate Gemini key at Magic Start click — ping /models endpoint, block UI until verdict, inline error message

Icon: switch from cyan/mint to gold-on-brown to match the photobooth DA

DA: middle ground center frame width (61vw, max 930px)

DA: shrink center frame slightly (max 880px / 58vw) so Change-mode link gets breathing room

DA: widen container so center frame isn't cropped + slightly shrink corner decorations

Convert deco assets PNG → WebP (27 MB → 2.5 MB, 90% reduction)

DA: drop the SAY REACHEESE banner + grow center frame to 980px/64vw + enlarge corner decorations

Mode B: pre-upload greeting audio to robot at prefetch time so click→speech is just the play handshake (~200ms)

Mode B: re-prefetch on setUserName + await pending prefetch in say() before falling through

Mode B: fetch first name from HF whoami + pre-cache greeting TTS when key is saved

Mode B: warmer personalised greeting using HF username + casual phrasing

Mode B: skip Gemini chat call for fixed greeting (saves 1-3s) + remove .container overflow (centred scrollbar bug)

Menu: stop stretching cards to full viewport height — restore natural size, push down with margin-top

Menu: restore card frames + hide corner decorations on menu view only

Menu: drop card frames + center cards in viewport + replace Magic wand with 3-star sparkle

Lock page scroll: viewport fixed 100dvh, container scrolls internally if content overflows

DA: pull corner decorations further inward (8%) and shrink slightly

DA: pull corner decorations inward (4-5% margins) for better breathing room

DA overhaul: photobooth corner decorations + gold SVG icons replace emojis + slim hero + mobile responsive pass

Mode B: route 'Hold still' through robot-speaker path (was using browser-only speak)

Gemini TTS: try 2.5 (free-tier accessible) first, fall back to 3.1 preview

Mode B: route TTS through Gemini + play on robot speaker (gemini-3.1-flash-tts-preview Puck voice)

Mode B: real multi-turn dialog with Gemini text chat — greeting + follow-up questions + ready signal

Mode B: dump all available TTS voices on first speak() for diagnosis

Mode B TTS: pick natural/neural voice when available + livelier pitch + script tweaks

Mode B speak(): dedupe voiceschanged-vs-timeout race that interrupted first utterance