Editing corners with handles on the output is unmanageable on a large LED
wall (7x4 m): the hand has to work on the giant screen. The Mappatura tab
now has a scaled replica of the output (aspect from the output display or
the LED area) where zones are dragged by corner or whole, with live sync
to the output (one IPC per frame). Keyboard on the focused editor: arrows
nudge 1 output pixel, Shift 10 px, Tab cycles corners, Esc = whole zone;
keys never leak to the global hotkeys. Selection is shared with the zone
list and highlighted on the output handles; a status line shows pixel
coordinates.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XVMQ2JhsJ77pGKqcvvFHqc
New controls in the Registra tab: a 'Posizione taglio' slider choosing WHICH
strip of the screen survives the cover-crop when the format differs from the
display (ffmpeg crop x/y driven by cropX 0..1), and a guide checkbox. The
guide is a dashed frame + dimmed surroundings drawn on the output window
(outside #stage, last in DOM, overlay only — composeFrame never captures
it), live-updated from slider/format and hidden when no crop applies.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XVMQ2JhsJ77pGKqcvvFHqc
captureStream at 60fps with an 80 Mbps cap. When the requested frame equals
the source and the first pass is H.264, the MP4 is a pure remux (-c copy):
the hardware realtime encode is the only compression in the final file. The
re-encode path (vertical reels, VP9 fallback) moves to CRF 16 preset slow at
constant 60fps.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XVMQ2JhsJ77pGKqcvvFHqc
MediaRecorder in Electron 33 muxes MP4 with avc1 High profile + AAC; on
macOS that encode runs on VideoToolbox hardware, whose realtime quality
ceiling is far above software VP9 — the residual smearing came from there.
Bitrate cap raised to 50 Mbps (hardware barely notices; the VP9 fallback
undershoots regardless). ffmpeg sniffs containers by content, so the .webm
temp name keeps working.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XVMQ2JhsJ77pGKqcvvFHqc
The realtime VP9 first pass at 12 Mbps was the bottleneck on full-screen
abstract motion (visible blocking the MP4 pass could not recover), and the
x264 pass added its own loss at CRF 20 veryfast. The WebM intermediate is
deleted after muxing, so it can afford 30 Mbps; the MP4 pass is offline, so
it can afford CRF 18 preset medium, constant 30 fps and 256k AAC.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XVMQ2JhsJ77pGKqcvvFHqc
ffmpegPath() short-circuited on the bare 'ffmpeg' candidate, so the absolute
Homebrew/system paths were never checked. From a terminal PATH covers it, but
the packaged app launched from Finder has a minimal PATH: stopping a
recording failed with 'spawn ffmpeg ENOENT'. Absolute candidates now win;
bare 'ffmpeg' is the last resort.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01XVMQ2JhsJ77pGKqcvvFHqc
A cue aimed at a connected display whose aux window was off landed nowhere
and the monitor stayed black. The cue now switches that output on (persisted,
row updates in the panel) and 'auxReady' re-sends the scene effect once the
freshly created window has loaded.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A playlist made only of ✨ Intermezzo scenes is the visuals-only sequencer
(effects/text/images on a timeline, chained and looping with repeat), but
playSilence tore down the live input source, so effects stopped reacting to
the DJ's real music. Now only file/track audio is silenced; a mic/line/
BlackHole input keeps feeding the analyser through every scene.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The per-monitor overlay text now reuses the main output's ticker wholesale
(same DOM structure and style.css classes): direction h/vertical/side-edges,
letter effects (wave/zoom/flash/rotate/updown), the 8 fonts, position, size,
weight, color, speed, and the audio-reactive glow — the complete Testo tab
option set. Scene 'effect' elements gain a 'Su' selector (main output / one
aux display / all screens); each target is an independent cue channel so
same-time cues aimed at different monitors all fire, and aux windows in
follow mode ignore the all-screens broadcast (they already follow).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Each aux row gets a 🔤 toggle opening a sub-row: text, position
(top/middle/bottom), static or marquee, size, speed, color, bold. The text
renders as an overlay above whatever the window shows — effect, image or
plain black — so a monitor can be a dedicated message board. Scroll duration
scales with text length for constant apparent speed.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The control panel's global 'select { width: 100% }' made the mode dropdown
swallow the whole row, pushing the family/preset selects off-screen.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New 'Uscite multiple' section in the Schermo tab: every display that hosts
neither the panel nor the main output gets a row with an on/off toggle and a
content mode — follow the main effect, a dedicated effect (family+preset,
camera/3D families excluded), a fullscreen image, or black. Aux windows are
lightweight visual-only renderers (aux.html/aux.js) fed with the audio
analysis relayed from the main output at ~30 Hz over the new 'afr' channel,
with a drift-corrected clock. Display hot-plug is now detected (list refresh
+ stale aux windows dropped); config persists per display id in localStorage.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Every 400ms progress tick rebuilt the whole list (now heavy with per-track
waveforms), so nodes were replaced between mousedown and mouseup and row
buttons never received their click while a track was playing. renderPlaylist
now computes a skeleton signature and, when unchanged, refreshes times,
progress bar and waveforms on the existing nodes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A lost dragend (dragged node removed from the DOM) left dragFrom/cueDragging
stuck, permanently deferring renderPlaylist: the list stopped redrawing and
row buttons acted on stale indexes (delete/play appeared dead). The guard now
unlocks itself when no .dragging element exists, and row actions resolve the
track index at click time by object identity.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
'Disattiva' in the Animazioni 3D tab restores the last non-3D preset
(tracked by applyEffect), answering 'how do I turn the 3D output off'
without hunting through the effect library.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New 'Zona Effetto': each zone runs its own preset on an independent
offscreen Visualizer (960x540, own WebGL context), stepped with the
shared clock and audio so every panel dances to the same music with a
different effect (fluid presets included; interactive/3D excluded).
Zone row gains family/preset selectors; engines are created/destroyed
as zones change and their canvases uploaded as zone textures each
frame. Guide updated with the performance note.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
A details/summary walkthrough: what mapping is, the three steps
(activate, create zones, drag corners on the output), every per-zone
option with a multi-panel Porzione example, and the persistence/
recording notes.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New Mappatura tab + mapping.js engine: each zone is a quad warped with a
projective homography (square->quad closed form, inverted per fragment)
on a dedicated WebGL2 canvas above the visual, black outside. Sources:
the live visual (with a source sub-rect for multi-panel LED walls) or an
image file (corner-pin onto physical surfaces). Edit mode draws handles
ON THE OUTPUT: corners (or whole quads) are dragged while looking at the
real projection, edits flow back to the panel which persists them. Zone
list with name, source portion, opacity and delete; included in
recordings; re-sent when the output restarts.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
color-scheme: dark makes Chromium render the native select popups and
scrollbars dark on Windows (they defaulted to white with our light text
on top), plus explicit dark backgrounds on select/option as a fallback.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- toFileURL and path-basename handling made Windows-safe (backslashes,
drive letters) so media/model loading works there
- Blender discovery scans Program Files\Blender Foundation; ffmpeg
candidates include common Windows paths
- build config: win/nsis target (x64, custom install dir), build/icon.ico
generated from the 1024 png, dist:win script
Cross-compiled from macOS: dist/DJ Visualizer Setup 1.6.0.exe (95MB) with
scripts unpacked and models/MediaPipe bundled. Untested on real Windows.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The fixed front-axis rotation degenerated on poses with the arms bent
forward (vector nearly parallel to the axis — why the mannequin showed
no effect on salsa-like moves). The swing axis is now computed per
frame as cross(current arm direction, character outward side), so the
arm is always pushed away from the torso regardless of the pose, and
the rotation naturally vanishes once the arm already points outward.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Dance clips are authored for the standard mannequin: on chunkier models
the same joint angles drive the arms through the torso (the ninja is
clean because its proportions match). New 'Apertura braccia' slider
(0-35deg, saved per model path): the upper arms get an extra outward
rotation about the world Z axis during clip playback (left +, right -),
solved back into each bone's local frame through the parent rotation.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Every loaded model (GLB or converted FBX) is added to 'I miei modelli'
in the Animazioni 3D tab: click a name to switch the dancer instantly
(active one highlighted), x removes it from the list (file untouched).
Persisted in localStorage; the current model is seeded on first run.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The hard clamp made out-of-scale clips ride its boundary and jerk the
model left/right. Hips translations now stay pinned horizontally (the
framing compensation cancelled that motion anyway — the moves live in
the legs and torso) and the vertical bounce saturates smoothly via tanh
at 0.8x the rest length. Same-rig baked clips render identically since
their root motion was already being compensated.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New FBX models could periodically vanish: foreign clips with
out-of-scale hip translations threw them past the camera frustum, then
the loop brought them back. Three guards: the unit-rescale triggers at a
1.75x ratio (was 3x), the sampled translation offset is hard-clamped to
1.2x the bone rest length, the residual vertical bounce is capped at
half the model radius, and the near/far planes widened (0.02-8x dist).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Some Mixamo clips carry translation keyframes on every bone; bound to a
model with different proportions they stretch/crush the limbs (bone
lengths must come from the target skeleton). Library binding now keeps
translations only for the Hips (root motion); rotations drive the rest.
Clips baked into a model keep their full channels.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The model button now accepts .fbx too: a new model:convert IPC runs
Blender with scripts/fbx2modelglb.py (mesh + skin kept, animations
dropped, embedded diffuse recovered even when the FBX references
missing external textures) into userData/models3d, then the resulting
GLB loads as usual. anim/model conversion share one convertFbx helper.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
'Modello 3D', 'Interattivo Modello' and 'Interattivo Avatar' no longer
appear in the Effetti family filter or the preset list: the new
'Effetto 3D' block in the Animazioni 3D tab has its own family/preset
selectors (palette · variant) and an Attiva button, applying through the
same applyEffect path. Saved pads/sequences/scenes keep working (preset
indices untouched).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The Auto VJ pill style (hidden checkbox, ragged flex wrap) didn't fit
the long clip names: #glb-anims now lays out as a two-column grid with
visible aligned checkboxes, ellipsis on overflow and green accent.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The 3D model controls outgrew the SVG section: model load + backdrop,
dance repertoire, extra animations and manual BPM/TAP now live in their
own tab between Effetti and Pad. All element ids unchanged, wiring
untouched.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
- models/dance_library.glb: skeleton-only GLB with the 13 Mixamo dances,
loaded at startup and bound BY BONE NAME to whatever rigged GLB the
user loads — new models dance immediately, no baking needed.
- Repertoire UI (Media tab): checkbox chips of the available clips, the
director only picks the ticked ones (none = all); persisted.
- 'Aggiungi animazioni': extra .glb clips load directly; .fbx are
converted on the fly by Blender via a new anim:convert IPC and the
bundled scripts/fbx2animglb.py (asarUnpacked for packaged builds).
- Manual BPM with TAP tempo: 0 = follow the detected beat, otherwise a
pure metronome drives the dance clock.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The uber-shader pass is now reusable (_renderUber): choosing 'fam:<i>'
in the Sfondo Modello 3D selector renders that family first (tinted with
the active preset's palette/hue/sat) and overlays the model on top. The
selector lists every family except fluids (already explicit options),
interactive and Modello 3D itself.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
New 'Sfondo Modello 3D' selector (Media tab): gradient or any of the
five fluid modes. The fluid sim paints the canvas first, then the model
renders on top clearing only the depth buffer (overlay mode skips its
own backdrop pass). Palette and speed are shared with the preset;
persisted and re-sent when the output restarts.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Expressive choreographies (Thriller, breakdance, flair, spins, moonwalk)
are authored one count per TWO beats: a per-clip pace factor (name
heuristic) halves their count rate. The beat clock also gained a
refractory window (60% of a period) so energetic tracks firing on both
kick and snare no longer double the detected tempo, which made every
clip race.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The pure exponential follower trailed the beat clock by its time
constant (~110ms), which read as the dancer being late on the kick. The
clip timeline now advances by itself at the metronome rate (feed-forward,
zero structural lag) while a gentler follower only corrects drift, with
a 50ms lead compensating beat-detection/display latency and a hard
resync when the error exceeds 1.5 beats.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The phase-locked playback felt mechanical: the smoothstepped phase made
the clip accelerate/brake every beat, and the hard clamp at phase 1
stalled motion when the next beat ran late. The beat clock now exposes a
linear phase that keeps running past 1, and the director chases the
beat-locked timeline with an exponential follower (~110ms) that absorbs
metronome re-syncs. The procedural sway keeps its eased phase (there the
easing reads as groove).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Clips used to play at their authored tempo, so the steps drifted against
the music. A shared beat clock (detected beats + free-running metronome
in the gaps) now drives the clip timeline directly: one choreography
count (0.5s of clip, ~120bpm authoring convention) per music beat, with
a smoothed phase inside each beat. Steps land on the kick at any track
BPM; the variant speed doubles the count rate; move switches and the
procedural dance reuse the same clock.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The global speed control multiplies the visual time, which froze the
dance clips at low values (only max speed showed motion). Clip playback,
the dance director and the procedural dance now run on a real-time
clock — only the preset variant speed scales the moves. The translation
unit-rescale also samples the peak of the first keyframes (robust when a
clip starts near the origin) and clamps the factor.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Recovers the embedded diffuse when the FBX references missing external
textures, discards animation-file objects keeping only their actions,
retargets bone-name prefixes (mixamorig vs mixamorig1) so clips recorded
for another character land on the base rig, and exports all clips as NLA
tracks for the dance director.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Clips merged from other Mixamo FBX files can carry hip translations in
a different unit (cm vs m): translation channels are auto-rescaled
against the node's rest pose (first-keyframe ratio > 3x). The dancer
also stays framed: the clip's root motion is cancelled horizontally
(hips world position vs rest) keeping 40% of the vertical bounce, so
travelling moves like moonwalks don't slide out of the camera.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
GLBs can carry several animation clips (merged from Mixamo FBX files
with scripts/merge via Blender NLA tracks). The director loops the
current move and switches to a random different one every 16 beats (or
~14s without beats) with a 0.45s pose crossfade (per-node quaternion
nlerp + translation lerp).
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Two paths for skinned GLBs with no live body driving them:
- animation clips in the GLB (e.g. Mixamo moves) are parsed (rotation/
translation channels, nlerp sampling) and looped at the effect speed;
- without clips, a beat-locked procedural dance drives the FK targets:
arms pump on alternate beats, hips sway, head nods, knees bounce, with
a free-running metronome keeping the groove between detected beats and
bass scaling the amplitude.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
The GLB loader now reads skins (JOINTS_0/WEIGHTS_0, inverseBindMatrices,
node hierarchy kept instead of baked); a skinned shader variant blends 4
joint matrices per vertex (up to 80 bones). _computeJoints solves FK
retargeting: for each mapped bone (mixamorig arms/forearms/up-legs/legs/
neck/spine) an extra local rotation aligns its chain child with the
world direction measured between the tracked landmarks — the user's
right side drives the character's Left bones (mirror). Interattivo
Avatar: when the loaded GLB has a skeleton it drives the rig (bounding
box from CPU-skinned samples, pelvis-follow track), else the procedural
armoured robot. Validated with a Mixamo Y-Bot: synthetic raised-arm pose
reproduced exactly.
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
