filter_template references were silently ignored in PP template test,
picture source test, and KC target test routes — they created a no-op
FilterTemplateFilter instead of expanding into the referenced template's
filters. Centralized expansion logic into PostprocessingTemplateStore.
resolve_filter_instances() and use it in all test routes + live stream
manager.
Also fixed empty template_id when adding filter_template filters: the
select dropdown showed the first template visually but onchange never
fired, saving "" instead. Now initializes with first choice's value and
auto-corrects stale/empty values at render time.
Other fixes: ScreenCapture dimensions now use actual image shape after
filter processing; brightness source label emoji updates.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Use i/(count-1) fraction (matching calibration dialog) so LEDs span
the full edge, and apply seg.reverse flag for correct numbering.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
audio_capture.py:
- Move _fft_bands from inner function to method (avoid per-frame closure)
- Pre-allocate channel split buffers and RMS scratch arrays
- Use in-place numpy ops (np.copyto, np.multiply) instead of copies
- In-place FFT smoothing instead of temp array allocation
- Cache loop-invariant values as locals
- Fix energy index to wrap-around instead of unbounded increment
audio_stream.py:
- Pre-compute interpolation arrays (band_x, led_x, full_amp, indices_buf,
vu_gradient) once on LED count change instead of every frame
- Pre-compute VU meter base/peak float arrays in _update_from_source
- Reuse full_amp and indices_buf buffers across frames
- In-place spectrum smoothing to avoid temp allocations
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The processing loop skipped resending when the CSS frame was
unchanged (static source). With a dynamic brightness value source
(e.g. audio), brightness can change every frame while colors stay
the same. Now compare effective brightness against previous value
and resend when it differs.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add resolve_audio_source() that accepts both MultichannelAudioSource
(defaults to mono mix) and MonoAudioSource. Update CSS and brightness
value source dropdowns to show all audio sources with type badges.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Pop send_latency_ms and rgbw from kwargs in WLED, Adalight, and
AmbiLED providers so mock-only fields don't leak through to
non-mock client constructors.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Instrument AudioCaptureStream with read/FFT timing and
AudioColorStripStream with render timing. Display audio-specific
timing segments (read/fft/render/send) in the target card
breakdown bar when an audio source is active.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Virtual device with configurable LED count, RGB/RGBW mode, and simulated
send latency. Includes full provider/client implementation, API schema
support, and frontend add/settings modal integration.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Stream real-time LED colors from running WLED targets to the browser via
binary WebSocket (RGB bytes, throttled to ~15 fps). Toggle button on
target card opens a compact canvas strip that renders each frame using
ImageData. Cached last frame is re-rendered after card reconciliation to
prevent flicker during auto-refresh.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add a 💡 button in the calibration modal header (CSS mode only) that
toggles the LED overlay visualization. Auto-stops overlay on modal close
if started from the dialog. Checks and reflects current overlay status
on modal open.
Fix serial devices creating a new connection on every edge test toggle,
which triggered Arduino bootloader resets. Now reuses the cached idle
client for all device types.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New source_type "api_input" allows external clients to push raw LED
color arrays ([R,G,B] per LED) via REST POST or WebSocket. Includes
configurable fallback color and timeout for automatic revert when no
data is received. Stream auto-sizes LED count from the target device.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace single "adaptive" type with adaptive_mode sub-selector by two
distinct source types in the dropdown. Removes the adaptive_mode field
entirely — the source_type itself carries the mode. Clearer UX with
"Adaptive (Time of Day)" and "Adaptive (Scene)" as separate options.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New "adaptive" value source type that automatically adjusts brightness
based on external conditions. Two sub-modes: time-of-day (schedule-based
interpolation with midnight wrap) and scene brightness (frame luminance
analysis via numpy BT.601 subsampling with EMA smoothing).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Background task samples system (CPU/RAM/GPU) and per-target (FPS/timing) metrics
every 1s into a 120-sample ring buffer (~2 min). New API endpoint
GET /system/metrics-history returns the buffer. Dashboard charts now seed from
server history on load instead of sessionStorage, surviving page refreshes.
Also removes emoji from brightness source labels.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Extend value source brightness modulation to Key Colors targets (matching LED target support).
Also fix stream subtab CSS selector collision that broke target subtab selection, and use 🔢 emoji
for value source UI elements.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Introduces a new Value Source entity that produces a scalar float (0.0-1.0)
for dynamic brightness modulation. Three subtypes: Static (constant),
Animated (sine/triangle/square/sawtooth waveform), and Audio-reactive
(RMS/peak/beat from mono audio source). Value sources can be optionally
attached to LED targets to control brightness each frame.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Segments are redundant now that the "mapped" CSS type handles spatial
multiplexing internally. Each target now references one color_strip_source_id
instead of an array of segments with start/end/reverse ranges.
Backward compat: existing targets with old segments format are migrated
on load by extracting the first segment's CSS source ID.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Audio sources moved to separate tab with dedicated CRUD API, store, and editor modal
- New "mapped" color strip source type: assigns different CSS sources to distinct LED sub-ranges (zones)
- Mapped stream runtime with per-zone sub-streams, auto-sizing, hot-update support
- Target editor auto-collapses segments UI when mapped CSS is selected
- Delete protection for CSS sources referenced by mapped zones
- Compact header/footer layout
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New scrcpy/ADB capture engine that captures Android device screens over
ADB using screencap polling. Supports USB and WiFi ADB connections with
device auto-discovery. Engine-aware display picker shows Android devices
when scrcpy engine is selected, with inline ADB connect form for WiFi
devices.
Key changes:
- New scrcpy_engine.py using adb screencap polling (~1-2 FPS over WiFi)
- Engine-aware GET /config/displays?engine_type= API
- ADB connect/disconnect API endpoints (POST /adb/connect, /adb/disconnect)
- Display picker supports engine-specific device lists
- Stream/test modals pass engine type to display picker
- Test template handler changed to sync def to prevent event loop blocking
- Restart script merges registry PATH for newly-installed tools
- All engines (including unavailable) shown in engine list with status flag
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Audio capture now produces per-channel FFT spectrum and RMS alongside
the existing mono mix. Each audio color strip source can select which
channel to visualize via a new "Channel" dropdown. This enables stereo
setups with separate left/right segments on the same LED strip.
Also shows the device LED count under the device selector in the target
editor for quick reference.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Each target now has a segments list where each segment maps a color strip
source to a pixel range (start/end) on the device with optional reverse.
This enables composing multiple visualizations on a single LED strip.
Old targets auto-migrate from the single source format on load.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Add new "audio" color strip source type with three visualization modes
(spectrum analyzer, beat pulse, VU meter) supporting WASAPI loopback and
microphone input via PyAudioWPatch. Includes shared audio capture with
ref counting, real-time FFT spectrum analysis, and beat detection.
Improve all referential integrity 409 error messages across delete
endpoints to include specific names of referencing entities instead of
generic "one or more" messages.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Composite sources stack multiple existing color strip sources as layers
with configurable blend modes (Normal, Add, Multiply, Screen) and per-layer
opacity. Includes full CRUD, hot-reload, delete protection for referenced
layers, and pre-allocated integer blend math at 30 FPS.
Also eliminates per-frame numpy allocations in color_strip_stream,
effect_stream, and wled_target_processor (buffer pre-allocation).
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Static, gradient, color cycle, and effect streams depend on LED count
and were being reconfigured per-consumer when shared. Now only picture
streams (expensive capture) are shared; count-dependent sources get
per-consumer instances keyed by css_id:target_id.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
LED skip: set first N and last M LEDs to black on a target. Color sources
(static, gradient, effect, color cycle) render across only the active
(non-skipped) LEDs. Processor pads with blacks before sending to device.
Rename standby_interval → keepalive_interval across all Python, API
schemas, and JS. from_dict falls back to old key for existing configs.
Remove legacy migration functions (_migrate_devices_to_targets,
_migrate_targets_to_color_strips) and legacy fields from target model.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Filter Template meta-filter: reference existing PP templates inside others
for composable, DRY filter chains. Filters are recursively expanded at
pipeline build time with cycle detection. New `select` option type with
dynamic choices populated by the API.
Keepalive optimization: serial devices (Adalight, AmbiLED) don't need
keepalive — they hold last frame indefinitely. Check `standby_required`
capability at processor start, skip keepalive sends for serial targets,
and hide keepalive metrics in the UI. Rename "Standby Interval" to
"Keep Alive Interval" throughout the frontend.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
New features:
- Procedural effect source type with fire, meteor, plasma, noise, and aurora algorithms
using palette LUT system and 1D value noise generator
- 12 predefined gradient presets (rainbow, sunset, ocean, forest, fire, lava, aurora,
ice, warm, cool, neon, pastel) selectable from a dropdown in the gradient editor
- Auto-crop filter: min aspect ratio parameter to prevent false-positive cropping
in dark scenes on ultrawide displays
Optimization:
- Static/gradient sources without animation: stream thread sleeps 0.25s instead of
frame_time; processor repolls at frame_time instead of 5ms (~40x fewer iterations)
- Inverted isinstance checks in routes to test for PictureColorStripSource only
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Backend: add batch endpoints for target states, metrics, and device
health to replace O(N) individual API calls per poll cycle.
Frontend: use batch endpoints in dashboard/targets/profiles tabs,
fix Chart.js instance leaks, debounce server event reloads, add
i18n active-tab guards, clean up ResizeObserver on pattern editor
close, cache uptime timer DOM refs, increase KC auto-refresh to 2s.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add dynamic "Hardware max ≈ N fps" recommendation below FPS slider,
computed from LED count (WLED: protocol timing) or baud rate (serial).
Reuses shared _computeMaxFps from devices.js with named constants.
- Fix gradient looking different across targets sharing the same stream:
configure() now uses max LED count across all consumers; _fit_to_device
uses np.interp linear interpolation instead of truncate/tile.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Replace per-frame sleep(remaining) with absolute next_frame_time
tracking so asyncio.sleep() overshoots are recovered in subsequent
frames, keeping average FPS on target. Skip first FPS sample to
avoid ~2000+ spike from near-zero init interval.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
All three non-picture color strip stream types had their animation
loops hardcoded at 30 FPS and lacked set_capture_fps(), so target
FPS changes had no effect. Now each stream reads self._fps per
iteration and exposes set_capture_fps() for the stream manager.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Replace animation Enable checkbox with None option in effect selector;
show effect description tooltip; disable speed slider when None selected
- Allow target FPS range 1-90 (was 10-90) across UI and backend validation
- Scope serial COM connections to target lifetime (no idle caching);
use temporary connections for power-off/test mode
- Fix serial black frame on stop: flush after write, delay after task
cancel to prevent race with in-flight thread pool write
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Remove static/idle color from entire stack (storage, API, processing, UI, CSS, locales)
- Simplify device power button to turn-off only (send black frame, no toggle)
- Send black frame on serial port close (AdalightClient.close)
- Fix settings modal serial port dropdown showing WLED devices due to stale deviceType
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
FPS is a consumption property (how fast to send to a device), not a
production property. Two targets sharing the same source may need
different FPS. This moves the fps field from PictureColorStripSource
to WledPictureTarget across the full stack.
The capture stream now auto-adjusts its rate to max(all connected
target FPS values) via ColorStripStreamManager tracking per-consumer
FPS. UI updates: FPS slider in target editor, FPS badge on target
cards, LED count repositioned in CSS editor, consistent speed icons.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
When two targets share the same color strip source, configure() resizes
the stream to the last caller's LED count. If a WLED device (934 LEDs)
starts after a serial device (fewer LEDs), the serial target sends the
full 934-LED frame over serial, massively inflating send time.
Add _fit_to_device() to truncate/tile colors to the target's actual
device LED count before sending, so each consumer only transmits what
its device needs regardless of the shared stream's current size.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Non-picture sources (static, gradient, color_cycle) returned empty
timing from get_last_timing(), causing timing_total_ms to be null and
hiding the entire timing section in the UI. Now timing_total_ms falls
back to send_ms when no CSS pipeline timing exists. Frontend timing
bar/legend segments are conditionally rendered to avoid null labels.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
The GET /devices/{id}/brightness endpoint was making an HTTP request to
the ESP32 over WiFi on every frontend poll (~3s), causing 150ms async
event loop jitter that froze the LED processing loop. Cache brightness
server-side after first fetch/set, add frontend dedup guard, reduce
get_device_info() frequency, and add processing loop timing diagnostics.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Change /api/v1/system/performance from async def to def so FastAPI
runs the blocking psutil + NVML GPU queries in a thread pool instead
of freezing the event loop (polled every 2s by dashboard)
- Batch profile engine's 3 separate run_in_executor detection calls
into a single _detect_all_sync() call, reducing event loop wake-ups
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Add 5 new animation effects (strobe, sparkle, pulse, candle, rainbow
fade) to both static and gradient color strip streams
- Fix FPS drops (30→25) by using 5ms re-poll on frame skip instead of
full frame_time, preventing synchronization misses between animation
thread and processing loop
- Double-buffer animation output arrays to eliminate per-frame numpy
allocations and reduce GC pressure
- Use uint16 integer math for gradient brightness scaling instead of
float32 intermediates
- Update animation type dropdowns and locale strings (en + ru)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
- Fix WGC capture_frame() returning stale frames (80k "frames" in 2s)
by tracking new-frame events; return None when no new frame arrived
- Add draw_border config passthrough with Win11 22H2+ platform check
- Add high_resolution_timer() utility (timeBeginPeriod/EndPeriod)
- Switch all processing loops from time.time() to time.perf_counter()
- Wrap all loops with high_resolution_timer() for ~1ms sleep precision
- Add animation speed badges to static/gradient color strip cards
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
BetterCam/DXcam engines now detect when the DXGI Desktop Duplication
interface is lost (display mode change, sleep/wake, UAC prompt, etc.)
and automatically reinitialize the camera with a 3-second cooldown
between attempts, instead of error-looping indefinitely.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
configure() was only called for Static and Gradient streams, leaving
ColorCycle at its default led_count=1 — all other LEDs sent as black.
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- color_cycle is now a top-level source type (alongside picture/static/gradient)
with a configurable color list and cycle_speed; defaults to full rainbow spectrum
- ColorCycleColorStripSource + ColorCycleColorStripStream: smooth 30 fps interpolation
between user-defined colors, one full cycle every 20s at speed=1.0
- Removed color_cycle animation sub-type from StaticColorStripStream
- Color cycle editor: compact horizontal swatch layout, proper module-scope fix
(colorCycleAdd/Remove now exposed on window, DOM-synced before mutations)
- Animation enabled + Frame interpolation checkboxes use toggle-switch style
- Removed Potential FPS metric from targets and KC targets metric grids
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Frame interpolation filter (frame_interpolation):
- New PostprocessingFilter with supports_idle_frames = True
- Backward-blend algorithm: blends frame N-1 → N over one capture
interval, producing smooth output on idle ticks at ≤1 frame of lag
- Detects new vs idle frames via cheap 64-byte signature comparison
- No options; registered alongside other built-in filters
ProcessedLiveStream idle-tick support:
- Detects supports_idle_frames filters at construction (_has_idle_filters)
- target_fps returns 2× source rate when idle filters are present
- _process_loop runs at 2× rate; idle ticks copy cached source frame
and run full filter chain, publishing result only when a filter
returned actual interpolated output (not a pass-through)
- Pass-through idle ticks leave _latest_frame unchanged so consumers
correctly deduplicate via object identity
KC target hot-settings:
- brightness, smoothing, interpolation_mode now read from self._settings
each frame instead of captured as stale locals at loop startup
- Changes take effect within one frame without stop/restart
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
Introduces a new 'static' source type that fills all device LEDs with a
single constant RGB color — no screen capture or processing required.
- StaticColorStripSource storage model (color + led_count=0 auto-size)
- StaticColorStripStream: no background thread, configure() sizes to device
LED count at processor start; hot-updates preserve runtime size
- ColorStripStreamManager dispatches static sources (no LiveStream needed)
- WledTargetProcessor calls stream.configure(device_led_count) on start
- API schemas/routes: source_type Literal["picture","static"]; color field;
overlay/calibration-test endpoints return 400 for static
- Frontend: type selector modal, color picker, type-aware card rendering
(🎨 icon + color swatch), LED count field hidden for static type
- Locale keys: color_strip.type, color_strip.static_color (en + ru)
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Target overlay works without active processing: route pre-loads calibration
and display info from the CSS store, passes to processor as fallback
- Fix server crash on repeated overlay: replace per-window tk.Tk() with single
persistent hidden root; each overlay is a Toplevel child dispatched via
root.after() — eliminates Tcl interpreter crashes on Windows
- Fix edge test colors not lighting up: always call set_test_mode regardless
of processing state (was guarded by 'not proc.is_running'); pass calibration
so _send_test_pixels knows which LEDs map to which edges
- Fix device reset on overlay stop: keep idle serial client cached after
clearing test mode; start_processing() already closes it before connecting
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
- Add explicit led_count to PictureColorStripSource (0 = auto from calibration)
- Stream pads with black or truncates to match led_count exactly
- Calibration dialog: show led_count input above visual editor in CSS mode
- Calibration dialog: pre-populate led_count with effective count (cal sum) when stored value is 0
- Calibration dialog: sync preview label live as led_count input changes
- CSS editor: group brightness/saturation/gamma into collapsible "Color Corrections" section
Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
alexei.dolgolyov