xLights – Light sequencer and Show scheduler

2026.15 — August 4, 2026

Enhancements

dkulp — Layout: controllers can be shown as a movable, resizable box in the 2D or 3D preview so they can be placed where they physically sit. Set via Show on Layout on the Controllers tab (Off by default); Off / Controller Tab Only / Layout Panel / Always control where the box appears. Position, size and rotation are editable in the controller properties alongside an optional name label (off by default)

cybercop23 — Added a View > Toolbars menu to independently show/hide each dockable toolbar (Main, Play, Windows, Paste, AC, View, Effects); visibility persists across restarts. The old standalone AC Lights Toolbar checkbox now lives here, grouped next to Show AC Ramps

cybercop23 — Preferences: the Random Effects and Toolbars pages now show each effect’s icon beside its name

cybercop23 — Layout: the Import Previews/Models/Groups dialog can now also import named Viewpoints (2D/3D camera presets).

cybercop23 — Layout: setting a model’s shadow target automatically links the target model’s start channel to the shadow model (@SHADOWMODEL:1) and clears its controller.
cybercop23 — Layout: added ability to freeform draw node selection and deselection for SubModels, Model Faces, and Model States panels
cybercop23 — Bulk Controller Upload: added controller upload status
cybercop23 — All Controller Columns dialog: reopens with the controller you last worked on selected
cybercop23 — Select Directories dialog now closes automatically after a successful change to the temporary or permanent show folder, or after opening the base show folder.
cybercop23 — Sequence Settings: the media trim fields accept longer values
heffneil — Replace Model(s): a Model Groups option decides what happens to the groups the replaced models belonged to – leave them alone, replace them with the source model’s groups, or merge the two
heffneil — Tools – Test: every tab now has a live name filter
AGFazio — Layout: a Controllers-tab preview right-click entry selects every model that has no controller assigned and highlights it in the preview
scott — Controller layout print preview: paper size and orientation are now chosen in the preview dialog itself rather than through the OS page-setup dialog
derwin12 — MovingHead: add save/recall presets for the Pattern tab, matching the existing Position and Dimmer presets
derwin12 — MovingHead: Left and Right join All/None/Evens/Odds for choosing which fixtures an edit applies to
derwin12 — Add an Other Preferences option for what double-clicking a model in the layout does.
derwin12 — The log now records how long a batch render took, whether it finished or was cancelled
mpl1337 — Node Layout grids can be zoomed from 10–300% or fitted to the window
dkulp — Fire effect has a new Style setting, and new Fire effects now default to “New Render Method”. It carries the flame from frame to frame so it rises and drifts like real fire instead of flickering in place, and renders roughly 10x faster again because the whole flame updates at once on the GPU (Metal on macOS/iPad, Vulkan on Windows/Linux). Existing sequences are migrated to “Old Render Method” so their look is unchanged
dkulp — FPP: support several LED panel matrices on one controller driven different ways at once. New “LED Panel Matrix – Hat/Cap/Cape” and “LED Panel Matrix – ColorLight” protocols name which kind a model targets, and the port number is the matrix number the controller’s LED Panels page shows. The original “LED Panel Matrix” still works and matches whatever is on that port
dkulp — Automation API: added getMedia, embedMedia and extractMedia so a script can inventory a sequence’s media and embed or extract it in bulk
dkulp — Automation API: added getMediaIssues and convertMedia so a script can find and convert media that isn’t playable on all platforms, without opening each sequence by hand
dkulp — Windows video decode: hardware decoding and the DirectX11 reader are now the default, and that reader converts and scales on the GPU with the colour space stated explicitly instead of leaving it to the system. Video-heavy sequences render 2-3x faster (the weaker the CPU, the bigger the gain); formats the reader cannot open still fall back to FFmpeg automatically
dkulp — Sequences open several times faster when they contain many file-based effects (Pictures, Video, Text, Shader, …): the per-effect file existence checks are now cached per load. On macOS each check was a round trip to the iCloud file provider, so picture-heavy sequences paid seconds re-checking the same files.
dkulp — Sequence load: file-based effects resolve their media once through the sequence’s media list, and the images are decoded in parallel on the background thread pool rather than one at a time. Images embedded in the .xsq are included, and an embedded multi-frame image no longer decodes every frame while the file is still being parsed. A sequence with a few hundred picture effects opens about twice as fast; rendered output is unchanged.
dkulp — Sequence load: embedded media no longer does a file-system path search per entry, which dominated the open of a sequence with a few thousand images
dkulp — Video: the decoder’s scaled-frame cache is now sized by a memory budget instead of a fixed 48 entries, so a clip reused by several effects at the same output size is decoded once rather than once per effect. Rendered output is unchanged.
dkulp — MacOS Video: a single large effect on a file made the decode-scale anti-alias headroom overshoot the source, so the file decoded at full native size and the scaling bought nothing, however small every other effect using it was. Such files now decode at the size actually rendered, and both decoder frame caches are bounded in bytes as well as frame count so a 4K file cannot cost 8x a 1080p one per cached frame. A sequence with 164 video effects over nine files, one of them 4K: peak memory 22GB -> 8GB, render CPU -35%
dkulp — Render: throttle frame concurrency when memory use approaches the limit the OS will kill the process at, and XL_RENDER_MEM=1 reports what a render is spending memory on
dkulp — Frame rendering, per-model buffers and every parallel_for now share one round-robin worker pool instead of running three pools of their own. The render nests all three, so the separate pools put about three times as many threads on the machine as there are cores, and the extra threads cost context switches rather than adding throughput. How much that was costing depends on the platform: macOS gains the most (render CPU down 20-30%, system time roughly halved), Windows a few percent. Output is byte-identical.
dkulp — Render: a row rendering frames in parallel now farms its whole run of frames to a round-robin worker pool instead of fixed-size windows with a barrier between them, so it is no longer resynchronised every few frames and clone memory follows the row’s concurrency, not the run length
dkulp — Render: a row carrying an effect whose state cannot be recomputed per frame (Strobe, Snowstorm, Fireworks and the like) no longer has its parallel run cut short – the state capture is now the head of a pipeline rather than a serial pre-pass, and the memory it needs is a fixed bound instead of one snapshot per frame. Output is unchanged.
dkulp — Model load, parallel output transmission and FPP Connect frame upload spread their work across the pool by index rather than by walking a linked list behind a lock, so they no longer serialise every worker on that lock. Output is byte-identical.
dkulp — Windows: Text and Shape effects now render through Direct2D/DirectWrite directly instead of via wxWidgets, roughly 3.6x faster on text-heavy sequences. Text using a font not installed on the machine may pick a different substitute than before
dkulp — Text effect renders much faster where the same text repeats: rendered text is now cached across rows and frames instead of per render buffer, so it survives the parallel render split (measured up to 4x faster on prop-heavy sequences on Windows). Output is unchanged.
dkulp — Pictures effect renders much faster: the image path is no longer re-resolved on every frame. On macOS that resolve reached iCloud and dominated prop-heavy sequences (one measured 7.7s -> 1.1s, and a 56-sequence show -5.9%). Output is unchanged
dkulp — Meteors: the Implode and Explode styles render several times faster on large buffers, use far less memory while rendering, and draw their trails across cores again (the other styles already did). Output is unchanged
dkulp — Ripple renders about 9x faster, and drawing to the render buffer got cheaper for every effect – VU Meter, SingleStrand, Marquee, Shape and Kaleidoscope all gained 10-38%. Output is unchanged.
dkulp — SingleStrand renders 15-25% faster. The settings it reads that cannot change during an effect are now decoded once instead of on every frame of every model, and the chase draw writes whole buffer columns directly. Output is unchanged.
dkulp — Shader: on Vulkan (Windows/Linux) the input upload now rides the frame’s own command buffer and the GPU wait is deferred until the pixels are actually needed, so rows overlap on the GPU instead of serializing. Renders a shader-heavy test sequence ~30% faster with identical output
dkulp — Sped up the number-aware name sort used by the model, group, view and import lists – it no longer copies and re-splits both names on every comparison (~2.3x on its own), and the Import Effects mapping tree no longer converts each name through wxString twice per comparison. Both showed up as multi-minute freezes while sorting a large mapping tree.
derwin12 — Replace Model(s): the candidate list is built once and filtered by hiding rows, instead of destroying and recreating every checkbox on each keystroke
- the dialog was unusably slow on a large show
dkulp — XL_GPU_SIZE_THRESHOLD diagnostic to sweep the GPU/CPU buffer-size break-even without a rebuild
dkulp — Crash reports: an exception that is not a C++ std::exception is now named rather than logged as “an unknown exception”, including the class and reason of a macOS/iOS system exception
dkulp — Each show folder now carries a random id in xlights_rgbeffects.xml so a show that submits many crash reports is counted once rather than once per report. It identifies the show only – no machine, user or location – and is written once when the show is first opened.
dkulp — The startup log now records the CPU model, physical/logical core counts, the attached displays and their scale factors, and (on macOS, which has no OpenGL renderer string) the GPU, so crash reports carry the hardware needed to reproduce threading and GPU issues
dkulp — Crash reports now include the startup machine configuration as its own machine_config.txt, so it survives the log rotating away – previously about 6% of reports arrived with no record of the machine at all
dkulp — Render: when a render or an abort will not finish, the log now names the models still outstanding along with the frame each one reached and what it is waiting on, instead of only counting them. A model stuck inside a frame-parallel window also reports the window, how much of it finished, and which frames are still being rendered. A render that stops making progress now says so on its own after 60 seconds rather than only when cancelled, headless included

Bug Fixes

dkulp — Fix crashes rendering after editing submodels. A model group holding a submodel kept pointing at the old one once its parent rebuilt them, so the next render could use a submodel that no longer existed
dkulp — Video: a GPU that cannot decode a file’s codec profile (common in a VM) reported it as a corrupt file and repeated the failed attempt, plus raw FFmpeg errors, for every video reader. It now names the codec and profile the hardware rejected, stops retrying that format, and FFmpeg’s own messages go to the log not the console
dkulp — Video: a narrow crop asked the decoder for a frame up to 100x the render buffer, which could exceed what a frame buffer can address and crash the render. The requested decode size is now bounded and an unusable one is refused
cybercop23 — Layout: Fix single model undo leaving stale tree pointers causing model to disappear from preview (#6817)
cybercop23 — Layout: exporting a model included every member of any model group referenced by the model, not just the one actually being exported; each exported model group is now restricted to the model included in that export.
cybercop23 — Layout: in the 3D preview, moving the mouse over a model while not editing models cleared the whole selection instead of only moving the highlight, and a click in empty space now deselects only when it was not the start of a camera drag. On the Controllers tab a single click selects the model under the pointer.
cybercop23 — Layout: a 3D object (.obj/.mtl) shipped with xLights is now found in the installed mesh library on whichever platform the show is opened on, so a layout using one survives moving between macOS and Windows
cybercop23 — Layout: reordering the columns of the model or controller list is now saved
kylegrymonprez — Layout: clicking the orange centre handle of a model in the 3D preview cycles the selection between overlapping models instead of opening the model’s edit dialog
AGFazio — Tree models now default to vertical strands
dkulp — Layout: “Set Center Offset Here” crashed when the group it was invoked on had gone away between the menu opening and the item being picked
dkulp — Layout: a polyline with no drop points placed its nodes at infinite coordinates
dkulp — Layout: crash when selecting in the model tree while models are being deleted or the tree is being rebuilt. Selection now skips tree items whose model has already been freed on every platform, instead of only on Linux
dkulp — Layout: fixed a crash while downloading or importing a model – the property grid refresh dispatched from the download progress dialog ran against the model the import had already replaced
dkulp — Fixed model groups keeping pointers to freed models: renaming a model through a replace left every group naming it pointing at the freed model, and rebuilding a model’s submodels (SubModels dialog save, export submodels to other models, custom model reverse) never repointed the groups that name them
dkulp — Fix a crash drawing a model preview when a graphics accumulator is null: the OpenGL and Metal draw paths now guard null like Vulkan already did
cybercop23 — Try fix floating sequencer panes shifting position on macOS after switching tabs (#6631)
derwin12 — Picking the perspective that is already selected now re-applies it, so it can be used to put the sequencer windows back where they belong
derwin12 — The timeline now rescales when a docked panel is shown or hidden, instead of keeping the hash-mark spacing it had at its previous width (#6765)
derwin12 — The timing-mark symbol pennant now draws at the bottom of the end mark (#6778) (#6790)
cybercop23 — The effect Fade In/Fade Out drop-downs list the sequence’s timing tracks again
derwin12 — Opening Sequence Settings no longer marks the sequence as changed unless something in it was actually changed (#6828) (#6833)
derwin12 — Fix typing negative numbers into effect slider/textbox pairs
MrPierreB — Effect drag: waveform time markers now show the full selection span rather than just the grabbed effect
MrPierreB — Effect drag: restore start/end/duration status bar text during move drag
MrPierreB — Clicking an effect in a multi-selection now narrows the selection to just that effect instead of keeping all effects selected.
derwin12 — Fix MovingHead Advanced: reset Path and Pattern on new effect
derwin12 — MovingHead: fix effect corruption copying and pasting between fixtures (#6806) (#6808), and the panel no longer paints over its own tab strip or comes up with the wrong height
derwin12 — Import Effects: two mapping rows sharing a name (a stacked duplicate) kept their insertion order only on macOS, so the tree sorted differently on Windows and Linux (#6835)
dkulp — Ripple: the Thickness setting now draws the thickness asked for. Every shape except Square grew its radius cumulatively across the thickness passes, so a thick ripple ballooned far past the intended size and mostly off the buffer. This changes how existing Ripple effects look.
dkulp — Text: vertical (Vert Up / Vert Down) XL-font text centres each character on its own glyph instead of left-justifying it, so a narrow character lines up with the wider ones stacked above and below it (#6789)
dkulp — Linux/iPad: fixed the font named in a Text or Shape effect being ignored when it has a multi-word name (stored quoted, e.g. ‘gill sans’) or was saved with a Mac character set, which fell back to a default font
dkulp — Shader: headless/batch rendering through the OpenGL path filled every Shader effect solid cyan – without a UI canvas the GL entry points were never loaded, so the capability check always failed. They now load the first time a render context is made current
dkulp — Shader: a shader that accumulates into gl_FragColor without first assigning it (common in Shadertoy conversions) read an undefined value, so it rendered differently every run and made the whole sequence irreproducible. The fragment output is now initialised before the shader’s own code runs
dkulp — Shader: the ISF DATE uniform came from the wall clock, so any sequence using a shader that reads it rendered differently every run. It now follows the sequence timeline, which is both reproducible and animates with the sequence rather than freezing at the time the render started
dkulp — Shader (Vulkan): a shader defining a function whose signature matches a GLSL built-in failed translation and rendered solid yellow; the built-in name is now renamed on a retry so the shader renders
dkulp — Shader: a shader Metal refused to build aborted the render instead of just not rendering that effect
derwin12 — Windows: fix a crash from an ISF shader whose slider declares a range too narrow to survive conversion (min ends up >= max) (#6755)
derwin12 — Pictures: a missing image file is now remembered per effect instead of being re-read every frame, and a model group holding a freed member no longer takes the render down with it (#6811)
dkulp — Video decode: hardware-decoded frames now take their colour space from the stream rather than being guessed from the frame size, and the scaler’s output buffer is cleared before use so any pixel the scaler skips cannot show as stray colour
dkulp — Video decode on Windows/Linux now asks the scaler for full-range RGB with the right matrix, snaps near-black pixels to true black, and uses an averaging filter for large reductions. Colours and blacks now match macOS closely, which matters for video composited with Transparent Black, and large downscales are faster
dkulp — Windows video decode: the DirectX11 reader picked the next frame instead of the nearest one for a requested time, so video ran up to a frame ahead of where macOS put it. Two separate causes, both fixed
dkulp — MacOS Video: a file whose display matrix encodes its pixel aspect as a scale made the decoder target a frame far larger than the source, so every cached frame was tens of MB. Renders using such a video could reach 40GB+ and be many times slower. Such files decode at native size again, as they did before decode-time scaling.
dkulp — MacOS Video: only H.264/HEVC keyframes were treated as safe decode entry points, so an all-intra file (MJPEG, ProRes) got exactly one, and every seek re-decoded the whole file from its first frame. Chains were held long enough that other effects using the same file timed out waiting, and a timeout was reported as end-of-video, which restarted looping effects early – so the same sequence rendered differently every time. Such files now render identically run to run, and a video-heavy sequence measured 94s -> 34s
dkulp — Embedding media now strips the show/media folder prefix from its stored path (and repoints the effects) instead of leaving an absolute path behind. Media folders are honoured, not just the show folder, and an image series embedded at load time is no longer skipped.
dkulp — GPU rendering: a canvas-mode layer with no active layers below it rendered on top of uninitialized GPU memory, so its output could differ run to run (mostly visible on Windows; latent on macOS). The blend scratch buffer is now zeroed, matching the CPU render path
dkulp — Headless render: a sequence whose predecessor’s render could not be drained is now reported as a failure rather than being loaded on top of the render workers still reading it
dkulp — FPP: LED panel matrix upload picked matrices by their position in the config rather than by their configured number, so it could write a start channel to the wrong matrix when they were not numbered sequentially
dkulp — FPP: re-uploading an unchanged LED panel matrix left the panel’s channels out of the controller’s channel ranges
dkulp — FPP: under full xLights control, a panel matrix xLights no longer drives stayed enabled on its old start channel
dkulp — Fixed a crash uploading outputs to a PixLite/PixCon whose vendor/model/variant has no capabilities definition – the port-count checks dereferenced the missing definition instead of reporting the upload error
dkulp — LOR: out-of-bounds write when changing a controller’s unit ID or channel count in the controller properties
dkulp — Fix crash editing or deleting an LOR device in controller properties when the property grid still held a stale device position
dkulp — Crash expanding a controller with no populated ports in the All Controller Columns dialog
dkulp — Crash pressing Ctrl+letter in the visualiser for a model that is not assigned to a controller
cybercop23 — Tools – Test: checking every controller spanned by the selection is now checked and connected
dkulp — Fix crash opening Tools – Test when a model uses a PWM protocol on a controller whose vendor/model has no capabilities definition
dkulp — Test: an RGB twinkle over a single selected channel spun until it ran out of memory
dkulp — Test: marking every 50th pixel read past the end of the selection when the selection was not a whole multiple of 50 pixels
dkulp — Fix crash when the value curve controls were re-enabled after a shader effect rebuilt its parameter list
scott — Linux: enable mDNS/Bonjour controller discovery (e.g. WLED) by linking against the avahi-compat-libdns_sd shim; previously the Linux build always compiled the discovery code out, silently, with no error
scott — Linux: hardware video decoding (VA-API, Preferences > Other) actually works now. The Linux build force-disabled it in code regardless of the setting, so the checkbox was hidden and had no effect even if re-enabled via config file
scott — Linux: the log file (xLights_spdlog.log) now lives in $XDG_STATE_HOME (default ~/.local/state/xLights) instead of /tmp, where it was wiped on every reboot and could vanish mid-session to a systemd-tmpfiles sweep
scott — Linux: fixed the build and startup on distributions whose C++ ABI differs from the prebuilt Map-from-Lights library, and a null model-metadata crash on launch
dkulp — Linux: Shader effects rendered as solid cyan on machines whose only Vulkan device is a software one (lavapipe/llvmpipe) and that have no usable OpenGL
- typically a headless render box. The device is now used for shaders, which have no CPU implementation, while compute effects stay on the faster CPU path; a headless render also enables GPU rendering if that is the only way a shader can be drawn
dkulp — macOS: fixed a crash when the Finder hands over more than one sequence at once – the save-changes prompt of the first open pumped the event queue and started the second open inside it, so the sequencer loaded against panes the first one was still closing; queued opens now run strictly one at a time
dkulp — macOS: the OS diagnostic payloads carried in crash reports were named with a localised clock, so in a 12-hour region the filename held bytes that made the payload impossible to extract from the report zip
dkulp — Windows: command-line output (–headless timings, –fseqcmp results, render diagnostics) now reaches the launching console instead of being discarded
dkulp — Fix a crash on app close resetting the effect panels to defaults, and track effect-panel window lifetime so a destroyed panel can never be reused

Changes

dkulp — Faces: the automatic eye blink is now computed independently per frame, so the render engine can render Faces frames in parallel. Blink timing may differ slightly from renders made with older releases.
dkulp — Layout: on the Controllers tab the preview now selects controller boxes rather than models. Models remain selectable from the controller tree and the Models tab
dkulp — Test: the pattern engine moved to shared core code so it is no longer carried in function-local statics that persisted across dialog open/close

Bug Fixes

Changes

Home

About xLights

How to donate

Current Release

2026.15 — August 4, 2026

Enhancements

xLights User Manual