diff --git a/skills/creative/touchdesigner-mcp/SKILL.md b/skills/creative/touchdesigner-mcp/SKILL.md index ee25f7e09c..a81e29e6dc 100644 --- a/skills/creative/touchdesigner-mcp/SKILL.md +++ b/skills/creative/touchdesigner-mcp/SKILL.md @@ -204,8 +204,9 @@ win.par.winopen.pulse() | `td_input_clear` | Stop input automation | | `td_op_screen_rect` | Get screen coords of a node | | `td_click_screen_point` | Click a point in a screenshot | +| `td_screen_point_to_global` | Convert screenshot pixel to absolute screen coords | -See `references/mcp-tools.md` for full parameter schemas. +The table above covers the 32 tools used in typical creative workflows. The remaining 4 tools (`td_project_quit`, `td_test_session`, `td_dev_log`, `td_clear_dev_log`) are admin/dev-mode utilities — see `references/mcp-tools.md` for the full 36-tool reference with complete parameter schemas. ## Key Implementation Rules @@ -338,6 +339,10 @@ See `references/network-patterns.md` for complete build scripts + shader code. | `references/operator-tips.md` | Wireframe rendering, feedback TOP setup | | `references/geometry-comp.md` | Geometry COMP: instancing, POP vs SOP, morphing | | `references/audio-reactive.md` | Audio band extraction, beat detection, envelope following | +| `references/animation.md` | LFOs, timers, keyframes, easing, expression-driven motion | +| `references/midi-osc.md` | MIDI/OSC controllers, TouchOSC, multi-machine sync | +| `references/particles.md` | POPs and legacy particleSOP — emission, forces, collisions | +| `references/projection-mapping.md` | Multi-window output, corner pin, mesh warp, edge blending | | `scripts/setup.sh` | Automated setup script | --- diff --git a/skills/creative/touchdesigner-mcp/references/animation.md b/skills/creative/touchdesigner-mcp/references/animation.md new file mode 100644 index 0000000000..2ce55dd5e8 --- /dev/null +++ b/skills/creative/touchdesigner-mcp/references/animation.md @@ -0,0 +1,221 @@ +# Animation Reference + +Patterns for time-based motion — keyframes, LFOs, timers, easing, expression-driven animation. + +Always call `td_get_par_info` for the op type before setting params. Param names below reflect TD 2025.32 but verify if errors fire. + +--- + +## Time Sources + +TD has three time references — pick the right one. + +| Expression | Behavior | Use for | +|---|---|---| +| `absTime.seconds` | Wall-clock seconds since TD started. Never resets. | Continuous motion, GLSL `uTime`, infinite loops | +| `absTime.frame` | Wall-clock frame count. | Frame-accurate triggers | +| `me.time.frame` | Local component frame count (resets on play/stop). | Per-COMP animation timeline | +| `me.time.seconds` | Local component seconds. | Same, in seconds | + +**Rule:** for shaders and continuous motion use `absTime.seconds`. For triggered/looping animations inside a COMP use `me.time.*`. + +--- + +## LFO CHOP — Cyclic Motion + +The simplest periodic driver. Fast, GPU-cheap, expression-friendly. + +```python +lfo = root.create(lfoCHOP, 'rot_driver') +lfo.par.type = 'sin' # 'sin' | 'cos' | 'ramp' | 'square' | 'triangle' | 'pulse' +lfo.par.frequency = 0.25 # cycles per second +lfo.par.amplitude = 1.0 +lfo.par.offset = 0.0 +lfo.par.phase = 0.0 # 0-1, useful for offsetting parallel LFOs +``` + +**Drive a parameter via export:** + +```python +op('/project1/geo1').par.rx.mode = ParMode.EXPRESSION +op('/project1/geo1').par.rx.expr = "op('rot_driver')['chan1'] * 360" +``` + +**Multiple synced LFOs (X/Y/Z rotation with phase offsets):** +Create one LFO with three channels and phase-offset each, or use three LFOs and offset their `phase` params (0.0, 0.33, 0.66). + +--- + +## Timer CHOP — Triggered Sequences + +For run-once animations, beat-locked sequences, or stage-based logic. + +```python +timer = root.create(timerCHOP, 'fade_timer') +timer.par.length = 4.0 # cycle length in seconds +timer.par.cycle = False # run once vs. loop +timer.par.outputseconds = True +``` + +Output channels: `timer_fraction` (0→1 across the cycle), `running`, `done`, `cycles`. + +**Start the timer:** +```python +timer.par.start.pulse() +``` + +**Drive a fade:** +```python +op('/project1/level1').par.opacity.mode = ParMode.EXPRESSION +op('/project1/level1').par.opacity.expr = "op('fade_timer')['timer_fraction']" +``` + +**Easing on the timer fraction** — apply in the expression itself: + +```python +# Smoothstep: ease in/out +expr = "smoothstep(0, 1, op('fade_timer')['timer_fraction'])" +# Cubic ease-out: 1 - (1-t)^3 +expr = "1 - pow(1 - op('fade_timer')['timer_fraction'], 3)" +``` + +--- + +## Pattern CHOP — Custom Curves + +For arbitrary waveforms (saw ramps, easing curves, custom envelopes). + +```python +pat = root.create(patternCHOP, 'envelope') +pat.par.type = 'gaussian' # 'gaussian' | 'ramp' | 'square' | 'sin' | etc. +pat.par.length = 60 # samples +pat.par.cyclelength = 1.0 # seconds at TD framerate +``` + +Combine with `lookupCHOP` to remap a 0-1 driver through a custom curve. + +--- + +## Animation COMP — Keyframe-Based + +For multi-keyframe motion graphics. Each animationCOMP holds channels with keyframes editable in the Animation Editor. + +```python +anim = root.create(animationCOMP, 'intro_anim') +# By default has channels chan1..chanN; access via: +# op('intro_anim').par.length, .par.play, .par.cue, etc. + +# Drive a parameter from a channel +op('/project1/text1').par.tx.mode = ParMode.EXPRESSION +op('/project1/text1').par.tx.expr = "op('intro_anim/out1')['chan1']" +``` + +**Keyframes are typically edited in the UI** (Animation Editor), but can be set via `keyframes` table internally. For programmatic keyframe creation, use `td_execute_python`: + +```python +# Get the channel CHOP inside an animationCOMP +ch = op('/project1/intro_anim/chans') +# Insert a key (advanced API — verify with td_get_par_info(op_type='animationCOMP')) +ch.appendKey('chan1', frame=0, value=0.0, expression=None) +ch.appendKey('chan1', frame=120, value=1.0) +``` + +For most use cases, drive params with LFO/Timer/Pattern CHOPs instead — simpler and scriptable. + +--- + +## Easing in Expressions + +TD's expression evaluator supports Python math. Common easing forms: + +```python +# Linear +"t" + +# Smoothstep (classic ease-in-out) +"smoothstep(0, 1, t)" + +# Ease-out cubic +"1 - pow(1 - t, 3)" + +# Ease-in cubic +"pow(t, 3)" + +# Ease-in-out cubic +"3*t*t - 2*t*t*t" + +# Bounce (manual, simplified) +"abs(sin(t * 6.28 * 3) * (1 - t))" +``` + +Where `t` is `op('fade_timer')['timer_fraction']` or any 0-1 driver. + +--- + +## Filter CHOP — Smoothing Existing Channels + +Smooth out jittery values (e.g., audio analysis, sensor data) before driving visuals. + +```python +filt = root.create(filterCHOP, 'smooth') +filt.par.filter = 'gaussian' # or 'lowpass' +filt.par.width = 0.5 # smoothing window in seconds +filt.inputConnectors[0].connect(op('raw_signal')) +``` + +**WARNING:** Do NOT use Filter CHOP on AudioSpectrum output in timeslice mode — it expands the sample count and averages bins to near-zero. See `audio-reactive.md`. + +--- + +## Lag CHOP — Asymmetric Attack/Release + +Different speeds for rising vs. falling values. Standard for visualizing audio envelopes. + +```python +lag = root.create(lagCHOP, 'env_smooth') +lag.par.lag1 = 0.02 # attack (rise time, seconds) +lag.par.lag2 = 0.30 # release (fall time, seconds) +lag.inputConnectors[0].connect(op('raw_envelope')) +``` + +Fast attack, slow release = classic VU-meter feel. + +--- + +## Per-Frame Driving via Script DAT + +For complex per-frame logic that doesn't fit expressions, use a `executeDAT` (`onFrameStart` callback) or a `chopExecuteDAT`. + +```python +# In an executeDAT (frameStart): +def onFrameStart(frame): + t = absTime.seconds + op('/project1/circle').par.tx = math.sin(t * 2.0) * 3.0 + op('/project1/circle').par.ty = math.cos(t * 2.0) * 3.0 + return +``` + +Heavy logic should still be in CHOPs (CPU-cheap, deterministic). Reserve scripts for one-shots or non-realtime branching. + +--- + +## Pitfalls + +1. **Frame rate dependency** — `me.time.frame` is in TD project frames (default 60). If your project rate changes, motion speed changes. Use `seconds` for rate-independent timing. +2. **Cooking budget** — every CHOP that drives a parameter cooks every frame. Consolidate drivers (one big mathCHOP > many small ones). +3. **Expression mode** — params default to `CONSTANT`. `par.X.expr = ...` is ignored unless `par.X.mode = ParMode.EXPRESSION`. +4. **Animation editor edits** — keyframes set via UI live in the animationCOMP's internal keyframe table. They survive save/reopen. Programmatic keys via `appendKey()` work but verify the API with `td_get_docs(topic='animation')` first. +5. **Looping animations** — for seamless loops, `length` must equal `cyclelength` and the start/end values must match. Otherwise expect a visible jump. + +--- + +## Quick Recipes + +| Goal | Simplest path | +|---|---| +| Continuous rotation | LFO CHOP `type='ramp'`, expr → `geo.par.rx` | +| Fade in over 2s | Timer CHOP `length=2`, smoothstep expr → `level.par.opacity` | +| Pulse on every beat | `triggerCHOP` from audio → drive scale via expression | +| 3D Lissajous orbit | Two LFOs with different freq, drive `tx`/`ty`/`tz` | +| Random jitter | `noiseCHOP` (low-freq) added to position | +| Timed scene switch | Timer CHOP → switchTOP/CHOP `index` | diff --git a/skills/creative/touchdesigner-mcp/references/midi-osc.md b/skills/creative/touchdesigner-mcp/references/midi-osc.md new file mode 100644 index 0000000000..23cbbd850a --- /dev/null +++ b/skills/creative/touchdesigner-mcp/references/midi-osc.md @@ -0,0 +1,211 @@ +# MIDI / OSC Reference + +External controller input and output — MIDI hardware, TouchOSC mobile UIs, OSC routing across the network. + +For audio-driven MIDI patterns (track triggers from spectrum analysis), see also `audio-reactive.md`. + +--- + +## MIDI Input — Hardware Controllers + +### Discovery + +List connected MIDI devices first. Use a `midiinDAT` to enumerate: + +```python +mdat = root.create(midiinDAT, 'mid_devices') +# Read available device names from the DAT after one cook +``` + +Or via Python directly: + +```python +# In td_execute_python +import td +devices = [d for d in op.MIDI.devices] # verify with td_get_docs('midi') +``` + +Verify the API with `td_get_docs(topic='midi')` since this varies between TD versions. + +### MIDI In CHOP + +Standard pattern: + +```python +midi_in = root.create(midiinCHOP, 'midi_in') +midi_in.par.device = 0 # device index from discovery +midi_in.par.activechan = True +``` + +Output channels follow the convention `chCcN` and `chCnN`: +- `ch1c74` — channel 1, CC 74 +- `ch1n60` — channel 1, note 60 (middle C) — value is velocity 0-127 + +**Map a CC to a parameter:** + +```python +op('/project1/bloom1').par.threshold.mode = ParMode.EXPRESSION +op('/project1/bloom1').par.threshold.expr = "op('midi_in')['ch1c74'][0] / 127.0" +``` + +**Map a note as a trigger:** + +Notes in `midiinCHOP` output velocity while held, 0 when released. Use a `triggerCHOP` to convert a held note into pulses: + +```python +trig = root.create(triggerCHOP, 'note_trig') +trig.par.threshold = 1 +trig.par.triggeron = 'increase' +trig.inputConnectors[0].connect(op('midi_in')) +# Filter to a single channel via a selectCHOP if desired +``` + +### MIDI Learn Pattern + +Build a reusable learn pattern when you don't know the controller's CC layout in advance: + +1. Drop a `midiinCHOP` and `selectCHOP` after it. +2. User wiggles the controller knob. +3. Use `td_read_chop` on the midiinCHOP to identify which channel is non-zero — that's the active CC. +4. Set the `selectCHOP.par.channames` to that channel name. +5. Save the mapping to a `tableDAT` so it persists across sessions. + +--- + +## MIDI Output + +```python +midi_out = root.create(midioutCHOP, 'midi_out') +midi_out.par.device = 0 +midi_out.par.outputformat = 'continuous' # 'continuous' | 'event' + +# Drive an output: send out a CC mapped from any 0-1 source +src = root.create(constantCHOP, 'cc_src') +src.par.name0 = 'ch1c20' +src.par.value0 = 0.5 +midi_out.inputConnectors[0].connect(src) +``` + +For note events specifically, use `event` mode and pulse the value with a `pulseCHOP` or `triggerCHOP`. + +--- + +## OSC Input — Network Control + +OSC is the more flexible cousin of MIDI. Used heavily for: +- TouchOSC / Lemur mobile control surfaces +- Show control systems (QLab, Watchout) +- Inter-application sync (Ableton via Max for Live, Resolume, etc.) + +### OSC In CHOP + +```python +osc_in = root.create(oscinCHOP, 'osc_in') +osc_in.par.port = 7000 # listen on UDP 7000 +osc_in.par.localaddress = '' # empty = all interfaces +osc_in.par.queued = False # immediate vs. queued processing +``` + +Each incoming OSC address becomes a channel. `/scene/1/intensity` becomes a channel named `scene_1_intensity` (TD sanitizes slashes to underscores). + +**Common gotcha:** TD only creates the channel after the FIRST message arrives at that address. Send a "hello" message from the controller during setup, or pre-declare channel names manually. + +### OSC In DAT (for raw events) + +Use a `oscinDAT` when you need full message access (multiple typed args, addresses with brackets/regex). + +```python +osc_dat = root.create(oscinDAT, 'osc_events') +osc_dat.par.port = 7001 +# Each row: timestamp, address, type tags, args... +``` + +Drive logic via a `datExecuteDAT` watching the `oscinDAT`: + +```python +def onTableChange(dat): + last = dat[dat.numRows - 1, 'message'] + parsed = last.val.split() + addr = parsed[0] + args = parsed[1:] + if addr == '/scene/trigger': + op('/project1/scene_switcher').par.index = int(args[0]) + return +``` + +--- + +## OSC Output — Sending to External Apps + +```python +osc_out = root.create(oscoutCHOP, 'osc_out') +osc_out.par.netaddress = '127.0.0.1' # destination IP +osc_out.par.port = 9000 + +# Channel names become OSC addresses +src = root.create(constantCHOP, 'send') +src.par.name0 = 'scene/intensity' # → /scene/intensity +src.par.value0 = 0.7 +osc_out.inputConnectors[0].connect(src) +``` + +**Channel-to-address mapping:** TD prepends `/` automatically. Use `/` in channel names to nest. + +For one-shot string/typed messages, use `oscoutDAT` and call `.sendOSC(address, args)`: + +```python +op('osc_out_dat').sendOSC('/scene/trigger', [1, 'fade']) +``` + +--- + +## TouchOSC / Mobile UI Pattern + +Common setup for live VJ control from a phone/tablet: + +1. **Configure TouchOSC layout** — assign each control an OSC address like `/vj/master`, `/vj/scene/1`, etc. +2. **Find your machine's LAN IP** — TouchOSC needs to point at it. +3. **TD listens** on `oscinCHOP.par.port = 8000` (or whichever). +4. **Map channels to params** via expressions: + +```python +op('/project1/master_level').par.opacity.mode = ParMode.EXPRESSION +op('/project1/master_level').par.opacity.expr = "op('osc_in')['vj_master']" +``` + +5. **Send feedback** to the controller via `oscoutCHOP` — useful for syncing state across multiple devices. + +--- + +## Network / Multi-Machine + +OSC over LAN works out-of-the-box. For multi-TD-instance sync (e.g., projection cluster): + +- One TD acts as **master**, broadcasts `/sync/...` over OSC +- Worker TDs run `oscinCHOP` listening on the same port +- Use UDP **broadcast address** (e.g., `192.168.1.255`) on the master's `oscoutCHOP.par.netaddress` to hit all peers + +For reliability over WAN, use `webserverDAT` or `websocketDAT` with an external relay instead — UDP loss is invisible. + +--- + +## Pitfalls + +1. **MIDI device indexing** — device `0` is whichever device TD enumerated first. Reorder may shift it. Pin by name when possible. +2. **OSC channel names** — TD doesn't create a channel until the first message lands. New channels invalidate cooked dependents on first arrival, causing a one-frame stutter. +3. **OSC queued mode** — `par.queued = True` defers processing to a single per-frame batch. Lower latency but messages arriving same frame collapse to the last value. Off for triggers, on for continuous knobs. +4. **MIDI clock vs. transport** — `midiinCHOP` reports clock if available. Use `midisyncCHOP` (if your TD version exposes it) or compute BPM from clock pulses (24 per quarter note). +5. **Latency** — wired MIDI is ~1-3ms. WiFi OSC is 10-30ms with jitter. Use wired for tight beat-locked work. +6. **Port conflicts** — only one process can bind a UDP port on most OS. If `oscinCHOP` shows no traffic, check that another app (Max, Ableton, etc.) isn't already listening on that port. + +--- + +## Quick Recipes + +| Goal | Op chain | +|---|---| +| Knob → bloom intensity | `midiinCHOP` → expression on `bloom.par.threshold` | +| Note → scene change | `midiinCHOP` → `triggerCHOP` → `selectCHOP` → drive `switchTOP.par.index` | +| Phone slider → master fader | TouchOSC `/master` → `oscinCHOP` → expression on output `level.par.opacity` | +| TD → Resolume scene trigger | `oscoutCHOP` channel `composition/layers/1/clips/1/connect` → Resolume listening on 7000 | +| Multi-projector sync | Master TD `oscoutCHOP` broadcast → workers `oscinCHOP` | diff --git a/skills/creative/touchdesigner-mcp/references/particles.md b/skills/creative/touchdesigner-mcp/references/particles.md new file mode 100644 index 0000000000..048e495545 --- /dev/null +++ b/skills/creative/touchdesigner-mcp/references/particles.md @@ -0,0 +1,245 @@ +# Particles Reference + +Particle systems in TouchDesigner — modern POPs (Particle Operators) and the legacy particleSOP path. + +For instancing static geometry (without per-instance lifetime/velocity), see `geometry-comp.md`. For GLSL-driven feedback simulations (no particle abstraction), see `operator-tips.md` (Feedback TOP section). + +Always call `td_get_par_info` for the op type before setting params. Param names below reflect TD 2025.32 — verify before relying on them. + +--- + +## Two Paths: POPs vs. SOPs + +| | **POP family** (modern) | **particleSOP** (legacy) | +|---|---|---| +| GPU? | Yes (compute) | No (CPU) | +| Particle count | 100k+ comfortably | ~5k before slowdown | +| API style | Source / Force / Solver / Render chain | Single op with many params | +| Use for | New projects, anything intensive | Quick demos, low counts, TD < 2023 | + +**Default to POPs.** Only fall back to particleSOP if a POP variant of an op you need doesn't exist. + +--- + +## POP Pipeline Overview + +A POP system is a chain of operators inside a `geometryCOMP`: + +``` +popSourceTOP / popSourceSOP ← spawn new particles + ↓ +popForceTOP (gravity, wind, etc.) + ↓ +popForceTOP (attractor, vortex, ...) + ↓ +popDeleteTOP (lifetime, bounds) + ↓ +popSolverTOP ← integrates velocity, updates positions + ↓ +[render via geometryCOMP / glslMAT instancing] +``` + +POP buffers carry standard channels: `P` (position), `v` (velocity), `life`, `id`, `Cd` (color), plus any custom channels you add. + +--- + +## Minimal POP Setup + +```python +# Create a geometry COMP to hold the POP network +geo = root.create(geometryCOMP, 'particles_geo') + +# 1. Source — emit particles from a point +src = geo.create(popSourceTOP, 'src') +src.par.birthrate = 500 # per second +src.par.life = 4.0 # seconds + +# 2. Gravity force +grav = geo.create(popForceTOP, 'gravity') +grav.par.forcetype = 'gravity' +grav.par.fy = -9.8 + +# 3. Lifetime cleanup +delp = geo.create(popDeleteTOP, 'cull') +delp.par.condition = 'lifeleq' # delete when life <= 0 +delp.par.value = 0 + +# 4. Solver +solv = geo.create(popSolverTOP, 'solver') +solv.par.timestep = 'frame' + +# Wire: source → force → delete → solver +src.outputConnectors[0].connect(grav.inputConnectors[0]) +grav.outputConnectors[0].connect(delp.inputConnectors[0]) +delp.outputConnectors[0].connect(solv.inputConnectors[0]) +``` + +The `popSolverTOP` output IS the live particle buffer. Render it via `glslMAT` instancing on a small SOP (sphere, point) as the "shape" of each particle. + +--- + +## Common Forces + +| Force type | Effect | Common params | +|---|---|---| +| `gravity` | Constant directional pull | `fx`, `fy`, `fz` | +| `wind` | Constant velocity addition | `wx`, `wy`, `wz` | +| `drag` | Velocity damping over time | `dragstrength` | +| `noise` | Curl-noise turbulence | `noiseamp`, `noisefreq`, `noiseseed` | +| `attractor` | Pull toward a point | `position`, `strength`, `falloff` | +| `vortex` | Swirl around an axis | `axis`, `strength` | +| `point` (custom) | GLSL-evaluated arbitrary force | via `popforceadvancedTOP` | + +Stack multiple `popForceTOP`s in series — each modifies velocity additively. + +--- + +## Lifecycle Patterns + +### Continuous emission (e.g. smoke plume) + +```python +src.par.birthrate = 800 +src.par.life = 6.0 # variance via 'lifevariance' +src.par.lifevariance = 1.5 +``` + +### Burst emission (e.g. explosion) + +```python +src.par.birthrate = 0 # no continuous emission +src.par.burst.pulse() # one burst on demand (verify param name) +src.par.burstcount = 5000 +src.par.life = 1.5 +``` + +### Beat-triggered burst + +Wire a `triggerCHOP` (from audio or MIDI) to pulse the burst: + +```python +op('/project1/audio_kick_trigger').outputConnectors[0].connect(...) +# Then via a chopExecuteDAT, on each kick: +def offToOn(channel, sampleIndex, val, prev): + op('/project1/particles_geo/src').par.burst.pulse() + return +``` + +--- + +## Rendering Particles + +### Point Sprites (simplest) + +```python +# Inside the geometryCOMP, render the solver output directly +# The geo's first SOP child becomes the geometry +# But for POPs, we typically render via glslMAT on a small "shape" + +# Simple billboard sphere per particle: +shape = geo.create(sphereSOP, 'shape') +shape.par.rad = 0.05 +shape.par.rows = 6; shape.par.cols = 6 # low-poly to keep it fast + +# Material that uses POP buffer for instancing +mat = root.create(glslMAT, 'particle_mat') +# Configure mat.par.instancingTOP = solver output (verify param name) +``` + +The exact instancing setup varies by TD version — call `td_get_hints(topic='popInstancing')` (or `popRender` / `instancing` — try a few). + +### GPU Sprites via glslcopyPOP + +For dense smoke/fire-like effects, use a `glslcopyPOP` that writes per-particle color/size from a compute shader, then render as point sprites with additive blending in a `renderTOP`. + +--- + +## Collisions + +```python +# Collision detection against an SOP +coll = geo.create(popCollideTOP, 'ground_coll') +coll.par.collidewithsop = '/project1/ground_geo' # path to colliding SOP +coll.par.bounce = 0.3 +coll.par.friction = 0.1 +# Insert between force and solver +``` + +For plane/box collisions only, use `popPlaneCollideTOP` (cheaper). + +--- + +## Custom Per-Particle Data + +Add a custom channel via `popAttribCreateTOP` (or by writing through `glslcopyPOP`): + +```python +# Add a "phase" attribute initialized random per-particle, used in render shader +attr = geo.create(popAttribCreateTOP, 'add_phase') +attr.par.attribname = 'phase' +attr.par.value0 = 'rand(@id)' # expression in TD's POP attribute language +``` + +Then in the render shader, `texture(sTDPOPInputs[0].phase, ...)` (or whichever sampler convention your TD version uses — verify with `td_get_docs(topic='pops')`). + +--- + +## Legacy particleSOP (Use Sparingly) + +For quick demos or low-count systems: + +```python +# Inside a geo +psrc = geo.create(addSOP, 'point_src') # source: a single point +psrc.par.points = '0 0 0' + +part = geo.create(particleSOP, 'particles') +part.par.life = 3.0 +part.par.birthrate = 100 +part.par.gravityy = -9.8 +part.par.windx = 0.5 +part.inputConnectors[0].connect(psrc) +``` + +CPU-bound. Beyond ~5,000 active particles you'll see frame drops. + +--- + +## Pitfalls + +1. **Particles don't appear** — usually a render-side issue. Check via `td_get_screenshot` on the solver output (renders the buffer as a TOP-like view in newer TD). Then check the `geometryCOMP`'s render path. +2. **Burst won't fire** — verify the `burst` param is a pulse, not a toggle. Pulses must use `.pulse()`, not `= True`. +3. **Particles teleport on first frame** — uninitialized velocity. Set `popSourceTOP.par.initialvelocityX/Y/Z` or zero them explicitly. +4. **Gravity feels wrong** — TD's "1 unit" depends on your scene scale. Start with `fy = -1.0` and scale up rather than using real-world 9.8. +5. **High birthrate = stuttering** — birthrate is per-second, not per-frame. At 60fps, `birthrate = 6000` is 100/frame which is fine; `birthrate = 600000` will tank. +6. **POP solver order matters** — forces apply in the order they appear in the chain. Putting gravity AFTER drag dampens gravity itself; usually not what you want. +7. **Instancing param name varies** — `mat.par.instancingTOP` vs. `mat.par.instanceop` vs. `mat.par.instances` differs across TD versions. Always check `td_get_par_info(op_type='glslMAT')`. +8. **Cooking dependency loops** — POP solvers create implicit time-loops. The "cook dependency loop" warning is expected and harmless for POPs. +9. **CHOP-driven force values** — when a force param is expression-bound to a CHOP (e.g., audio-reactive gravity), make sure the CHOP cooks before the solver. If not, force lags by one frame. + +--- + +## Performance Targets + +| Particle count | Setup | Frame budget @ 60fps | +|---|---|---| +| < 1k | particleSOP fine | trivial | +| 1k - 10k | POPs, simple forces | ~2-5ms | +| 10k - 100k | POPs, GPU-only forces | ~5-15ms | +| 100k+ | `glslcopyPOP`, custom compute | ~10-25ms | +| 1M+ | Custom GPU buffer, no POP framework | depends on shader | + +Use `td_get_perf` to find which op in the POP chain is the bottleneck. + +--- + +## Quick Recipes + +| Goal | Pipeline | +|---|---| +| Smoke plume | `popSourceTOP` (point) → gravity + wind + noise → `popDeleteTOP` (life) → solver → glslMAT instancing | +| Beat-triggered burst | `triggerCHOP` (audio) → chopExecuteDAT pulses `popSourceTOP.par.burst` | +| Fireworks shell | Burst at point → drag + gravity → secondary burst on lifetime threshold | +| Snow/rain | Continuous emission across XZ plane (high y), gravity + small wind, infinite life box-deleted | +| Sparks | Burst, very short life (0.3s), bright additive render, motion blur via feedback | +| Audio particles | Birthrate driven by audio envelope, color driven by frequency band | diff --git a/skills/creative/touchdesigner-mcp/references/projection-mapping.md b/skills/creative/touchdesigner-mcp/references/projection-mapping.md new file mode 100644 index 0000000000..9b2fb5863f --- /dev/null +++ b/skills/creative/touchdesigner-mcp/references/projection-mapping.md @@ -0,0 +1,211 @@ +# Projection Mapping Reference + +Multi-window output, surface mapping, edge blending, and projector calibration patterns for installation/event work. + +For HUD layouts and on-screen panel grids, see `layout-compositor.md`. For wireframe/test-pattern generation, see `operator-tips.md`. + +--- + +## Window COMP — Output to a Display + +The `windowCOMP` is how TD pushes pixels to a real display. + +```python +win = root.create(windowCOMP, 'output_window') +win.par.winop = '/project1/final_out' # path to the TOP being displayed +win.par.winw = 1920 +win.par.winh = 1080 +win.par.winoffsetx = 0 # screen-space offset +win.par.winoffsety = 0 +win.par.borders = False # no chrome +win.par.alwaysontop = True +win.par.cursor = False # hide cursor in fullscreen +win.par.justify = 'fillaspect' # 'fill' | 'fitaspect' | 'fillaspect' | 'native' +win.par.winopen.pulse() # OPEN the window +``` + +To target a specific physical display, set `par.location`: + +```python +win.par.location = 'secondary' # 'primary' | 'secondary' | 'monitor1' | 'monitor2' | ... +``` + +Or set absolute coordinates using `winoffsetx/y` matched to your OS display layout. + +**Always pulse `winopen` — setting params alone doesn't open the window.** + +--- + +## Multi-Window Output + +For multi-projector or multi-display setups, create one `windowCOMP` per output, each pointing at a different TOP. + +```python +for i, screen_top in enumerate(['out_left', 'out_center', 'out_right']): + w = root.create(windowCOMP, f'win_{i}') + w.par.winop = f'/project1/{screen_top}' + w.par.winw = 1920; w.par.winh = 1080 + w.par.winoffsetx = i * 1920 + w.par.winoffsety = 0 + w.par.borders = False + w.par.alwaysontop = True + w.par.cursor = False + w.par.winopen.pulse() +``` + +For ultra-wide single-output spans, use ONE windowCOMP at e.g. 5760×1080 spanning three projectors via the GPU's mosaic/spanning mode (Nvidia Mosaic, AMD Eyefinity), then split content via `cropTOP` per screen inside TD. + +--- + +## 4-Point Corner Pin (Quad Warp) + +The simplest projection mapping primitive — warping a rectangle onto a quadrilateral. + +```python +# Source content +src = op('/project1/scene_out') + +# Manual: cornerPinTOP (TD has this built-in) +cp = root.create(cornerPinTOP, 'corner_pin') +cp.par.tlx = 0.05; cp.par.tly = 0.10 # top-left (normalized 0-1) +cp.par.trx = 0.95; cp.par.try = 0.08 # top-right +cp.par.brx = 0.93; cp.par.bry = 0.92 # bottom-right +cp.par.blx = 0.07; cp.par.bly = 0.94 # bottom-left +cp.inputConnectors[0].connect(src) +``` + +Alternative: use a `geometryCOMP` with a `gridSOP` and bend the verts in vertex GLSL. More flexible (curved surfaces) but more setup. + +Verify TD 2025.32 param names with `td_get_par_info(op_type='cornerPinTOP')`. + +--- + +## Bezier / Mesh Warp (Curved Surfaces) + +For non-flat surfaces (domes, columns, curved walls), use a subdivided mesh and per-vertex displacement. + +### Pattern: Grid Mesh + GLSL Displacement + +```python +# Subdivided grid in a geo +geo = root.create(geometryCOMP, 'warp_geo') +grid = geo.create(gridSOP, 'warp_grid') +grid.par.rows = 32 # higher = smoother curve +grid.par.cols = 32 +grid.par.sizex = 2; grid.par.sizey = 2 + +# Texture the source onto it +mat = root.create(constMAT, 'warp_mat') # use constMAT for unlit projection +mat.par.maptop = '/project1/scene_out' # source TOP + +geo.par.material = mat.path + +# Render to a TOP that goes to the projector window +cam = root.create(cameraCOMP, 'cam_proj') +cam.par.tz = 4 + +render = root.create(renderTOP, 'projection_out') +render.par.camera = cam.path +render.par.geometry = geo.path +render.par.outputresolution = 'custom' +render.par.resolutionw = 1920; render.par.resolutionh = 1080 +``` + +For per-vertex offsets, write a vertex GLSL on the constMAT (or use `glslMAT`) and read displacement values from a CHOP via uniform. + +Calibration is iterative: render a checkerboard from `scene_out`, project it, photograph the projection, manually nudge corner/grid points until aligned. + +--- + +## Edge Blending (Multi-Projector Overlap) + +When two projectors overlap, the overlap region is twice as bright. Blend by ramping each projector's edge alpha to 0 across the overlap zone. + +### GLSL Edge Blend Shader + +Per-projector output pass that fades the inside edge to black: + +```glsl +// edge_blend_pixel.glsl +out vec4 fragColor; +uniform float uBlendLeft; // overlap width on left edge (0-0.5, 0=no blend) +uniform float uBlendRight; +uniform float uGamma; // typically 2.2 — perceptual ramp + +void main() { + vec2 uv = vUV.st; + vec4 col = texture(sTD2DInputs[0], uv); + + float aL = (uBlendLeft > 0.0) ? smoothstep(0.0, uBlendLeft, uv.x) : 1.0; + float aR = (uBlendRight > 0.0) ? smoothstep(0.0, uBlendRight, 1.0 - uv.x) : 1.0; + float a = pow(aL * aR, uGamma); + + fragColor = TDOutputSwizzle(vec4(col.rgb * a, 1.0)); +} +``` + +Apply this to each overlap-touching projector's output. Tune `uBlendLeft` / `uBlendRight` to match your physical overlap. + +For top/bottom blends or cylindrical setups, extend the shader with `uBlendTop` / `uBlendBottom`. + +--- + +## Calibration Patterns + +Useful test patterns for aligning projectors. Build a `switchTOP` selecting one of these, route to all projector windows during setup. + +```python +# Solid white — for brightness/uniformity check +white = root.create(constantTOP, 'cal_white') +white.par.colorr = 1.0; white.par.colorg = 1.0; white.par.colorb = 1.0 + +# Centered crosshair — for keystone alignment +gridcross = root.create(textTOP, 'cal_cross') +gridcross.par.text = '+' +gridcross.par.fontsizex = 200 + +# Fine grid — for warp/mesh alignment (use rampTOP + math + threshold, or build via GLSL) +# Color bars for projector color calibration +bars = root.create(rampTOP, 'cal_bars') +bars.par.type = 'horizontal' +``` + +Or use the bundled `testpatternTOP` if your TD version includes it. + +--- + +## Projection Audit Workflow + +When debugging a multi-screen setup: + +1. Render a unique color and label per output (`textTOP` saying "LEFT", "CENTER", "RIGHT"). +2. Check that each window is sourcing the correct path: `td_get_operator_info(path='/project1/win_0')`. +3. Verify display assignment: walk to each projector and confirm visually. +4. Check resolution: physical projector native res vs. TD output res — mismatches cause scaling artifacts. +5. Cook flag: `td_get_perf` — if a window's source TOP isn't cooking, the projector shows last frame frozen. + +--- + +## Pitfalls + +1. **Window won't open** — you forgot `winopen.pulse()`. Setting params alone doesn't open it. +2. **Wrong display** — `par.location='secondary'` depends on OS display order. Set `winoffsetx/y` to absolute coords as a more reliable override. +3. **Cursor visible** — set `par.cursor = False` BEFORE opening, or close+reopen. +4. **Black projection** — usually a cooking issue. Verify `final_out` TOP is cooking via `td_get_perf`. Check `td_get_errors` recursively from `/`. +5. **Tearing / vsync** — `windowCOMP` honors `par.vsync`. For projection always set `vsync='vsync'` (default). Tearing means GPU is over-budget — reduce render resolution. +6. **Aspect mismatch** — projector native is often 1920×1200 (16:10) not 1080. Use `justify='fitaspect'` or render at native projector res. +7. **Non-Commercial license** — caps total resolution at 1280×1280. For real installation work you need Commercial. Pro license adds 4K+. +8. **Multiple monitors on macOS** — `windowCOMP` honors macOS Spaces. Disable Spaces or pin TD to a specific display in System Settings before showtime. + +--- + +## Quick Recipes + +| Goal | Approach | +|---|---| +| Single fullscreen output | One `windowCOMP`, `justify='fillaspect'`, `winopen.pulse()` | +| 3-projector wide span | 3 `windowCOMP` + per-output `cropTOP` from one wide source | +| Single quad surface | `cornerPinTOP` → `windowCOMP` | +| Curved/dome | Subdivided gridSOP with vertex GLSL → `renderTOP` → `windowCOMP` | +| Edge blend overlap | GLSL fade shader per projector → `windowCOMP` | +| Calibration mode | `switchTOP` between scene and test patterns, hot-key triggered |