Houdini POP particles — verified recipes

Default approach: use Houdini's native systems (POPs, stock SOPs/DOPs) over custom solutions; faking is fine when abstract/art-directed.

POP particle sims — verified recipe

In this build there are no SOP-level POP nodes; build a dopnet with POP DOPs inside. Collide with Static Object DOPs, not popcollisiondetect:

popobject ─────────────────────────────► popsolver::2.0 [0: Object]
popsource::2.0 → popwind → popcolor ───► popsolver     [1: Pre-Solve]
staticobject(s) ─┐
popsolver ───────┴► merge (affectortype 'ordered', colliders first) → output

• Viewing: wiring the dopnet straight into a SOP merge shows EVERY DOP object, including static-object collider proxies (they hide the particles). Use dopimport with objpattern='popobject1', importstyle='fetch' for the display branch.
• Read particles with dopnet.simulation().findObject('popobject1') — once static objects are in the sim, objects()[0] is a collider, not the particles.

Collisions — what actually contains and bounces (verified)

• Working combo: popsolver.collisionresponse='none' (= default bounce, uses bounce on popobject/staticobject) + staticobject collisiondetection='surface', mode='rayintersect' on clean, closed proxy geometry. 0% leaks, real bounce.
• For a container (cylinder bore): don't collide against the production mesh. Build a proxy — tube (poly, capped) at the bore radius from head face to below BDC. Verified in ray mode with a plain capped tube: 0% leaks to frame 900 at 20k particles with clumping on. Moving parts (piston) as a separate staticobject with animategeo=1; unpack Alembic + convert to polygons first.
• Failed approaches (don't repeat):
  • Volume/SDF collision (collisiondetection='volume') on a non-watertight mesh: bad sign, leaks.
  • Inverted SDF container (invert=1) on a tube: the volume ends at the tube's bbox = the wall, no solid voxels beyond → ~50% leak. A hollow solid (outer tube + reversed inner) also leaked in VOLUME mode (it contains fine in surface/ray mode).
  • collisionresponse='none' with volume collision: leaks. 'slide' contains but particles glue to walls/crown (no bounce) and accumulate there.
  • popcollisiondetect defaults to response='none' = detect only; ray against a production mesh with a hairline piston/bore slit leaks through the slit.
  • popinteract with a POSITIVE position force (repulsion) pushes particles through colliders. A NEGATIVE force (attraction) is safe and gives clumping — see below.
  • popaxisforce type 'sphere' pulls toward its centre POINT, not toward the axis.
• Collision on thin open surfaces (valves) works with ray mode.

Emission & motion (verified)

1. Sims must be stepped in order (for f in range(start, N): hou.setFrame(f)).
2. emittype='allpoint' births ONE particle per point per step, ignoring rate. Use 'point' with constantactivate=1, constantrate=<per second>.
3. Set popsource.dopobjects='*', usecontextgeo=0, soppath=<emitter SOP>.
4. Velocity isn't inherited unless v is in inheritattrib. Simplest: initvel='set', velx/y/z + varx/y/z for cone spread.
5. Spray deceleration — verified recipe (physics-based, not "viscous"): no popforce; popwind light (airresist ~0.5, amp ~0.12, swirlsize ~0.25) → popspeedlimit (dospeedmax=1, uselocal=1) following Hiroyasu–Arai: speed ~constant until breakup time, then penetration ∝ √t → speed ∝ 1/√t, floor at ambient-gas drift (~1/10 of spray speed): speedmax = @age < tb ? 3.0 : max(v0 * sqrt(tb / @age), 0.08); (tb 0.12, v0 1.3). Verified: mean speed 0.84 → 0.59 → 0.34 → 0.17 across age buckets; droplets reach walls/crown.
   • Avoid an abrupt cap (@age<0.25 ? 3 : 0.04) — it reads as a viscous stop.
   • popwind airresist saturates (3.5 → effective decay ~1.4/s) and popdrag doesn't stack with popwind (same drag attributes, last wins) — don't rely on drag for spray deceleration.
   • Gas follows the piston ("stretch" wind) — works, with two conditions: popwind uselocalwind=1, wind += set(0, vpist * s, 0) with s = height fraction (1 at crown, 0 at roof), vpist from spare parms ptop/ptop_prev (hscript bbox() on the piston and on a timeshift $F-1 copy). (1) Convert to sim time: vpist = Δy*60/timescale, with a spare parm tscale = ch("../popsolver1/timescale"). (2) The speed cap floor must exceed the gas speed (0.25, not 0.08). An earlier attempt failed (93% on the crown) because it combined the stretch with an abrupt 0.04 cap. Verified: cloud trails the crown by ~0.09, wall contact 33% → 7%.
   • Entrainment by age: in the same VEXpression, airresist = fit(@age, 0.3, 1.5, 0.1, 1.5); — fresh spray flies free, mature droplets follow the gas.
   • Wall/piston rebound: bounce ~0.5, friction ~0.1 on popobject AND each staticobject (Bai–Gosman: hot walls → rebound regime with energy loss).
   • Diagnose drift with an age-bucket table (mean speed, mean_vy, depth per age range).
   • Count impacts with hittotal (cumulative). hitnum is per-step only. 5b. Speed ramping (slow-mo): keyframe popsolver.timescale (e.g. 1.0 → 0.35 with ease() keys). Verified: age advances 0.35/60 per frame after the ramp. The dopnet's own timescale does not animate — keys are ignored and it evaluates constant. 5c. Clumping ("fuel-rich zones"): popinteract with positionforce -0.25, coreradius 0.015, falloffradius 0.05 (= clump scale), velforce 0. Verified: clump index 12.8 → 29.9, 0% leaks. Measure clumping as the index of dispersion (var/mean of counts in 0.05 cells within the cloud's 5–95% height range; 1 = random) against a bypassed baseline. Costs ~2× per frame. The index grows with density, so compare runs at the same particle count only; across counts compare median neighbours within a radius and % empty cells. Prefer this over popflock, which aligns neighbours into swarm-like groups.
6. Stale DOP cache: press dopnet.parm('resimulate').pressButton() before verifying.
7. Colour over the shot: popcolor colortype='ramp', updatecolor=1; its localramp VEX can use @Frame, noise(@P), rand(@id). 7b. Hollow-cone spray (piezo injector look): emitter = ring of ~128 points at the nozzle tip, each with v along the cone surface (half-angle ±wall thickness), speed factor rand 0.4–1.4 so droplets stop at different distances; popsource inheritattrib='v Cd ...', initvel='add', small var. Verified: fresh droplets 30–45° off-axis, median 36°, none <25°.
8. Emitter that follows a part: detail wrangle on the part's geo, centroid of the lowest points, removepoint + removeprim everything, addpoint at the tip; clamp below the chamber roof with a spare parm if the tip is recessed in the head.
9. Timed bursts: keyframe popsource.constantrate 0 → rate → 0 (bezier()). Set dopnet.startframe to the shot's first frame.
10. When the scene is rescaled, rescale emitter position, velocities, variance, forces and pscale.
11. Measure verification with distribution stats (height/radius bands, % outside the container), not just counts — a narrow spray can look "contained" because it never reached the walls.

Keeping this skill current

After any session that used Houdini/Cinema 4D via MCP, or whenever something new is verified:

1. Add gotchas/recipes ONLY if tested and working in a real scene; mark anything else "(unverified)".
2. Correct entries that proved wrong — don't append contradictions.
3. Put each finding in the skill it belongs to (see the indexes in houdini-mcp and cinema4d-mcp); project-specific facts go to that project's notes (see project-notes), never into a skill.
4. Keep each SKILL.md under ~300 lines; split into a new sub-skill or references/ when it grows.
5. Re-package every changed skill (keep names) and, if the skills live in a git repo, commit and push. Tell the user in one line what changed.