agentleFS
dima-rdnvdima-rdnv/claude-skills/houdini-flipbook/SKILL.md

houdini-flipbook

Skill2 starsChanged 6 days agoMITView on GitHub

Rendering Houdini viewport flipbooks and OpenGL ROP passes to image sequences through MCP / hython — real alpha from hython, a camera copied from Cinema 4D, volume visualization for smoke / fire passes. Use for flipbooks, playblasts, OpenGL renders or Houdini passes to composite. Load with houdini-mcp.

The file

We split this skill into the rules it gives an agent

5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Verify framing: render the same frame of an engine part in both and compare silhouettes (matched within a few pixels at 1080p).

    line 82
  2. Distinctive:

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

    line 109
  3. Distinctive:

    Add gotchas/recipes ONLY if tested and working in a real scene.

    line 110
  4. Very specific:

    Put each finding in the skill it belongs to (see the indexes in houdini-mcp and cinema4d-mcp).

    line 112
  5. Very specific:

    Keep each SKILL.md under ~300 lines.

    line 114

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Viewport flipbooks and OpenGL passes (verified Houdini 21 Indie, macOS)

Flipbook from the GUI (verified: 1920×1080 PNGs)

sv = hou.ui.paneTabOfType(hou.paneTabType.SceneViewer)
vp = sv.curViewport()
prev = vp.settings().colorScheme()
vp.settings().setColorScheme(hou.viewportColorScheme.Dark)   # dark background
s = sv.flipbookSettings().stash()
s.frameRange((start, end)); s.frameIncrement(1)
s.output(os.path.join(out_dir, 'name_$F4.png'))
s.outputToMPlay(False)
s.useResolution(True); s.resolution((1920, 1080))
sv.flipbook(vp, s)
# vp.settings().setColorScheme(prev)   # restore, unless the user prefers it dark
  • A dark background reads better for particles. If the user wants it kept dark, set it on all SceneViewer viewports: for vp in pt.viewports(): vp.settings().setColorScheme(Dark).
  • Flipbooks the CURRENT viewport view and the displayed node — set the display node first.
  • It cooks frames in order (sims build correctly) but blocks the bridge for minutes. Monitor from a shell: file count in the output folder, sips -g pixelWidth -g pixelHeight.
  • Output folders: follow the user's convention (see the project's notes), e.g. a new <YYMMDD HHMM> <index> folder per flipbook built from datetime.now().

OpenGL ROP passes for compositing (verified)

  • /out → opengl ROP. Parms: camera; vobjects '' + forceobjects (renders only these objects, their display flag off is fine); alights '' + forcelights; trange 1, f1/f2/f3 (delete their keyframes before set()); picture; mkpath 1; aamode 'aa8'; colorcorrect 'lut_gamma' (gamma 2.2) or 'none' (linear); volumequality; particle ('points', 'discs', 'litsphere', …).
  • For alpha, render with hython. The same ROP rendered from the GUI wrote alpha = 1 everywhere (opaque viewport background, PNG and EXR). Rendered by hython (rop.render() in a script, see houdini-mcp) it wrote a transparent background: PNG = straight (unpremultiplied) RGBA, alpha = volume coverage. It ran headless on macOS at ~2–4 fps for 1080p volume frames; two jobs in parallel.
  • GUI-only fallback for a matte: a second object with the same volume, albedo 0 and emission = extinction (emitfield density with the same scale and ramp mode as the density) → RGB equals coverage (its mean matched the hython alpha within ~2 %). Lit pass + matte = a premultiplied pair.
  • Emission-only passes (fire glow): an 8-bit straight-alpha PNG clamps them (glow turned into white fog). Render linear EXR (colorcorrect none) with almost no absorption and add it in the compositor (AE: blending Add, alpha Ignore).
  • A SOP Solver feeding two ROPs: render each ROP in its own hython process. Rendered one after the other in one session, the second ROP's first frames came out with the solver's state (and geometry) from a much later frame — the solver cache had dropped the early frames during the first ROP (long shots only: 400+ solver steps; short shots were fine). Check: the alpha of two passes of the same geometry must match (hoiiotool a.exr --ch A b.png --ch A --diff → mean error ~1e-6). Separate processes (or a file cache of the solver) fixed it; they can run in parallel.
  • 30 fps edit from a 60 fps scene: f3 = 2 and picture ending .`padzero(4, $F/2)`.png → file numbers are the 30 fps frames.
  • Geometry passes (particles as small instanced octahedra): object shop_materialpath → principledshader::2.0 with reflect 0, rough 1; delete point Cd so the material colour wins.
  • The OpenGL ROP ignores emission from point colour (emitcolor_usePointColor): a pass meant to show per-point colours as emission came out plain white.
  • Flat point-colour pass (verified): material basecolor (1,1,1) + basecolor_usePointColor 1, reflect 0, rough 1, emitint 0; light it with ONE ambient light object through forcelights, light_intensity 10. GL ambient contributes 0.1 × intensity (measured: 1 → 0.10, 9.66 → 0.962), so 10 gives rendered RGB = Cd exactly, with no shading falloff. ROP: shadows 0, hqlighting 0, colorcorrect 0 (raw values). Lighting clamps at 1.0, so keep Cd ≤ 1 (scale a glow up in the compositor, or write it as EXR and Add it).
  • Pass types built this way: a multiply pass (white = untouched, tint/char darker; straight-alpha PNG, Multiply in the compositor) and an emission pass (black = nothing; linear EXR, Add) — two ROPs on the same geometry with different Cd.

Same camera as Cinema 4D (verified)

  • Map the world spaces first by comparing a few parts' bboxes in both apps. Seen: C4D → Houdini = ×0.1 (a transform on the imported Alembic) and z negated (C4D's Alembic export flips handedness).
  • C4D cameras look down +Z, Houdini cameras down −Z. With z flipped, a C4D camera with identity rotation maps to a Houdini camera with rotation 0 at (x, y, −z) × scale.
  • Rotated / moving cameras (verified, pass matched the C4D render exactly): export the camera's global matrix per frame from C4D (off, v1, v2, v3). In Houdini, with F = diag(1, 1, −1): rows = (v1·F, v2·F, −(v3·F)), translate = (x, y, −z) × scale → hou.Matrix4(...).explode() → linear keys on tx…rz.
  • Lens: Houdini focal = C4D focal length; aperture = sensor width when the camera fits horizontally. Redshift Camera object (type 1057516) ids: 500 focal (mm), 7002 sensor size (vector, 36×24), 7005 Fit (0 Fill-Crop, 1 Horizontal, 2 Vertical, 3 Overscan-Fit, 4 Square), 7012 shift.
  • Bake the animation per frame: in C4D sample track.GetCurve().GetValue(c4d.BaseTime(f, fps)) for every frame and write JSON with open(); in Houdini parm.setKeyframes([...]) with linear() keys (9001 keys in one call, instant).
  • Verify framing: render the same frame of an engine part in both and compare silhouettes (matched within a few pixels at 1080p).
  • Cutaway views: geometry hidden in the C4D render (a cut-open liner) still exists as a Houdini collider. Drop points the cut view can't show (front half, behind solid moving parts — e.g. below the piston crown via a bbox() spare parm) before rendering.

Volume visualization for GL passes (verified)

volumevisualization SOP after the cache reader (object_merge of the sim's file SOP):

  • densityfield density. densityrampmode 'none' = raw density × densityscale; 'clamp' with rangemin/rangemax = min(d / range, 1) × scale (tames very dense frames, e.g. a burn at density 90).
  • Albedo: cdfield + cdramp (black smoke ~0.07–0.12 grey; white haze ~0.8).
  • Glow: emitfield (e.g. temperature) + emitrangeoverride, emitrange1/2, emitrampmode, emitscale; colour from emitcdfield + emitcdramp (dark red → orange → pale yellow).
  • Lights: a distant hlight::2.0 (light_type 'distant') + a dim envlight, via forcelights.
  • Starting values for a ~1-unit chamber at 0.01 voxels: black smoke (density 1–4) raw ×1.5; white haze clamp 0..2 ×2; glow temperature 0..1 ×3–4.
  • White haze over a dark plate: thin lit haze averages to grey over a dark background and reads as soot/black smoke (turning shadows off and albedo up barely changed it). Verified fix: make it self-lit — emitfield density, emitscale ~0.9, densityscale 3, white albedo and emission ramps, shadows 0 — and put the layer on Screen in the compositor so it can only brighten.

Checking passes

  • Alpha present? ffmpeg -i f.png -vf "alphaextract,format=gray,signalstats,metadata=mode=print" -f null - 2>&1 | grep YAVG
  • Over the other app's frame: ffmpeg -i bg.png -i pass.png -filter_complex "[0]scale=1920:1080[b];[b][1]overlay=format=auto" out.png (straight alpha). Look at it before rendering the full range.

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; in a public repo keep personal names and client details out of the text. Tell the user in one line what changed.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.