Detailed view of one creature or object by index. Where
--list-zone-spawns shows headline fields in a table, --info-spawn
dumps every field for a single record:
Creatures: id, name, modelPath, displayId, position, orientation,
level, health/mana, faction, scale, behavior mode, wander/aggro/
leash radii, respawn time, flag set (hostile/questgiver/vendor/
trainer), patrol-path waypoint count.
Objects: uniqueId, path, type (m2/wmo), position, rotation, scale.
Both kinds support --json for downstream scripts. Useful for
spot-checking individual records discovered via list-zone-spawns or
audit-zone-spawns.
Verified: random-populated zone → info-spawn creature 0 shows
Spider with all behavior fields populated; object 0 shows mushroom
WMO with rotation/scale.
Offsets each vertex along its current normal by heightmap brightness
× scale. UVs determine where each vertex samples the heightmap;
sampling uses bilinear filtering with UV-wrap so repeating UVs
tile correctly.
Pairs naturally with --gen-mesh-grid: gen a flat grid, then
--displace-mesh with a noise PNG to create procedural terrain. Or
use it on a sphere to make a bumpy planet, on a cylinder for tree-
bark deformation, etc.
Output reports the actual delta range produced so the user can
gauge the scale. A hint suggests running --smooth-mesh-normals
afterward since the post-displacement shading would otherwise
follow the original (now-stale) flat normals.
Verified pipeline: --gen-mesh-grid (32×32, 1089 verts) → --gen-
texture-noise → --displace-mesh (scale 2) → bounds X/Y stay ±5
while Z spans 0..1.93.
Flat plane on the XY plane subdivided into NxN cells. Useful for
LOD demos, deformable surfaces (later --displace passes), and
testbench geometry that needs many triangles for benchmarking.
Default size 1.0 (centered on origin), subdivisions hard-capped at
256 so a typo can't generate 130k+ vertices accidentally.
Vertex count is (N+1)² and triangles are 2N², both reported in the
output so the user sees the budget. Verified: N=16 size=4 produces
289 verts / 512 tris.
EditorApp::auditSpawnsAgainstTerrain(threshold) counts every
creature + object whose Z is more than `threshold` yards off the
sampled terrain. Returns the issue count; non-mutating.
Generate menu gets a new "Audit Spawns Against Terrain" item that
runs the audit at the default 5y threshold and shows a toast: a
clean count if everything's fine, or the issue count + a hint to
run "Snap All" if not. Surfaces placement bugs without dropping
to the CLI.
Pairs with the existing "Snap All Spawns to Ground" so the
workflow stays inside the editor: audit → see count → snap →
audit again.
Generates <count> random zones in a single pass. Names default to
"Zone1, Zone2..."; tile coordinates step outward from (32, 32) in
a small raster so the zones don't overlap. Each zone gets a
unique sub-seed (seed + n) so its random content differs from its
siblings.
Args:
<count> required (1..100)
--prefix S name prefix (default "Zone")
--seed N base seed (default 100)
--creatures N per zone (default 20)
--objects N per zone (default 10)
--items N per zone (default 25)
Useful for spinning up a multi-zone playtest project from a single
command. Verified: count=3 with prefix Test produces Test1/Test2/
Test3 each with 4 creatures + 2 objects + 5 items, all on separate
tile coordinates around (32,32).
Project-wide companion to --list-zone-spawns. Combines creatures +
objects across every zone into one listing keyed by (zone, kind,
name) with position + a per-kind extra column (level for creatures,
scale for objects). Both human (table) and --json output.
Useful for project-wide review and for piping into spreadsheets via
--json — the zone column lets a pivot table group/filter by zone
which the per-zone command can't.
Composes scaffold-zone + random-populate-zone + random-populate-
items into a single invocation. Useful for "I just want a complete
test zone, don't make me chain three commands" — common case for
playtest fixtures and CI scenario setup.
Args:
<name> required (becomes the slug)
[tx ty] optional (default 32 32)
--seed N default 42 (items use seed+1 so they don't
correlate with creature/object placements)
--creatures N default 20
--objects N default 10
--items N default 25
Each sub-step's output streams through. If any step fails the
generator stops and reports which one. Slug-cleans the name to
match scaffold-zone's expectations.
Verified: TestZone with 5/3/8 counts produces a complete zone dir
containing creatures.json, items.json, objects.json, the .whm/.wot
tile pair, and zone.json.
One-command survey of "what's in this zone" without running both
--info-creatures and --info-objects separately. Pretty-prints two
tables (creatures and objects) with idx / level-or-type / position /
key-field columns. Also supports --json for downstream scripts.
Useful for quick spot-checks during placement and for piping into
review notes — the JSON form makes it easy to grep/jq for
specific entries.
Adds a small "Play" button next to the music and ambience dropdowns
that hands the selected file off to the OS default audio player —
'open' on macOS, 'xdg-open' on Linux, 'start ""' on Windows.
Buttons only appear when a file is set.
Avoids pulling SDL_mixer (or any new dep) into the editor binary
just for the preview workflow. Designers already have a working
audio player; this just routes the file there with one click
instead of having them hunt through Data/Sound/Music in their file
manager.
Closes the third leg of the original "dropdown + browse + play
audio to hear it" ask: dropdown is the auto-scanned combo from the
prior commit; browse-via-real-files works through the dropdown;
play preview works now via OS shell.
Project-wide companion to --audit-zone-spawns. Spawns the binary
per-zone (only those with creatures.json or objects.json),
streams the per-zone reports, and exits 1 if any zone has spawns
flagged.
Optional --threshold N flag is forwarded to each sub-invocation so
projects with stricter or looser placement requirements get
consistent audits across zones.
CI-friendly pre-release placement check: drop into a pipeline,
fail the build if any spawn is more than N yards off terrain.
Adds EditorApp::snapAllSpawnsToGround() and a menu item under
Generate, mirroring the --snap-zone-to-ground CLI for users who
want the action without context-switching to a terminal.
Walks every NPC + object, casts a downward ray from baseZ+500y,
and writes the hit Z into spawn.position.z. Patrol waypoints on
each NPC are snapped too. Marks the project dirty + auto-save
pending so the change persists without a manual Ctrl+S.
Existing per-selection "Snap Ground" button on the right panel is
unchanged; this is the bulk version for "I just edited terrain
under a populated area."
Non-destructive companion to --snap-zone-to-ground. Loads the
zone's terrain, samples the height under each creature + object,
and flags any whose Z is more than <threshold> yards (default 5)
off from the terrain.
Useful for surveying placement issues before deciding whether to
run --snap-zone-to-ground (which would silently rewrite every
spawn). The flagged report shows kind / delta / spawnZ / terrainZ /
name so the user can spot whether the spawn is buried, floating,
or just out of bounds.
Threshold configurable via --threshold N. Exit 1 if any issue is
flagged so CI can gate on placement validity.
Verified: zone with FloatyMurloc at Z=5000 (vs terrain Z=99.9)
correctly flags it with +4900.1 delta; GroundedWolf at Z=100 is
within threshold and not flagged.
Project-wide companion to --snap-zone-to-ground. Spawns the binary
per-zone (only zones that actually have creatures.json or
objects.json — pure-terrain zones are skipped to avoid noise),
streams the per-zone output, then aggregates a summary.
Useful after a project-wide terrain regen, or after batch-running
--random-populate-zone across multiple zones — one command snaps
all the spawns at once.
Verified: 2-zone test where only z1 has populated content correctly
selects z1 alone, snaps 3 creatures + 1 object, exits 0.
Project-wide companion to --info-zone-audio. Walks every zone in
<projectDir>, reads the audio fields, emits a per-zone yes/no
table for music + ambience plus the volume sliders. Counts how
many zones have music and how many have any ambience set so the
header tells you at a glance how much audio work remains.
Useful for spotting zones still missing audio assignment before a
release pass — rather than opening each zone in the editor to
check.
Verified: 2-zone project where only A has audio configured →
header reports "with music: 1, with ambience: 1" and the per-zone
table shows A=yes/yes, B=no/no.
Walks every creature in creatures.json and every object in
objects.json, samples the actual terrain height at each spawn's
(x, y), and writes that into the spawn's Z. Useful after terrain
edits or after --random-populate-zone if the spawn baseZ doesn't
match the carved terrain.
Height lookup: loads every WHM tile listed in zone.json, then for
each spawn finds the chunk containing its (x, y) and uses the
chunk's average heightmap height + base Z. Average rather than
bilinear because spawns don't need sub-yard precision and the
average dodges sampling-induced spikes near chunk seams.
Verified: random-populate-zone followed by snap-zone-to-ground on a
fresh tile snaps 4 creatures + 2 objects without errors. Brings
command count to 235.
Right-triangular prism that climbs along +X. Footprint is size×size
on the XY plane (centered in X, +Y from 0 to size); rises from
Z=0 at -X up to Z=size at +X. Useful for ramps onto platforms,
simple roof slopes, cliff-face placeholders.
Five faces emitted as quads (top slope, bottom, back-tall vertical
wall, two side triangles encoded as degenerate quads so the
indexing stays uniform): 20 verts / 10 tris (some side tris are
zero-area; renderers skip them cleanly).
Top-slope normal is computed from the actual rise/run so shading
shows the slope angle correctly. Help text on both --gen-mesh and
--gen-mesh-textured updated to advertise the new shape — primitive
set is now cube/plane/sphere/cylinder/torus/cone/ramp + the
gen-mesh-stairs/-from-heightmap commands.
The Zone Audio panel previously only offered five hardcoded preset
paths. Replaced with a recursive directory walk of <data>/Sound/
Music and <data>/Sound/Ambience for any .mp3/.wav/.ogg files.
Results cached after the first scan; a Refresh button forces a
rebuild for when the user drops new files in.
Two ImGui combos populate from the scan:
- "Music File" — picks from Sound/Music
- "Ambience File" — picks from Sound/Ambience
Each combo has a "(none)" option to clear. Selecting an entry
updates both the manifest field and the existing manual text input
buffer below, so users can fine-tune the path after picking from
the combo.
EditorApp gets a public getDataPath() so the UI can reach the
configured asset root without exposing the rest of the private
state.
Audio playback preview is the remaining piece — needs SDL_mixer
or similar wired into the editor; not in this commit.
Prints the music track, day/night ambience tracks, and the two
volume sliders stored in zone.json. Also supports --json. Useful
for spot-checking that the right audio assets are wired up before
bake/export, and for CI to assert the zone has audio configured.
Empty fields render as "(none)" in the human view; JSON form emits
the actual empty string. Brings command count to 234.
Previously every random-populated creature and object dropped at
baseZ (the manifest's flat height), which sat below the carved
terrain on hills and floated above it in valleys. After
generateCompleteZone the terrain is far from flat, so the spawns
showed up clipping or floating.
Added an inline groundZ(x, y) helper that casts a downward ray
from baseZ+500y at the spawn's (x, y) and uses the hit position's
Z. Falls back to baseZ if the ray misses terrain (shouldn't happen
inside the loaded tile bbox but is safer than NaN).
Both creature and object loops now consult groundZ. Snap-to-ground
is the runtime placer's default behavior anyway, so this lines up
with the rest of the editor's expectations.
Adds EditorApp::randomPopulateZone(creatureCount, objectCount, seed)
and a Generate-menu submenu with sliders for both counts plus a
seed input. Same logic and bestiary as the --random-populate-zone
CLI: 12 creature templates with level jitter, 5 placeholder WMO
prop types with randomized rotation/scale.
Spawns are placed within the loaded tile's world bbox, marked
dirty + auto-save pending so the populated content persists across
restarts.
The CLI command remains for CI / scripting / batch workflows; the
menu serves designers who want to populate from the GUI without
context-switching to a terminal.
Same-seed reproducibility: a hover hint reminds the user that the
seed determines the output deterministically.
The path tool gains a "Append selected <object|NPC> as path point"
button that captures the currently-selected entity's world position
into pathPoints_. Workflow: place an inn + a town hall, select inn
→ click append, select town hall → click append, click Apply Path.
The road is then drawn between the two object positions.
When no object/NPC is selected, the button area shows a hint
instead. The button auto-kicks pathCapture_ from None into Waiting*
on first use so the user doesn't have to remember the setup
sequence.
Works with the multi-point polyline path tool from the previous
commit — append three or more objects to route a road through them
in order.
Seeded random items.json populator. Pulls a quality-banded prefix
("Worn" → "Eternal" depending on rolled quality) plus a noun
("Sword", "Tome", "Cloak"...) for plausible names, then randomizes
itemLevel and stack size around quality-appropriate baselines.
Useful for playtest loot tables that need bulk content without
hand-typing each entry. Reproducible from --seed; respects
--max-quality so a "trash loot" pass won't accidentally drop
artifacts.
Verified on a fresh zone: 8 items rolled with the seed=5 produces
a sensible spread (1 epic Tome, 1 epic Cuirass, 1 uncommon Bow,
5 common/poor) and unique IDs starting at 1. Brings command count
to 233.
Walks <zoneDir>/zone.json to compute the world AABB the zone
occupies, then drops N random creatures and M random objects with
positions inside that bbox. Seeded LCG so the same seed always
produces the same population — useful for reproducible playtest
scenarios and for CI fixtures that need deterministic content.
Default flags: --seed 42 --creatures 20 --objects 10. Creatures
draw from a small built-in bestiary (Wolf/Boar/Bear/Spider/Bandit/
Kobold/Murloc/Skeleton/Wisp/Goblin/Stag/Crab) with level jitter
around a per-name baseline. Objects pick from a generic placeholder
prop set (Tree/Boulder/Bush/Stump/Mushroom). Both lists are inline
so users can extend them without dragging in the asset browser.
Verified: 5-creature 3-object run produces a creatures.json with
real names + level-9 wolves + behavior=2 (Wander) + an objects.json
with real WMO paths, positions inside the zone bbox, randomized
rotation + scale, and unique IDs. Brings command count to 232.
Path capture replaced with std::vector<glm::vec3> pathPoints_. The
flow is now: click "Click Start Point" → click terrain for first
point → keep clicking for additional waypoints → press Apply when
done, or Cancel to scrap. Apply iterates each consecutive pair as a
segment, calling carveRiver + paintAlongPath + fillWaterAlongPath
(or flattenRoad for road mode) per segment.
Hard-capped at 64 captured points so a runaway click handler can't
unboundedly grow the polyline. Toast at end reports the segment
count so users see the polyline took.
PathCapture states extended: None / WaitingStart / WaitingEnd /
WaitingMore. Backwards-compatible getPathStart()/getPathEnd()
return first/last points so the existing path-preview wiring keeps
working.
The river path tool used to carve the channel and texture the banks
but never added a water layer — users had to manually run
"Fill Water" afterward, which floods the entire tile. The fix is a
new TerrainEditor::fillWaterAlongPath() method that adds water
layers only to chunks the river segment passes through (within
width + chunk-half-diagonal of the line).
Per-chunk water height is set to the chunk's post-carve minimum
terrain height + 0.5y offset so the water sits visibly in the
channel without overflowing onto banks.
The river apply path now invokes carveRiver → paintAlongPath →
fillWaterAlongPath in sequence. Toast updated to mention all three.
Multi-point rivers are still next on the list — the underlying
math takes a single segment today, so a polyline river needs a UI
revamp to capture N points + a per-segment loop. Ack'd, not done
this commit.
CreatureSpawn::behavior defaulted to Stationary, so newly placed
NPCs would never move at runtime. The "NPCs do not patrol or
wander, they stay put" complaint was a default-value issue, not a
missing feature — switching the default to Wander (radius 10y)
gets normal NPCs roaming out of the box.
Added a tooltip on the Behavior combo making it explicit that this
is the runtime mode (the editor's preview doesn't run movement
logic; the behavior kicks in once the zone ships and the server
consumes the exported SQL). Lists what each mode means:
Stationary / Patrol / Wander / Scripted.
Existing zones with explicitly-set Stationary NPCs are preserved
on load — only fresh defaults are affected.
Three issues fixed in one pass:
1. Z-offset was 35 yards — nameplates floated way above the actual
NPC position. Dropped to 3.5 yards (just above an average
creature's head) so the label reads as attached.
2. Horizontal alignment was a fixed -30 px offset, which only looked
centered for ~6-character names. Switched to ImGui::CalcTextSize
and centering on the projected position.
3. Nameplates were always-on, which got noisy fast on populated
zones. Added "Nameplate on selected only" checkbox in the NPC
panel (default ON); when unchecked, all NPCs show their names.
Vertical position also slightly tightened (sy - textSz.y - 2 vs
sy - 10) so the text baseline aligns rather than floating below.
The "Create Crater at Cursor" button used the brush's last hover
position, which was stale by the time the user clicked the button —
the cursor had to move onto the button itself, dragging the brush
position along with it. Result: crater spawned somewhere between
where the user wanted and the button.
New flow: button arms a "click on terrain to place crater" mode
(captures the user's chosen radius/depth/rim at arm time). The next
left-click anywhere on terrain spawns the crater at the actual click
position, regardless of brush state. Button label changes to "ARMED
— click on terrain to place" while waiting, with a Cancel button +
Esc to dismiss.
Misses on terrain leave the mode armed so the user can retry without
re-pressing the button.
Camera key events were gated on io.WantCaptureKeyboard, which goes
true whenever any ImGui panel has focus — not just when typing into
a text widget. Hovering the cursor over a side panel silently
disabled the flycam, which read as "WASD doesn't move the camera."
Switched the gate to io.WantTextInput, which is true only while a
text widget is actively accepting input. Now WASD/QE/Shift work
exactly when they should: always, except while you're typing into a
field.
Previously generateCompleteZone() applied a hardcoded heightband
texture set (Tanaris sand → Elwynn grass → Barrens rock →
Dragonblight snow) regardless of which biome the user picked in the
"New Terrain" dialog. The biome's textures were correctly applied
by createNewTerrain, then immediately overwritten by the
hardcoded auto-paint pass. Result: every "Create + Generate" run
looked the same, masquerading as "the first biome sticks."
Fix: store the active biome on EditorApp and read it in
generateCompleteZone(), pulling the four-band texture set + slope
accent from getBiomeTextures() instead of hardcoding. Now selecting
Forest vs Desert vs Snow vs ... actually changes the painted
textures.
Estimated bytes-per-category breakdown for a single WOM. Numbers
are based on the in-memory struct sizes (vertex 40B, index 4B,
bone 22B, batch 16B, anim keyframe 44B, plus texture path strings)
— not the actual on-disk encoding (which has framing overhead) —
but the relative shares are accurate and help users decide where
shrinking efforts pay off.
Surfaces actionable tips when a category dominates: animations >
vertices → suggest --strip-mesh --anims with the saved KB; bones
non-trivial → suggest --bones for static placement; vertices
dominate → suggest a lower-poly variant.
Verified on procedural cube: 960B vertices (85%) + 144B indices
(13%) + 16B batches + 8B textures, total 1128 estimated vs 1184
on-disk (~5% framing overhead). Tip correctly fires for
vertex-dominated mesh. Brings command count to 231.
Extracts a grayscale heightmap PNG from a row-major W×H heightmap
mesh. Y values are normalized to 0..255 using the mesh bounds so
the output PNG uses the full dynamic range.
User supplies W and H explicitly: arbitrary meshes aren't
necessarily heightmap-shaped, and taking the dimensions as args
avoids guessing wrong on a mesh with vertex count W*H but a
different layout.
Round-trips with --gen-mesh-from-heightmap modulo the 1-byte
quantization step.
Verified: noise PNG (32×32) → heightmap mesh → exported PNG (32×32)
chain executes cleanly, height 0..1.984 mapped to 0..255 grayscale,
both PNGs share the same dimensions. Brings command count to 230.
Recomputes per-vertex normals as the area-weighted average of
incident face normals. The cross-product magnitude is twice the
triangle area, so larger faces contribute more to the local
direction — gives a clean smooth-shaded result on curved surfaces.
Use cases:
- Imported geometry has no normals (--import-obj leaves them zero
or face-flat).
- Custom transforms have desynced normals from positions.
- Faceted-by-construction meshes (cube, stairs) need a smooth
re-shade for stylistic reasons.
Degenerate verts (unreferenced or with sum that cancels to zero —
e.g., the two poles of a UV sphere) fall back to (0,1,0) rather
than leaving NaN; reported separately so the user sees how many.
Verified: sphere → 219 of 221 normalized + 2 degenerate poles
handled cleanly; minimal triangle → 3/3 normalized. Brings command
count to 229 (kArgRequired 210).
Converts a grayscale PNG into a heightmap mesh. Each pixel becomes
one vertex; brightness becomes Y. Mesh is centered on the XZ plane
with X spanning [-W*scaleXZ/2, +W*scaleXZ/2] and Z spanning the
same for H.
Defaults: scaleXZ=0.1 (a 64×64 PNG covers a 6.4×6.4 yard patch),
scaleY=2.0 (full white pixels rise 2 yards above black). Capped at
512×512 source images to keep vertex count bounded (the cap raises
to 262K verts for the largest legal input).
Normals are computed from finite differences against the height
field, giving smooth shading across the surface. Single batch +
one empty texture slot ready for downstream binding via
--add-texture-to-mesh.
Pairs naturally with --gen-texture-noise (synthetic terrain) or
any artist-painted heightmap exported from Photoshop / GIMP.
Adds stb_image.h include — implementation is already linked via
the existing tools/editor/stb_image_impl.cpp.
Verified: 64×64 noise heightmap → 4096 verts (= 64²) + 7938 tris
(= 2×63²), 6.4×6.4 span, height range 0..1.99; 1024×1024 input
rejected with size-cap error. Brings command count to 228.
Synthesizes a two-color stripe pattern PNG. Stripe width in pixels,
plus direction (diagonal default, or horizontal/vertical). Color
endpoints share the same RRGGBB / RGB hex syntax as other texture
gens.
Useful for caution tape, marble bands, hazard markers, racing-style
start/finish flags — patterns that --gen-texture's checker/grid
don't capture.
Verified: 128×128 diagonal yellow/black at 32px ('caution tape')
and 256×256 horizontal red/white at 16px both written successfully
with correct headers reported. Brings command count to 227 (and
the texture-synthesis family to 6: solid + checker + grid + linear
gradient + radial gradient + noise + stripes).
Concatenates two WOMs into a single output. Vertex buffer is the
two inputs spliced; the second mesh's indices are offset by the
first mesh's vertex count; texture slots from both are appended;
batches from the second mesh have their indexStart shifted by the
first's index count and their textureIndex shifted by the first's
texture-slot count.
Single-batch / no-batch inputs are auto-promoted to one synthetic
batch each so the merged output is always a well-formed v3.
Bones/animations are NOT merged — that requires skeleton retargeting
which is out of scope. If either input has bones, the merged output
is treated as static (bones cleared, weights reset to identity-on-
bone-0) so renderers don't read mismatched indices.
Verified: cube (24v/12t/1 batch) + sphere translated to +X (221v/
384t/1 batch) → combined (245v/396t/2 batches/2 textures), bounds
union (-0.5..3.5 in X), clean v3 reload via --info-mesh. Brings
command count to 226.
Mirrors every vertex position and normal across the chosen axis
(x, y, or z). Negating just one position component reverses face
winding (the triangle's signed area flips), so the second and
third index of every triangle is also swapped to keep front-faces
forward and lighting correct. Bone pivots mirror too.
Useful for "I have a left arm, mirror it for the right arm" content
reuse — the open-format mesh ecosystem can build symmetric content
from one half-asset without artist round-trips.
Verified: cube translated to (+5,0,0) mirrored across X → bounds
correctly land at (-5.5,-0.5,-0.5)..(-4.5,0.5,0.5); 24 vertices
touched / 12 triangles flipped reported; bad axis 'w' rejected
exit 1. Brings command count to 225.
Single CSV with every item across every zone in <projectDir>. The
zone name is the first column so a pivot table can group by it;
remaining columns mirror --export-zone-csv items output (index, id,
name, quality, itemLevel, displayId, stackable).
Saves running the per-zone CSV exporter N times and concatenating
manually. Reuses the existing CSV-escape lambda so commas and
quotes in names round-trip cleanly.
Verified: 2-zone project (A: Sword, "Helm, with comma" / B: Crown)
emits 3 rows with the comma'd name correctly double-quoted; zone
column lets a downstream pivot group correctly. Brings command
count to 224 (kArgRequired 205).
Copies items from one zone to another. Two modes:
Default (replace): destination items.json becomes a verbatim copy
of the source. Useful when zoneB is a fresh copy of zoneA's loot
table.
--merge: appends source items to existing destination items, but
preserves the destination's existing IDs by re-id'ing source
entries on collision. The destination's items take precedence; the
source's entries get fresh IDs picked from the smallest unused
positive integer.
Verified: merge with id-1 collision (dest has Boot id=1; source has
Sword id=1, Helm id=2) → Sword becomes id=2, then Helm becomes id=3
because 2 was just claimed; "re-ided: 2 (id collisions)" reported.
Replace mode wholesale-overwrites as expected. Brings command count
to 223.
Synthesizes a radial gradient PNG fading from <centerHex> at the
image center to <edgeHex> at the corner. Distance is normalized so
the corner is t=1 (works for non-square images), with a smoothstep
curve giving a soft falloff rather than a harsh disc edge.
Useful for spell glow rings, vignettes, soft-edged decals, and
particle-emitter masks — the common "circular blob" cases that
linear gradients can't produce.
Verified: 64×64 white-center / black-edge "glow" PNG written
successfully; invalid 'notacolor' hex rejected with exit 1.
Brings command count to 222.
Two convenience commands for the common "fix imported mesh" cases:
--center-mesh <wom-base>
Translates the mesh so the bounds center lands at the origin.
Useful when an OBJ comes in with its pivot in some weird corner
and the user wants center-pivoted placement. Pure translation —
shape unchanged, radius preserved.
--flip-mesh-normals <wom-base>
Inverts every vertex normal. Use case: an OBJ with flipped
winding renders inside-out — flipping normals fixes shading
without re-winding the index buffer (which would also need
batch-aware care). Also useful for skybox-like inside-facing
meshes.
Verified: cube translated to (5,0,0) → center brings it back to
±0.5 with shift line "(-5, -0, -0)" reported; flip reports 24
vertices touched. Brings command count to 221 (kArgRequired 202).
Rotates every vertex position and normal around the chosen axis
(x, y, or z) by <degrees>. Bone pivots also rotate so the skeleton
stays in sync. Bounds are recomputed from rotated positions
(axis-aligned bbox can grow under non-90° rotations).
Standard right-hand rule: positive degrees rotate counter-clockwise
when looking down the axis from positive infinity. Math is
hand-rolled (no glm::rotate dependency) for clarity at the call
site.
Together with --scale-mesh and --translate-mesh this completes the
basic transform set. Useful for "imported mesh is on its side" or
"the orientation flipped between Blender and Wowee" cleanup.
Verified: 2×2 plane on XY rotated 90° around X correctly lands on
XZ (bounds (-1,0,-1)..(1,0,1)); bad axis 'w' rejected with exit 1.
Brings command count to 219 — kArgRequired entry 200.
Synthesizes a deterministic smooth-noise PNG from a seed. Useful
for terrain detail overlays, dirt/grass blends, magic-fog
backdrops — anywhere a "natural-looking" pseudo-random texture
beats a flat color or grid.
Algorithm: 16×16 random lattice (LCG with numerical-recipes
constants), bilinearly interpolated per pixel with smoothstep on
the cell-local coords so seams don't show as bands. Cheaper than
perlin and produces a similar visual signal at this resolution.
Output is grayscale (R==G==B) so users can tint externally with
their own pipeline. Default 256×256, configurable W/H. Identical
seeds produce byte-identical PNGs across platforms (LCG is
dependency-free and platform-stable).
Verified: seed=42 + seed=42 → diff finds zero byte differences;
seed=42 + seed=99 → diff -q reports "differ" as expected. Brings
command count to 218.
Drops bones and/or animations from a WOM in place. Use case: a
model imported with full skeleton + anims that will only ever be
placed as static decoration — there's no point shipping the bone
data, and stripping it shrinks the file substantially.
Flags:
--bones drop bones AND animations (anims reference bones, so
keeping anims with no bones is meaningless). Vertex
skinning weights reset to identity-on-bone-0 so a
renderer that expects them doesn't read stale indices.
--anims drop only animations (keep bones for posed-but-static
use)
--all shorthand for --bones
Default (no flags) refuses with exit 1 so the user explicitly opts
in to destruction. Unknown flags also fail. Reports before/after
counts and byte delta so the user sees what they saved.
Verified: no-flags refuses; --all on already-static cube reports
0→0 / +0 bytes correctly; --foo rejected. Brings command count
to 217.
Two in-place mesh transforms covering the most common authoring
fix-ups:
--scale-mesh <wom-base> <factor>
Multiplies every vertex position, bone pivot, animation
translation keyframe, and bounds (min/max/radius) by <factor>.
Normals are unchanged (uniform scale preserves direction).
Useful for "I imported this OBJ but it's the wrong size" fixes.
Factor must be positive + finite.
--translate-mesh <wom-base> <dx> <dy> <dz>
Offsets vertices, bone pivots, and bounds by (dx, dy, dz).
Animation keyframes are bone-local so they're left alone — only
pivots shift. Radius stays constant (rigid translation).
Verified: unit cube scale 3x → bounds ±1.5, radius 2.598;
translate (10, 0, 0) → bounds (8.5,-1.5,-1.5)..(11.5,1.5,1.5);
negative scale (-1) rejected with exit 1. Brings command count
to 216.
One-shot composite of what --info-batches and per-mesh stats spread
across multiple commands. Reports name, version, counts (vertices/
triangles/bones/anims/batches/textures), bounds (min/max/radius),
per-batch detail (indexStart/indexCount/triangles/blend mode/texture
name), and the full texture-slot table.
Both human and --json output modes. Useful authoring command: pass
a WOM and see everything about it without running three sub-
commands.
Verified on a --gen-mesh-textured cube (size 1.5): correctly
reports v3 / 24 verts / 12 tris / 1 batch / 1 texture, bounds
±0.75, batch 0 wired to cube.png with opaque blend. Brings command
count to 214.
Project-wide companion to --export-zone-items-md. Walks every zone
that has items.json, emits one ITEMS.md document with: project
header + total + zone-with-items count, project-wide quality
histogram, then per-zone sections each containing a table
(ID/name/quality/ilvl/displayId/stack).
Easier to scan than running --export-zone-items-md N times — and
the project-wide histogram surfaces the loot landscape across the
whole project at a glance.
Verified on 2-zone project (1 common in zoneA + 1 legendary in
zoneB): histogram shows both qualities best-first, per-zone
sections render with correct counts. Brings command count to 213.
Synthesizes a linear two-color gradient PNG. Direction is "vertical"
(top→bottom, default) or "horizontal" (left→right). Colors use the
same RRGGBB / RGB hex syntax as --gen-texture, with optional
leading '#'.
Useful for the common "fade" cases that solid/checker/grid don't
cover: sky strips, UI backgrounds, glow rings, dirt-on-grass terrain
blends.
Verified: sky.png (87CEEB→FFFFFF vertical 256×256) and a 64×32
horizontal red→blue both write valid PNGs with correctly-parsed
endpoint RGB values reported back; invalid 'notacolor' hex fails
exit 1. Brings command count to 212.
