Skill effects (Assets/Data/Effect/eftNNNN.eft + eftNNNN.alr)
Exported by
converter/effectconv.py <eftNNNN.alr> <dir>
→ effect.json (plus
pdconv <eft.alr> <dir> --textures
for tex_NNN.png); played by
godot/scripts/combat/pd_effect.gd
from <data>/effect/eftNNNN/.
Checked on the straight fire shot’s effect
(record 0x00 = its file number) against the
playtester’s arsenal-preview capture, and the
converter runs over all 477 effect files without
an error.
Correction (docs 11-10
xboxfx): the game does not play these
loose files. Both builds load effects from the
packs Effect/e.i +
aN.e, and the loose files are an
older revision that differs from the packs in
415 effects (“Effect packs and the Xbox / PC
effect comparison”). The layouts below hold for
both; counts taken over the loose files do not
describe the game.
- Not particle data in the exe’s
sense: an effect
.alris an ordinary ALR (Esper-RE’s layout) whose header offset table points at models only (no 0x5 animation chunks). Each model has a 0x1 material chunk (u32, then 0x4C-byte entries; shaderspdeftbb01andpdmtex01), a 0x3 joint chunk, a 0x16 vertex-buffer chunk and a list of elements. - Element = one 0x13 chunk (0x80 B)
followed by either a 0x14 chunk or a 0x2 index
buffer.
- 0x14 (0x2C B) names the
element’s kind in Esper-RE’s
6-characters-per-u32 text encoding (the same as
material shader names):
particle01,particle02,msparticle00/01,light01/02,tale01(trail),flair01,blade01/02,msbillbeam00,ykfade00(counts over all files: 1600particle01, 730msparticle01, 487light02, 339particle02, 97flair01, 77tale01). Then u16 material (Esper-RE’smaterial_idx) and u16 = the element’s joint (docs 11-12: not a running index; see “Effect instances, slots and joint attach”). The kinds match the exe’s shader set (Part1V/Part2V,Tale01V,Flair1V,Blade01V…). - 0x2 makes it a
mesh element: the index
buffer’s
texture_idxis a material entry,transform_idxa joint (its position places the mesh),vertex_bufthe 0x16 entry. Vertex format 0x09 (20 B): position 3 × f32, UV 2 × s16 / 4096, BGRA8 color (soft edges: grey ramps to black, alpha 255). Primitive 5 = list, 6 = strip. For lists, every index of the buffer counts (nidx); the header field read as the triangle count is the number of vertices used (the fire orb’s core sphere: 325 vertices, 1836 indices = 612 triangles, closed except at its UV seam). Taking only that many triangles dropped half of every effect mesh — the “hollow ring of wedges” of the first build. - 0x13 layout: 12 × f32
(
raw_f), two BGRA8 colors, two more BGRA8 colors (blink colors, see below), 16 × f32 (raw_m).
- 0x14 (0x2C B) names the
element’s kind in Esper-RE’s
6-characters-per-u32 text encoding (the same as
material shader names):
- Decoded with evidence:
raw_f[2],raw_f[3]= the element’s frame window (start, end; game ticks, 1/30 s). In the cast slot the small fireball meshes run 2–11 and the big set 12–24; in the capture (15 fps) the orb appears at the hand and grows for ~11 captured frames (~0.7 s), then is thrown.- The two colors are the start and end color (BGRA; e.g. orange → transparent on embers, black → orange on the light, white → transparent on the hit petals).
.eft(412 B, Esper-REeft.h): the six 0x1C entries at 0xF4 are slots,[count, 0, model, …, −1…](count observed: 1; model numbers index the ALR’s models). The 12 × 0x14 “high” entries are per-model settings (second u32 29 or 27 = the draw layer, docs 11-12).- (effects1: this paragraph and the joint list below describe the loose revision; the played pack entry differs, see “The fire shot in the packs” below.)
- Fire shot slots (from what each model holds
and the capture): 0 forming at the
hand (light, embers, growing fireball
meshes, 30 frames), 1 flight
(light,
tale01trail, an 83-vertex streak mesh on a joint 7.5 units behind, fire spheres, embers; 30 frames, looped while the shot flies), 2 hit (3 flame-petal meshes on joints turned 90°, frames 5–25, fading white → clear, plus a 5-frame flash; the petals lie 0.5–1.5 m on the effect’s back side (−Z in the game) and stretch to 2× along it; the exe plays it with the shot’s own matrix, not turned (see “Shot end”), so they splash back toward the shooter; an earlier reading of a battle capture had turned it 180°; seen side-on it is ~1–1.2 m), 3 launch smoke (three dark emitters, frames 0–15). The slot numbers mean the same for every effect, and slot 3 is the release on the caster, not at the shot (docs 11-12, “Effect instances, slots and joint attach”).
- Particles
(
particle01,particle02): decoded, see “Effect particles” below. Still inferred (pd_effect.gd marks them):msparticle00/01(own exe classes, read with the old particle guess); trails are decoded (“Effect trails” below); lightraw_f[0]= range.light02is a real dynamic light (it fills an entry of the exe’s light list: colour bytes × 1/255, position from the element’s matrix, range fromraw_f[0]→raw_f[1]; no draw, whatever its material says). Its update (0x14010c990) runs for start ≤ frame < end + 1 with an envelope weight w: over the firstraw_f[7]frames w rises through the curve int(raw_f[10]) (0: t / L, 1: t(t + 1) / L², 2: t(2L − t − 1) / L², the 0.5 and 2.0 constants at 0x1402f7d40 / 0x1402f2a10), over the lastraw_f[7]it falls the same way with curves 1 and 2 swapped and t counted from the end, and it is 1 between. Color bytes = start + (end − start) × w (truncated, × 1/255, rgb), range = lerp(raw_f[0],raw_f[1], w) game units. Unknown: the 0x3 joint rotations of effects. Mesh fields: see “Effect drawing”. - Where the rest is computed:
the exe, not the shaders.
Part1V_vs.csoonly expands a point into a quad (quad_pos/quad_uvtables in cbufferQuadMap, size = a per-vertex pixel size × (1/853, 1/480) in clip space); positions, sizes and colors arrive per vertex from the CPU. The template stringsEft0000.eft/Eft0000.alrin PDUWP.exe are only default names copied into 0x88-byte effect records (near RVA 0x18ca98); the loader itself reads the packs (docs 11-10 xboxfx, “Effect packs”).