Material constant binding and UV tiling

Material definitions store authoring values; the engine uploads shader constants after applying type-dependent rules. Sekiro’s programs name individual material fields, while Elden Ring commonly addresses the SAT_MtdParam array. Observed

SAT positions and values

For a colour with stored RGB c, alpha a and intensity i, the observed upload is RGB = pow(c, 2.2) * i when the metaparam’s sRGB flag is set, and RGB = c * i when clear. Alpha remains a. The power precedes intensity; a colour-like name or type alone does not imply the power. Observed

Sekiro commonly stores intensity as a separate <colour-name>-Intensity parameter; Elden Ring stores it as the fifth colour component. Observed

Stored runtime override payloads are documented separately under MaterialEx action data; that payload layout does not establish its complete application order or transition rule. Unknown · Elden Ring

Texture groups and centered tiling

The metaparam joins each Texture2D_N to a UV group. The corresponding definition parameter is <group>_CommonUV-UVParam; the row of FC_TextureTilingScale is indexed by the actual bound register, not N. Observed

For scale (u,v), captured rows are (u,v,(1-u)/2,(1-v)/2), giving uv' = uv * scale + (1-scale)/2. Scaling is about the texture center. Observed

Source of scale	Evidence and scope	Confidence
FLVER texture scale before definition scale	Three discriminating Sekiro buffer contents support this precedence	Observed · Sekiro
Nonzero definition texture scale before UV group	Four discriminating Sekiro buffer contents support this precedence	Observed · Sekiro
Definition sampler scale or UV-group scale	Captured Elden Ring character rows match both paths	Observed · Elden Ring
Metaparam default when no authored UV-group value exists	Plausible fallback, but the cited Sekiro captures do not distinguish it	Inferred · Sekiro
Elden Ring non-unit FLVER scale precedence	The install has 1,203 bindings with a FLVER texture scale, in 288 materials of 22 MATBINs, but no reliably identified captured draw uses one; AEG006_031 to AEG006_045 among them are placed in no MSB	Unknown · Elden Ring

The blood amount

Both games’ programs declare FC_BloodAmount, one float in the material buffer cbMtdParam (b3), under the same name and at a different byte offset in each game. Observed The name and the blood decal it blends in suggest the engine raises it per instance as a character takes damage, rather than taking it from the material definition. Inferred

Game	Programs that declare it	Byte offset in cbMtdParam	Confidence
Sekiro	12,412 of the 21,304 programs of the shader bundles, 12,117 of them pixel programs; 48 more, none of them pixel programs, hold the name in no constant buffer’s layout	96 in every one	Proven · Sekiro · the reflection of every program of the shader bundles
Elden Ring	3,004 of 43,138 programs, 2,638 of them pixel programs	48 in every one	Observed · Elden Ring · the programs whose bytes hold the name, in the install with the Shadow of the Erdtree data, Steam build 25827938

In Elden Ring’s character G-buffer programs that read it, the amount blends the surface colour v towards the blood decal sampled from g_DecalDiffuseTexture: amount * ((decal.a * v - v) + decal.rgb) + v. At 0 the surface is unchanged. Observed · Elden Ring · the lifted value graphs of nine Stormveil materials’ programs: C[DetailBlend] 0 and 13, two M[AMSN] _Skin_PaintDecal programs and CS[c4750][VA_Frame]_Flag

only in Sekiro In five frames at the Ashina Castle gate, every drawn character material carries its definition’s g_BloodAmount there unchanged (1 on 760 draws, 0.9 on 5, 0 on 15), and every map and object material 0 (it has no such parameter). None of the pixel programs those draws use reads the member: their reflection marks it unused. Observed · Sekiro · five captured frames, 780 character and 2,225 map and object draws whose material buffer holds the word

In Elden Ring every character material carries 1 there: across seven Stormveil sessions, all 1,318 reads of the word by pixel programs of the C[...] families (DetailBlend in all its permutations, AMSN, Fur) and the P[ChrCustomize] hair, skin and body-hair programs are 1.0, and the 28 reads by P[ChrCustomize][Eye] and 7 by P[ChrCustomize][EyeAO] are 0.0. Observed · Elden Ring · seven captured sessions Of the 62 distinct G-buffer pixel programs of one frame, 25 read it as a value of an output lane: every C[...] program, the P[ChrCustomize] skin and eye programs, M[AMSN]_Skin_PaintDecal, M[AMSN][Ov_N]_Skin_PaintDecal and S[AMSN]_Glow. Observed · Elden Ring · traces of one frame Map and object materials hold 0 in most draws (all 191 of M[AMSN_V][Ov_N]_Gbuf), but a few programs read other values from the word (for example 1, 10 and 0.706 in M[AMSN_V]_Gbuf), which these captures do not explain. Observed · Elden Ring That every character material holds exactly 1 (and the eyes 0) in seven sessions of one walk suggests a definition value or an engine default, not a damage measure; the sessions did not record the character’s health. Inferred · Elden Ring

What amount does a character carry at run time?

Sekiro’s recorded frames hold the authored value and no drawn program reads it; Elden Ring’s character programs read a word that is 1 in all seven undamaged sessions. Whether damage changes the word in either game needs a frame of a wounded character, and the Elden Ring definition’s own FC_BloodAmount value (to tell a stored 1 from an engine-written one) is not read here.

Sekiro

SAT positions and values

In Sekiro’s programs SAT_MtdParam is a structure of named members. The metaparam’s MtdParam start values are zero there and do not locate them; the members’ offsets come from the program’s reflection, with nested members flattened. Observed · Sekiro

The _cloth families (Character_AMSN_[DetailBlend]_cloth, Character_AN_Blend_AMSN_[Fresnel]_cloth, Fur_NTC_cloth) name their material parameters, and their programs’ cbMtdParam members, with the base family’s prefix (Character_AMSN__DetailBlend__snp_0_*), not one built from their own shader name. A lookup by the _cloth shader’s name matches no member. Observed · Sekiro · the cloth materials of the Wolf’s default outfit

Sekiro’s non-unit intensity evidence comes from five independently attributed material-buffer blobs matched by live texture-tiling rows. Examples include colour (1,1,0.86274517) at intensity 10 uploading (10,10,7.22672558), and (0.78039223,0.64705884,0.56862748) at intensity 0.5 uploading approximately (0.289773375,0.191887334,0.144408002). These distinguish gamma-before-intensity from both raw multiplication and gamma-after-intensity; they are not a census of every material. Observed · Sekiro

Over five frames at the Ashina Castle gate, with each draw tied to its FLVER mesh and material through the mesh buffer (the buffer is the FLVER’s data block as stored), the definition’s values reach cbMtdParam as follows: 43,417 float parameters as stored, bit for bit; 3,515 colours whose RGB is pow(c, 2.2) * i and not c * i, none the other way, and 7,891 that cannot tell the two apart (0 or 1), alpha as stored in all of them. Observed · Sekiro · five captured frames The sky dome’s material m[sky]_skydome_night is the exception: 9 of its colours and 9 of its floats hold other values than its definition’s in every frame, so the engine writes them at run time. Observed · Sekiro · five captured frames

The definition’s alpha reference g_AlphaRef (two integers a, b) reaches FC_AlphaRef converted: x = (a - 0.0001) / 255, or -100 when a is 0, and y = b / 255. Observed[3,989 draws of five frames, bit for bit for a = 64, 70, 96, 100 and 128 and b = 128 and 192]{games=“sekiro”} The pixel programs discard where the tested alpha is below x (alpha inputs), so -100 is no test and the small offset keeps a texel of exactly a / 255. Inferred · Sekiro Raising x to 2 in a replay of one of the G-buffer pass’s depth-only draws (its colour writes masked off) leaves the depth it wrote in the game, 23.8% of the screen, unwritten. Observed · Sekiro · one replayed draw

Texture groups and tiling

With draws tied to their mesh and material through the mesh buffer, the FC_TextureTilingScale row of every material texture a pixel program samples, at its register, is (u, v, (1-u)/2, (1-v)/2), centred, in all 17,040 rows of five frames at the Ashina Castle gate. Observed · Sekiro · five captured frames The scale comes from:

Case	Scale in the row	Rows	Confidence
The FLVER binding of the texture’s kind has a path	the FLVER binding’s scale (all 203 rows where it is not 1 match it, and 438 rows whose definition scale is not 1 hold the binding’s 1)	641	Observed · Sekiro
The binding’s path is empty	the definition’s texture scale (the two floats after its path)	5,707	Observed · Sekiro
The binding’s path is empty and the definition’s scale is 0	1	10	Observed · Sekiro
The definition has group_N_CommonUV-UVParam parameters	one group’s scale (240 rows), the definition’s texture scale (20) or 1 (210, among them the character families’ base textures)	470	Observed · Sekiro

Which group each texture belongs to is the metaparam’s. Read from the rows instead, in the 121 rows of two Ashina frames whose texture has no FLVER path and whose material has group parameters, Character_AMSN_[DetailBlend] textures 1 and 6 take group_2, textures 2 and 5 take group_3 and textures 10 and 12 take group_4, each with no row that contradicts it (the other group’s non-unit scale never appears in them), while textures 0, 3 and 7 (the base colour, normal and mask) take no group: their rows are 1 whichever scale a group holds. Observed · Sekiro · 121 rows of two captured frames Each pair is a detail colour with its normal map. Inferred · Sekiro A comparison with the metaparam’s own table, which is not read here, would make it a rule.

Elden Ring

SAT positions and values

The metaparam’s MtdParam start is an absolute offset in 32-bit words from the beginning of the constant buffer, not from the beginning of SAT_MtdParam. An array starting at byte 144 therefore begins at word 36. Observed · Elden Ring

Elden Ring capture comparisons distinguish gamma conversion on 124 RGB lanes, raw multiplication on eight unflagged lanes, and integer bit patterns on three nonzero booleans. The session contains 260 SAT draw groups, 97 tied to material definitions by shaders and textures; four eye lanes remain unexplained. Observed · Elden Ring

In full, the 260 groups are 97 tied, 70 untied, 73 with no SAT binding or program and 20 without a captured constant buffer, and 1,049 lanes are classified. Of the type-13 RGB lanes whose sRGB flag is set, 124 fit only the power form, none only the raw form, and 215 cannot tell them apart (values 0 or 1). Of the unflagged ones, 8 fit only the raw form, none only the power form, and 1 cannot tell. Observed · Elden Ring An independent check of the unflagged M_AMSN_V__cheap_A__snp_0_color_2 in AEG004_113, AEG004_110 and AEG004_171_exLOD matches all 12 RGBA outputs to raw RGB × intensity with alpha unchanged, within 4.03e-8; among 15 other materials, 21 flagged RGB lanes at intensities 1, 2, 3, 3.5, 4, 5 and 20 match pow(c, 2.2) × i within 3.97e-7. Observed · Elden Ring

A true Elden Ring SAT boolean uploads word 0x00000001, not float 1.0’s 0x3F800000. Zero-valued booleans cannot distinguish these encodings. Observed · Elden Ring

Beside SAT_MtdParam, some programs read a SAT_CAParam block. The plain M[AMSN_V] G-buffer pixel program declares four float members of it at bytes 272, 276, 280 and 284 of cbMatDynParam and multiplies the sampled albedo’s RGB by the first three. The bundle’s metaparam lists the matching CAParams as components 0 to 4 of color_0, each with default 1; the program does not read the fifth, the intensity. The names differ by one character: the metaparam’s has an extra underscore before snp. Observed[the installed program’s reflection and IR, and its bundle]{games=“elden-ring”} In the recorded Stormveil frame that program’s 47 readable material buffers hold (1, 1, 1, 1) there, the defaults. Observed · Elden Ring · one frame The M[AMSN_V][Mb3] and M[AMSN_V][Ov_N] bundles have no CAParams. Observed · Elden Ring

Texture groups and tiling

Over the 299,105 texture bindings of Elden Ring’s asset and map FLVERs, 81,470 have FLVER scale 1 with a UV-group scale other than 1, 1,058 have a FLVER scale other than 1 with UV-group scale 1, and 10 have both other than 1 (8 of them equal). Another 34,845 belong to no UV group: fixed-function and MultiBlend materials. Observed · Elden Ring

Elden Ring draws with the same shader and textures can have different tiling constants. A Limgrave capture has both 10/35/10 and 19/55/24 scale groups under identical declared texture resources, so shader-plus-texture identity alone does not establish material identity. Observed · Elden Ring Nor can that family’s material-constant buffer tell its materials apart: all 10 parameters of its metaparam are renderer-only, at offset 0. Observed · Elden Ring

The eye pixel program’s material buffer holds colour 100/255 and scalar 0.024705887 on all 17 of its 30 captured draws that have one, and its textures tie equally to the c9506 eye and the FG_A_1500 left and right MATBINs. Observed · Elden Ring

Which runtime inputs account for unmatched tiling and eye words?

Sekiro’s tiling rows are all explained (above), the group case by the inferred membership. Elden Ring’s non-unit FLVER scale precedence has no decisive capture; placed assets whose two scales disagree would decide it, such as AEG004_805 in m60_42_37_00 (FLVER scale 2, UV group scale 1) and AEG007_200 in m60_23_19_01 (160 against 30). The eye program is P[ChrCustomize][Eye]_Gbuf, the character-creator family: its two draws per frame (one with the boolean set) hold the same words, colour 100/255 and scalar 0.024705887 among them, in seven sessions, so nothing in the frames tells a stored value from one set at run time. The family name makes a customisation override likely, but a frame with a different appearance chosen would decide it. Unknown