Water surfaces

Rendered water is a forward material drawn after the opaque composite. Its program reads a copy of the lit picture drawn so far and of the depth, and draws what lies behind the surface through the depth of water in front of it, with moving normal maps, reflections and the sun on top. This is apart from the collision surfaces that make water walkable or swimmable. Observed

Which materials draw water

In Elden Ring a material states water by the MATBIN parameter g_GXFT_WaterInteraction (a boolean). In the install’s material binder, 115 of the 20,619 MATBINs set it true, and they belong to 24 shader families: the lakes and swamps (M[Lake], M[PlacidLake] and its LOD variants, M[MuddyWater], M[PoisonSwamp] and its variants, M[MossSwamp]_Emissive_Glow), the rivers and oceans (M[Water][river], M[Water][ocean2] to [ocean7]), M[Oil_Liquid], M[ShadowErdtree_BlackBlood] and the cutscene water families. Proven · Elden Ring · every MATBIN of the material binder

The flag goes with what the family’s forward (_FwdSdw) program reads: over the 952 shader bundles of material families examined, every family whose MATBINs set the flag has a forward program that reads the water simulation’s normal map g_WaterSimulationNormalTexture (t57), and no family whose MATBINs leave it false or unset has one that does (743 family and flag pairs, none disagreeing). M[MossSwamp]_Emissive_Glow sets the flag and has no forward program among those bundles. Observed · Elden Ring · 952 material bundles, every forward program’s reflection

g_GXFT_Refraction is not a water marker: it is also set on clouds, ice, lava and bonfire materials. The waterfall families (M[Waterfall*]) refract and do not set g_GXFT_WaterInteraction. Observed · Elden Ring · every MATBIN of the material binder

Sekiro’s MTDs state the same flag. In the install’s allmaterialbnd, 28 of the 3,170 MTDs set g_GXFT_WaterInteraction true and none sets it false: M[Interaction]_Tessellation_Water (22), M[Interaction]_Tessellation_PoisonWater (2), M[Interaction]_Water (2), GXFlver_WaterBumpTriInteraction and GXFlver_WaterBumpTriInteractionFloatObj (1 each). The flowing water (M[Flow]_Water, 11; M[Flow]_Tessellation_Water, 2), the waterfalls (M[Mb1]_SmallWaterfall_test, 16) and the waterfall foam (M[WaterFall]_Foam, 5) do not state it. Proven · Sekiro · every MTD of the binder Its forward programs that read the water simulation’s normal map (at t58) are those of M[Interaction]_Tessellation_Water, M[Interaction]_Tessellation_PoisonWater and M[Interaction]_Water; the flowing-water families (M[Flow]_Water, M[Flow]_Tessellation_Water) read the frame’s colour and depth without it. Observed · Sekiro · 1,053 forward programs’ reflection

What the frame binds to a water draw

In one recorded Elden Ring frame (Stormveil, 2560x1440, five frames, 39 forward water draws: two M[PlacidLake] and one M[Water][ocean4] a frame), every water draw binds the following. Observed · Elden Ring · one capture of five frames

Register	Name	What the frame binds	Confidence
t45, t55	SceneColor_0_Screen_0, g_RefractionTexture	One full-screen R11G11B10 copy of the lit composite, made right before the water draws, which then draw into the composite itself	Observed · Elden Ring
t56	g_DepthTexture	A copy of the G-buffer depth	Observed · Elden Ring
t57	g_WaterSimulationNormalTexture	A 512x512 RGBA8 target the frame draws first, while water is drawn (GXWaterInteractionNormalMap2)	Observed · Elden Ring
t44	g_RLRTexture	The screen-space reflection chain’s result, half resolution RGBA16F with six levels	Observed · Elden Ring
t33	g_SSAOTexture	A 1x1 RGBA8 texture, not the ambient occlusion target	Observed · Elden Ring

The draws blend SRC_ALPHA, INV_SRC_ALPHA and test depth GREATER_EQUAL. M[PlacidLake]’s screen-space reflection prepass (FwdSsr) draws earlier, among the reflection chain’s steps. Observed · Elden Ring · one capture of five frames

A lake’s forward program

M[PlacidLake]_FwdSdw (Elden Ring) draws a lake as follows. Observed · Elden Ring · the installed program’s DXIL

• Waves. Normal map A (Texture2D_2, tiled by cbTextureTilingScale row 1) scrolls by float_11 * time * float2_2; its blue channel offsets the coordinate of normal map B (Texture2D_0, row 0), which scrolls by float2_4 * time * float_11 + float2_5 * uv. The tangent-space normal is A’s xy (pulled towards 0.5 by TEXCOORD4.y) plus (B - 0.5) * float_9, decoded as twice that less one: its x turns the normal along the bitangent (the normal crossed with the tangent, times the tangent’s w) and its y along the tangent, and the third component is sqrt(1 - saturate(x² + y²)). Inside the water simulation’s square (TEXCOORD0.zw strictly between 0 and 1) the simulation normal (t57 sampled there) replaces it by its alpha.
• What is behind. The screen coordinate is offset by the normal, by float_1 * 10 / (depth + 10), and the frame’s colour copy is read there; where the offset sample is in front of the surface, the straight coordinate is read instead.
• Absorption. What is behind is multiplied by exp2(-1.4427 * sigma * max(0, behind - surface)) per channel, with sigma = float3_0 + float3_2 and the two depths along the view axis: the surface’s from its position’s w, what is behind from the depth copy as -z / (x - stored) with x and z from cbSceneParam. m10_00_692, Stormveil’s lake, states sigma = (0.215, 0.115, 0.09) per metre.
• Fresnel. The transmitted light is weighted by 0.96 - 0.96 * (1 - |N.V|)^5: Schlick’s form with a reflectance of 0.04 at normal incidence.
• Caustics. A mask (Texture2D_3) sampled at the world xz of what is behind (from the depth copy), moved by a flow map (Texture2D_4) in two phases of frac(2 * float_8 * time), adds hourColour * behind * mask * float_7 * float_16 faded by the distance (float_26, float_27) and the screen-space shadow. hourColour is color_4 within five hours of noon, color_3 in the morning and color_5 in the afternoon, color_2 beyond seven hours.
• Reflection. Screen-space reflection over the two image-based-light cubes, the sun’s specular and the tiled point and spot lights’ specular.
• Body. A diffuse body FC_DifMapMulCol * saturate(color_0) * aColor[FC_ColorSlotNo] (the instance record’s colour in the slot cbMatDynParam names), lit by the sun (cbLight’s Col_0 times saturate(N.Vec_0) / pi, times min(1 + FC_SsaoWeight * (shadow - 1), 1)), is mixed over the rest by saturate(saturate((float_19 + (distance - float_17) / max(float_18 - float_17, 0.001)) * float_21) * float_5): a blend by the distance from the camera, not by the depth of water. m10_00_692 states 120 and 700 for the two distances, 0.2, 6 and 0.2 for the rest, so its body is blended in from 4 metres from the camera, by 0.012 at 10 metres, up to a fifth from 101 metres on. Its body is black: every recorded M[PlacidLake] draw binds color_0 as (0, 0, 0), the rest of its material constants equal to m10_00_692’s, a colour being its stored value raised to the power 2.2 and multiplied by its fifth value (color_2’s (0.506, 0.900, 1, 1, 8) is bound as (1.786, 6.347, 8)); color_0’s fifth value is 0. So far away the lake passes less of what is behind it, down to 0.8 of it, rather than showing a colour of its own. Observed · Elden Ring · the material constants of every lake draw in five recorded frames
• Sun highlight. The sun’s specular takes a smoothness of sqrt(smoothness * (1 - 0.075 / Vec_0.w)) from the light set’s Vec_0.w (see the frame).
• Alpha. The draw’s alpha is the instance’s times saturate((behind - surface) * float_3), a fade over the shore.

What does the water simulation draw?

GXWaterInteractionNormalMap2 draws the 512x512 normal map that interacting water reads near the player. What it draws from (the player’s and objects’ contacts) and how its square follows the camera are not described.

Do the other water families follow the lake’s model?

The lake’s program is read in full. The rivers, oceans, swamps and Sekiro’s interaction water are known to read the same frame inputs; whether their absorption, Fresnel and shore fade take the same form is not established.