Re-implement subsurface scattering.

The size settings are more "just works", with default scale and depth scale
values that don't need much tweaking.

Additionally, a "skin" mode was added so skin looks better.
EDIT: Cleaned up SSR filter shader a bit.

15 files changed, 388 insertions, 17 deletions

diff --git a/servers/rendering/rasterizer.h b/servers/rendering/rasterizer.h
index 6676b00723..d77963b3dd 100644
--- a/servers/rendering/rasterizer.h
+++ b/servers/rendering/rasterizer.h
@@ -266,6 +266,9 @@ public:
 	virtual void screen_space_roughness_limiter_set_active(bool p_enable, float p_curve) = 0;
 	virtual bool screen_space_roughness_limiter_is_active() const = 0;
 
+	virtual void sub_surface_scattering_set_quality(RS::SubSurfaceScatteringQuality p_quality) = 0;
+	virtual void sub_surface_scattering_set_scale(float p_scale, float p_depth_scale) = 0;
+
 	virtual bool free(RID p_rid) = 0;
 
 	virtual void update() = 0;
diff --git a/servers/rendering/rasterizer_rd/rasterizer_effects_rd.cpp b/servers/rendering/rasterizer_rd/rasterizer_effects_rd.cpp
index 3b22cb4d9d..9abdf55dfd 100644
--- a/servers/rendering/rasterizer_rd/rasterizer_effects_rd.cpp
+++ b/servers/rendering/rasterizer_rd/rasterizer_effects_rd.cpp
@@ -461,6 +461,53 @@ void RasterizerEffectsRD::screen_space_reflection(RID p_diffuse, RID p_normal, R
 	RD::get_singleton()->compute_list_end();
 }
 
+void RasterizerEffectsRD::sub_surface_scattering(RID p_diffuse, RID p_diffuse2, RID p_depth, const CameraMatrix &p_camera, const Size2i &p_screen_size, float p_scale, float p_depth_scale, RenderingServer::SubSurfaceScatteringQuality p_quality) {
+
+	RD::ComputeListID compute_list = RD::get_singleton()->compute_list_begin();
+
+	int32_t x_groups = (p_screen_size.width - 1) / 8 + 1;
+	int32_t y_groups = (p_screen_size.height - 1) / 8 + 1;
+
+	Plane p = p_camera.xform4(Plane(1, 0, -1, 1));
+	p.normal /= p.d;
+	float unit_size = p.normal.x;
+
+	{ //scale color and depth to half
+		sss.push_constant.camera_z_far = p_camera.get_z_far();
+		sss.push_constant.camera_z_near = p_camera.get_z_near();
+		sss.push_constant.orthogonal = p_camera.is_orthogonal();
+		sss.push_constant.unit_size = unit_size;
+		sss.push_constant.screen_size[0] = p_screen_size.x;
+		sss.push_constant.screen_size[1] = p_screen_size.y;
+		sss.push_constant.vertical = false;
+		sss.push_constant.scale = p_scale;
+		sss.push_constant.depth_scale = p_depth_scale;
+
+		RD::get_singleton()->compute_list_bind_compute_pipeline(compute_list, sss.pipelines[p_quality - 1]);
+
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_diffuse), 0);
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_diffuse2), 1);
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth), 2);
+
+		RD::get_singleton()->compute_list_set_push_constant(compute_list, &sss.push_constant, sizeof(SubSurfaceScatteringPushConstant));
+
+		RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
+
+		RD::get_singleton()->compute_list_add_barrier(compute_list);
+
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_diffuse2), 0);
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_uniform_set_from_image(p_diffuse), 1);
+		RD::get_singleton()->compute_list_bind_uniform_set(compute_list, _get_compute_uniform_set_from_texture(p_depth), 2);
+
+		sss.push_constant.vertical = true;
+		RD::get_singleton()->compute_list_set_push_constant(compute_list, &sss.push_constant, sizeof(SubSurfaceScatteringPushConstant));
+
+		RD::get_singleton()->compute_list_dispatch(compute_list, x_groups, y_groups, 1);
+
+		RD::get_singleton()->compute_list_end();
+	}
+}
+
 void RasterizerEffectsRD::merge_specular(RID p_dest_framebuffer, RID p_specular, RID p_base, RID p_reflection) {
 
 	RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(p_dest_framebuffer, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_READ, RD::INITIAL_ACTION_KEEP, RD::FINAL_ACTION_DISCARD, Vector<Color>());
@@ -1063,8 +1110,7 @@ void RasterizerEffectsRD::render_sky(RD::DrawListID p_list, float p_time, RID p_
 
 RasterizerEffectsRD::RasterizerEffectsRD() {
 
-	{
-		// Initialize blur
+	{ // Initialize blur
 		Vector<String> blur_modes;
 		blur_modes.push_back("\n#define MODE_GAUSSIAN_BLUR\n");
 		blur_modes.push_back("\n#define MODE_GAUSSIAN_GLOW\n");
@@ -1356,6 +1402,21 @@ RasterizerEffectsRD::RasterizerEffectsRD() {
 		ssr_scale.pipeline = RD::get_singleton()->compute_pipeline_create(ssr_scale.shader.version_get_shader(ssr_scale.shader_version, 0));
 	}
 
+	{
+		Vector<String> sss_modes;
+		sss_modes.push_back("\n#define USE_11_SAMPLES\n");
+		sss_modes.push_back("\n#define USE_17_SAMPLES\n");
+		sss_modes.push_back("\n#define USE_25_SAMPLES\n");
+
+		sss.shader.initialize(sss_modes);
+
+		sss.shader_version = sss.shader.version_create();
+
+		for (int i = 0; i < sss_modes.size(); i++) {
+			sss.pipelines[i] = RD::get_singleton()->compute_pipeline_create(sss.shader.version_get_shader(sss.shader_version, i));
+		}
+	}
+
 	RD::SamplerState sampler;
 	sampler.mag_filter = RD::SAMPLER_FILTER_LINEAR;
 	sampler.min_filter = RD::SAMPLER_FILTER_LINEAR;
@@ -1412,4 +1473,8 @@ RasterizerEffectsRD::~RasterizerEffectsRD() {
 	roughness_limiter.shader.version_free(roughness_limiter.shader_version);
 	cubemap_downsampler.shader.version_free(cubemap_downsampler.shader_version);
 	filter.shader.version_free(filter.shader_version);
+	ssr.shader.version_free(ssr.shader_version);
+	ssr_scale.shader.version_free(ssr_scale.shader_version);
+	ssr_filter.shader.version_free(ssr_filter.shader_version);
+	sss.shader.version_free(sss.shader_version);
 }
diff --git a/servers/rendering/rasterizer_rd/rasterizer_effects_rd.h b/servers/rendering/rasterizer_rd/rasterizer_effects_rd.h
index 32eb8c8cda..27f051d194 100644
--- a/servers/rendering/rasterizer_rd/rasterizer_effects_rd.h
+++ b/servers/rendering/rasterizer_rd/rasterizer_effects_rd.h
@@ -48,6 +48,7 @@
 #include "servers/rendering/rasterizer_rd/shaders/ssao.glsl.gen.h"
 #include "servers/rendering/rasterizer_rd/shaders/ssao_blur.glsl.gen.h"
 #include "servers/rendering/rasterizer_rd/shaders/ssao_minify.glsl.gen.h"
+#include "servers/rendering/rasterizer_rd/shaders/subsurface_scattering.glsl.gen.h"
 #include "servers/rendering/rasterizer_rd/shaders/tonemap.glsl.gen.h"
 
 #include "servers/rendering_server.h"
@@ -480,6 +481,29 @@ class RasterizerEffectsRD {
 		RID pipeline;
 	} ssr_scale;
 
+	struct SubSurfaceScatteringPushConstant {
+
+		int32_t screen_size[2];
+		float camera_z_far;
+		float camera_z_near;
+
+		uint32_t vertical;
+		uint32_t orthogonal;
+		float unit_size;
+		float scale;
+
+		float depth_scale;
+		uint32_t pad[3];
+	};
+
+	struct SubSurfaceScattering {
+
+		SubSurfaceScatteringPushConstant push_constant;
+		SubsurfaceScatteringShaderRD shader;
+		RID shader_version;
+		RID pipelines[3]; //3 quality levels
+	} sss;
+
 	RID default_sampler;
 	RID default_mipmap_sampler;
 	RID index_buffer;
@@ -571,6 +595,7 @@ public:
 
 	void screen_space_reflection(RID p_diffuse, RID p_normal, RS::EnvironmentSSRRoughnessQuality p_roughness_quality, RID p_roughness, RID p_blur_radius, RID p_blur_radius2, RID p_metallic, const Color &p_metallic_mask, RID p_depth, RID p_scale_depth, RID p_scale_normal, RID p_output, RID p_output_blur, const Size2i &p_screen_size, int p_max_steps, float p_fade_in, float p_fade_out, float p_tolerance, const CameraMatrix &p_camera);
 	void merge_specular(RID p_dest_framebuffer, RID p_specular, RID p_base, RID p_reflection);
+	void sub_surface_scattering(RID p_diffuse, RID p_diffuse2, RID p_depth, const CameraMatrix &p_camera, const Size2i &p_screen_size, float p_scale, float p_depth_scale, RS::SubSurfaceScatteringQuality p_quality);
 
 	RasterizerEffectsRD();
 	~RasterizerEffectsRD();
diff --git a/servers/rendering/rasterizer_rd/rasterizer_scene_high_end_rd.cpp b/servers/rendering/rasterizer_rd/rasterizer_scene_high_end_rd.cpp
index 6094fbb36f..685c30b5b8 100644
--- a/servers/rendering/rasterizer_rd/rasterizer_scene_high_end_rd.cpp
+++ b/servers/rendering/rasterizer_rd/rasterizer_scene_high_end_rd.cpp
@@ -1795,6 +1795,14 @@ void RasterizerSceneHighEndRD::_render_scene(RID p_render_buffer, const Transfor
 	render_list.clear();
 	_fill_render_list(p_cull_result, p_cull_count, PASS_MODE_COLOR, render_buffer == nullptr);
 
+	bool using_sss = render_buffer && scene_state.used_sss && sub_surface_scattering_get_quality() != RS::SUB_SURFACE_SCATTERING_QUALITY_DISABLED;
+
+	if (using_sss) {
+		using_separate_specular = true;
+		render_buffer->ensure_specular();
+		using_separate_specular = true;
+		opaque_specular_framebuffer = render_buffer->color_specular_fb;
+	}
 	RID radiance_uniform_set;
 	bool draw_sky = false;
 
@@ -1894,8 +1902,8 @@ void RasterizerSceneHighEndRD::_render_scene(RID p_render_buffer, const Transfor
 
 	RENDER_TIMESTAMP("Render Opaque Pass");
 
-	bool can_continue_color = !scene_state.used_screen_texture && !using_ssr && !scene_state.used_sss;
-	bool can_continue_depth = !scene_state.used_depth_texture && !using_ssr;
+	bool can_continue_color = !scene_state.used_screen_texture && !using_ssr && !using_sss;
+	bool can_continue_depth = !scene_state.used_depth_texture && !using_ssr && !using_sss;
 
 	{
 
@@ -1904,10 +1912,15 @@ void RasterizerSceneHighEndRD::_render_scene(RID p_render_buffer, const Transfor
 
 		//regular forward for now
 		Vector<Color> c;
-		c.push_back(clear_color.to_linear());
 		if (using_separate_specular) {
+			Color cc = clear_color.to_linear();
+			cc.a = 0; //subsurf scatter must be 0
+			c.push_back(cc);
 			c.push_back(Color(0, 0, 0, 0));
+		} else {
+			c.push_back(clear_color.to_linear());
 		}
+
 		RID framebuffer = using_separate_specular ? opaque_specular_framebuffer : opaque_framebuffer;
 		RD::DrawListID draw_list = RD::get_singleton()->draw_list_begin(framebuffer, keep_color ? RD::INITIAL_ACTION_KEEP : RD::INITIAL_ACTION_CLEAR, will_continue_color ? RD::FINAL_ACTION_CONTINUE : RD::FINAL_ACTION_READ, depth_pre_pass ? (using_ssao ? RD::INITIAL_ACTION_KEEP : RD::INITIAL_ACTION_CONTINUE) : RD::INITIAL_ACTION_CLEAR, will_continue_depth ? RD::FINAL_ACTION_CONTINUE : RD::FINAL_ACTION_READ, c, 1.0, 0);
 		_render_list(draw_list, RD::get_singleton()->framebuffer_get_format(framebuffer), render_list.elements, render_list.element_count, false, using_separate_specular ? PASS_MODE_COLOR_SPECULAR : PASS_MODE_COLOR, render_buffer == nullptr, radiance_uniform_set, render_buffers_uniform_set);
@@ -1950,10 +1963,9 @@ void RasterizerSceneHighEndRD::_render_scene(RID p_render_buffer, const Transfor
 
 	if (using_separate_specular) {
 
-		if (scene_state.used_sss) {
+		if (using_sss) {
 			RENDER_TIMESTAMP("Sub Surface Scattering");
-
-			//_process_sss()
+			_process_sss(p_render_buffer, p_cam_projection);
 		}
 
 		if (using_ssr) {
@@ -2620,6 +2632,8 @@ RasterizerSceneHighEndRD::RasterizerSceneHighEndRD(RasterizerStorageRD *p_storag
 		actions.render_mode_defines["diffuse_lambert_wrap"] = "#define DIFFUSE_LAMBERT_WRAP\n";
 		actions.render_mode_defines["diffuse_toon"] = "#define DIFFUSE_TOON\n";
 
+		actions.render_mode_defines["sss_mode_skin"] = "#define SSS_MODE_SKIN\n";
+
 		bool force_blinn = GLOBAL_GET("rendering/quality/shading/force_blinn_over_ggx");
 
 		if (!force_blinn) {
diff --git a/servers/rendering/rasterizer_rd/rasterizer_scene_rd.cpp b/servers/rendering/rasterizer_rd/rasterizer_scene_rd.cpp
index af86d145ad..e4c5f1c984 100644
--- a/servers/rendering/rasterizer_rd/rasterizer_scene_rd.cpp
+++ b/servers/rendering/rasterizer_rd/rasterizer_scene_rd.cpp
@@ -3203,6 +3203,26 @@ void RasterizerSceneRD::_free_render_buffer_data(RenderBuffers *rb) {
 	}
 }
 
+void RasterizerSceneRD::_process_sss(RID p_render_buffers, const CameraMatrix &p_camera) {
+
+	RenderBuffers *rb = render_buffers_owner.getornull(p_render_buffers);
+	ERR_FAIL_COND(!rb);
+
+	bool can_use_effects = rb->width >= 8 && rb->height >= 8;
+
+	if (!can_use_effects) {
+		//just copy
+		return;
+	}
+
+	if (rb->blur[0].texture.is_null()) {
+		_allocate_blur_textures(rb);
+		_render_buffers_uniform_set_changed(p_render_buffers);
+	}
+
+	storage->get_effects()->sub_surface_scattering(rb->texture, rb->blur[0].mipmaps[0].texture, rb->depth_texture, p_camera, Size2i(rb->width, rb->height), sss_scale, sss_depth_scale, sss_quality);
+}
+
 void RasterizerSceneRD::_process_ssr(RID p_render_buffers, RID p_dest_framebuffer, RID p_normal_buffer, RID p_roughness_buffer, RID p_specular_buffer, RID p_metallic, const Color &p_metallic_mask, RID p_environment, const CameraMatrix &p_projection, bool p_use_additive) {
 
 	RenderBuffers *rb = render_buffers_owner.getornull(p_render_buffers);
@@ -3547,6 +3567,19 @@ void RasterizerSceneRD::render_buffers_configure(RID p_render_buffers, RID p_ren
 	_render_buffers_uniform_set_changed(p_render_buffers);
 }
 
+void RasterizerSceneRD::sub_surface_scattering_set_quality(RS::SubSurfaceScatteringQuality p_quality) {
+	sss_quality = p_quality;
+}
+
+RS::SubSurfaceScatteringQuality RasterizerSceneRD::sub_surface_scattering_get_quality() const {
+	return sss_quality;
+}
+
+void RasterizerSceneRD::sub_surface_scattering_set_scale(float p_scale, float p_depth_scale) {
+	sss_scale = p_scale;
+	sss_depth_scale = p_depth_scale;
+}
+
 int RasterizerSceneRD::get_roughness_layers() const {
 	return roughness_layers;
 }
@@ -4101,6 +4134,9 @@ RasterizerSceneRD::RasterizerSceneRD(RasterizerStorageRD *p_storage) {
 	screen_space_roughness_limiter_curve = GLOBAL_GET("rendering/quality/filters/screen_space_roughness_limiter_curve");
 	glow_bicubic_upscale = int(GLOBAL_GET("rendering/quality/glow/upscale_mode")) > 0;
 	ssr_roughness_quality = RS::EnvironmentSSRRoughnessQuality(int(GLOBAL_GET("rendering/quality/screen_space_reflection/roughness_quality")));
+	sss_quality = RS::SubSurfaceScatteringQuality(int(GLOBAL_GET("rendering/quality/subsurface_scattering/subsurface_scattering_quality")));
+	sss_scale = GLOBAL_GET("rendering/quality/subsurface_scattering/subsurface_scattering_scale");
+	sss_depth_scale = GLOBAL_GET("rendering/quality/subsurface_scattering/subsurface_scattering_depth_scale");
 }
 
 RasterizerSceneRD::~RasterizerSceneRD() {
diff --git a/servers/rendering/rasterizer_rd/rasterizer_scene_rd.h b/servers/rendering/rasterizer_rd/rasterizer_scene_rd.h
index 11a8555d22..3f71f7dbc1 100644
--- a/servers/rendering/rasterizer_rd/rasterizer_scene_rd.h
+++ b/servers/rendering/rasterizer_rd/rasterizer_scene_rd.h
@@ -93,6 +93,7 @@ protected:
 
 	void _process_ssao(RID p_render_buffers, RID p_environment, RID p_normal_buffer, const CameraMatrix &p_projection);
 	void _process_ssr(RID p_render_buffers, RID p_dest_framebuffer, RID p_normal_buffer, RID p_roughness_buffer, RID p_specular_buffer, RID p_metallic, const Color &p_metallic_mask, RID p_environment, const CameraMatrix &p_projection, bool p_use_additive);
+	void _process_sss(RID p_render_buffers, const CameraMatrix &p_camera);
 
 	void _setup_sky(RID p_environment, const Vector3 &p_position, const Size2i p_screen_size);
 	void _update_sky(RID p_environment, const CameraMatrix &p_projection, const Transform &p_transform);
@@ -698,6 +699,9 @@ private:
 	RS::DOFBlurQuality dof_blur_quality = RS::DOF_BLUR_QUALITY_MEDIUM;
 	RS::DOFBokehShape dof_blur_bokeh_shape = RS::DOF_BOKEH_HEXAGON;
 	bool dof_blur_use_jitter = false;
+	RS::SubSurfaceScatteringQuality sss_quality = RS::SUB_SURFACE_SCATTERING_QUALITY_MEDIUM;
+	float sss_scale = 0.05;
+	float sss_depth_scale = 0.01;
 
 	mutable RID_Owner<CameraEffects> camera_effects_owner;
 
@@ -1110,6 +1114,10 @@ public:
 	virtual bool screen_space_roughness_limiter_is_active() const;
 	virtual float screen_space_roughness_limiter_get_curve() const;
 
+	virtual void sub_surface_scattering_set_quality(RS::SubSurfaceScatteringQuality p_quality);
+	RS::SubSurfaceScatteringQuality sub_surface_scattering_get_quality() const;
+	virtual void sub_surface_scattering_set_scale(float p_scale, float p_depth_scale);
+
 	int get_roughness_layers() const;
 	bool is_using_radiance_cubemap_array() const;
 
diff --git a/servers/rendering/rasterizer_rd/shaders/SCsub b/servers/rendering/rasterizer_rd/shaders/SCsub
index ade0418bd2..04a43e3251 100644
--- a/servers/rendering/rasterizer_rd/shaders/SCsub
+++ b/servers/rendering/rasterizer_rd/shaders/SCsub
@@ -25,4 +25,5 @@ if "RD_GLSL" in env["BUILDERS"]:
     env.RD_GLSL("screen_space_reflection.glsl")
     env.RD_GLSL("screen_space_reflection_filter.glsl")
     env.RD_GLSL("screen_space_reflection_scale.glsl")
+    env.RD_GLSL("subsurface_scattering.glsl")
     env.RD_GLSL("specular_merge.glsl")
diff --git a/servers/rendering/rasterizer_rd/shaders/scene_high_end.glsl b/servers/rendering/rasterizer_rd/shaders/scene_high_end.glsl
index 07f4770b14..955ae2e588 100644
--- a/servers/rendering/rasterizer_rd/shaders/scene_high_end.glsl
+++ b/servers/rendering/rasterizer_rd/shaders/scene_high_end.glsl
@@ -1697,6 +1697,9 @@ FRAGMENT_SHADER_CODE
 
 #else
 
+#ifdef SSS_MODE_SKIN
+	sss_strength = -sss_strength;
+#endif
 	diffuse_buffer = vec4(emission + diffuse_light + ambient_light, sss_strength);
 	specular_buffer = vec4(specular_light, metallic);
 
diff --git a/servers/rendering/rasterizer_rd/shaders/screen_space_reflection_filter.glsl b/servers/rendering/rasterizer_rd/shaders/screen_space_reflection_filter.glsl
index 671e289ed0..1a5dd5ab55 100644
--- a/servers/rendering/rasterizer_rd/shaders/screen_space_reflection_filter.glsl
+++ b/servers/rendering/rasterizer_rd/shaders/screen_space_reflection_filter.glsl
@@ -69,8 +69,6 @@ float gauss_weight(float p_val) {
 	return mix(gauss_table[idx], gauss_table[idx + 1], c);
 }
 
-#define GAUSS_WEIGHT(m_val) gauss_table[clamp(int(m_val * float(GAUSS_TABLE_SIZE - 1)), 0, GAUSS_TABLE_SIZE - 1)]
-
 #define M_PI 3.14159265359
 
 vec3 reconstructCSPosition(vec2 S, float z) {
@@ -105,17 +103,14 @@ void do_filter(inout vec4 accum, inout float accum_radius, inout float divisor,
 			break;
 		}
 
-		float contrib = 0.0;
 		if (d < radius) {
-			contrib += gauss_weight(d / radius);
-		}
 
-		if (contrib > 0.0) {
-			accum += imageLoad(source_ssr, tc) * contrib;
+			float w = gauss_weight(d / radius);
+			accum += imageLoad(source_ssr, tc) * w;
 #ifndef VERTICAL_PASS
-			accum_radius += r * contrib;
+			accum_radius += r * w;
 #endif
-			divisor += contrib;
+			divisor += w;
 		}
 	}
 }
diff --git a/servers/rendering/rasterizer_rd/shaders/subsurface_scattering.glsl b/servers/rendering/rasterizer_rd/shaders/subsurface_scattering.glsl
new file mode 100644
index 0000000000..41f8fde3ca
--- /dev/null
+++ b/servers/rendering/rasterizer_rd/shaders/subsurface_scattering.glsl
@@ -0,0 +1,198 @@
+/* clang-format off */
+[compute]
+
+#version 450
+
+VERSION_DEFINES
+
+
+
+layout(local_size_x = 8, local_size_y = 8, local_size_z = 1) in;
+
+/* clang-format on */
+
+#ifdef USE_25_SAMPLES
+const int kernel_size = 13;
+
+const vec2 kernel[kernel_size] = vec2[](
+		vec2(0.530605, 0.0),
+		vec2(0.0211412, 0.0208333),
+		vec2(0.0402784, 0.0833333),
+		vec2(0.0493588, 0.1875),
+		vec2(0.0410172, 0.333333),
+		vec2(0.0263642, 0.520833),
+		vec2(0.017924, 0.75),
+		vec2(0.0128496, 1.02083),
+		vec2(0.0094389, 1.33333),
+		vec2(0.00700976, 1.6875),
+		vec2(0.00500364, 2.08333),
+		vec2(0.00333804, 2.52083),
+		vec2(0.000973794, 3.0));
+
+const vec4 skin_kernel[kernel_size] = vec4[](
+		vec4(0.530605, 0.613514, 0.739601, 0),
+		vec4(0.0211412, 0.0459286, 0.0378196, 0.0208333),
+		vec4(0.0402784, 0.0657244, 0.04631, 0.0833333),
+		vec4(0.0493588, 0.0367726, 0.0219485, 0.1875),
+		vec4(0.0410172, 0.0199899, 0.0118481, 0.333333),
+		vec4(0.0263642, 0.0119715, 0.00684598, 0.520833),
+		vec4(0.017924, 0.00711691, 0.00347194, 0.75),
+		vec4(0.0128496, 0.00356329, 0.00132016, 1.02083),
+		vec4(0.0094389, 0.00139119, 0.000416598, 1.33333),
+		vec4(0.00700976, 0.00049366, 0.000151938, 1.6875),
+		vec4(0.00500364, 0.00020094, 5.28848e-005, 2.08333),
+		vec4(0.00333804, 7.85443e-005, 1.2945e-005, 2.52083),
+		vec4(0.000973794, 1.11862e-005, 9.43437e-007, 3));
+
+#endif //USE_25_SAMPLES
+
+#ifdef USE_17_SAMPLES
+const int kernel_size = 9;
+const vec2 kernel[kernel_size] = vec2[](
+		vec2(0.536343, 0.0),
+		vec2(0.0324462, 0.03125),
+		vec2(0.0582416, 0.125),
+		vec2(0.0571056, 0.28125),
+		vec2(0.0347317, 0.5),
+		vec2(0.0216301, 0.78125),
+		vec2(0.0144609, 1.125),
+		vec2(0.0100386, 1.53125),
+		vec2(0.00317394, 2.0));
+
+const vec4 skin_kernel[kernel_size] = vec4[](
+		vec4(0.536343, 0.624624, 0.748867, 0),
+		vec4(0.0324462, 0.0656718, 0.0532821, 0.03125),
+		vec4(0.0582416, 0.0659959, 0.0411329, 0.125),
+		vec4(0.0571056, 0.0287432, 0.0172844, 0.28125),
+		vec4(0.0347317, 0.0151085, 0.00871983, 0.5),
+		vec4(0.0216301, 0.00794618, 0.00376991, 0.78125),
+		vec4(0.0144609, 0.00317269, 0.00106399, 1.125),
+		vec4(0.0100386, 0.000914679, 0.000275702, 1.53125),
+		vec4(0.00317394, 0.000134823, 3.77269e-005, 2));
+#endif //USE_17_SAMPLES
+
+#ifdef USE_11_SAMPLES
+const int kernel_size = 6;
+const vec2 kernel[kernel_size] = vec2[](
+		vec2(0.560479, 0.0),
+		vec2(0.0771802, 0.08),
+		vec2(0.0821904, 0.32),
+		vec2(0.03639, 0.72),
+		vec2(0.0192831, 1.28),
+		vec2(0.00471691, 2.0));
+
+const vec4 skin_kernel[kernel_size] = vec4[](
+
+		vec4(0.560479, 0.669086, 0.784728, 0),
+		vec4(0.0771802, 0.113491, 0.0793803, 0.08),
+		vec4(0.0821904, 0.0358608, 0.0209261, 0.32),
+		vec4(0.03639, 0.0130999, 0.00643685, 0.72),
+		vec4(0.0192831, 0.00282018, 0.00084214, 1.28),
+		vec4(0.00471691, 0.000184771, 5.07565e-005, 2));
+
+#endif //USE_11_SAMPLES
+
+layout(push_constant, binding = 1, std430) uniform Params {
+
+	ivec2 screen_size;
+	float camera_z_far;
+	float camera_z_near;
+
+	bool vertical;
+	bool orthogonal;
+	float unit_size;
+	float scale;
+
+	float depth_scale;
+	uint pad[3];
+}
+params;
+
+layout(set = 0, binding = 0) uniform sampler2D source_image;
+layout(rgba16f, set = 1, binding = 0) uniform restrict writeonly image2D dest_image;
+layout(set = 2, binding = 0) uniform sampler2D source_depth;
+
+void do_filter(inout vec3 color_accum, inout vec3 divisor, vec2 uv, vec2 step, bool p_skin) {
+
+	// Accumulate the other samples:
+	for (int i = 1; i < kernel_size; i++) {
+		// Fetch color and depth for current sample:
+		vec2 offset = uv + kernel[i].y * step;
+		vec4 color = texture(source_image, offset);
+
+		if (abs(color.a) < 0.001) {
+			break; //mix no more
+		}
+
+		vec3 w;
+		if (p_skin) {
+			//skin
+			w = skin_kernel[i].rgb;
+		} else {
+			w = vec3(kernel[i].x);
+		}
+
+		color_accum += color.rgb * w;
+		divisor += w;
+	}
+}
+
+void main() {
+
+	// Pixel being shaded
+	ivec2 ssC = ivec2(gl_GlobalInvocationID.xy);
+
+	if (any(greaterThan(ssC, params.screen_size))) { //too large, do nothing
+		return;
+	}
+
+	vec2 uv = (vec2(ssC) + 0.5) / vec2(params.screen_size);
+
+	// Fetch color of current pixel:
+	vec4 base_color = texture(source_image, uv);
+	float strength = abs(base_color.a);
+
+	if (strength > 0.0) {
+
+		vec2 dir = params.vertical ? vec2(0.0, 1.0) : vec2(1.0, 0.0);
+
+		// Fetch linear depth of current pixel:
+		float depth = texture(source_depth, uv).r * 2.0 - 1.0;
+		float depth_scale;
+
+		if (params.orthogonal) {
+			depth = ((depth + (params.camera_z_far + params.camera_z_near) / (params.camera_z_far - params.camera_z_near)) * (params.camera_z_far - params.camera_z_near)) / 2.0;
+			depth_scale = params.unit_size; //remember depth is negative by default in OpenGL
+		} else {
+			depth = 2.0 * params.camera_z_near * params.camera_z_far / (params.camera_z_far + params.camera_z_near - depth * (params.camera_z_far - params.camera_z_near));
+			depth_scale = params.unit_size / depth; //remember depth is negative by default in OpenGL
+		}
+
+		float scale = mix(params.scale, depth_scale, params.depth_scale);
+
+		// Calculate the final step to fetch the surrounding pixels:
+		vec2 step = scale * dir;
+		step *= strength;
+		step /= 3.0;
+		// Accumulate the center sample:
+
+		vec3 divisor;
+		bool skin = bool(base_color.a < 0.0);
+
+		if (skin) {
+			//skin
+			divisor = skin_kernel[0].rgb;
+		} else {
+			divisor = vec3(kernel[0].x);
+		}
+
+		vec3 color = base_color.rgb * divisor;
+
+		do_filter(color, divisor, uv, step, skin);
+		do_filter(color, divisor, uv, -step, skin);
+
+		base_color.rgb = color / divisor;
+	}
+
+	imageStore(dest_image, ssC, base_color);
+}
diff --git a/servers/rendering/rendering_server_raster.h b/servers/rendering/rendering_server_raster.h
index 8480aa30fc..267efdbacb 100644
--- a/servers/rendering/rendering_server_raster.h
+++ b/servers/rendering/rendering_server_raster.h
@@ -541,6 +541,8 @@ public:
 	BIND5(environment_set_fog_height, RID, bool, float, float, float)
 
 	BIND2(screen_space_roughness_limiter_set_active, bool, float)
+	BIND1(sub_surface_scattering_set_quality, SubSurfaceScatteringQuality)
+	BIND2(sub_surface_scattering_set_scale, float, float)
 
 	/* CAMERA EFFECTS */
 
diff --git a/servers/rendering/rendering_server_wrap_mt.h b/servers/rendering/rendering_server_wrap_mt.h
index 13dad3ec18..d0f711838c 100644
--- a/servers/rendering/rendering_server_wrap_mt.h
+++ b/servers/rendering/rendering_server_wrap_mt.h
@@ -456,6 +456,8 @@ public:
 	FUNC5(environment_set_fog_height, RID, bool, float, float, float)
 
 	FUNC2(screen_space_roughness_limiter_set_active, bool, float)
+	FUNC1(sub_surface_scattering_set_quality, SubSurfaceScatteringQuality)
+	FUNC2(sub_surface_scattering_set_scale, float, float)
 
 	FUNCRID(camera_effects)
 
diff --git a/servers/rendering/shader_types.cpp b/servers/rendering/shader_types.cpp
index ba7b992e51..78fa2c2690 100644
--- a/servers/rendering/shader_types.cpp
+++ b/servers/rendering/shader_types.cpp
@@ -168,6 +168,8 @@ ShaderTypes::ShaderTypes() {
 
 	shader_modes[RS::SHADER_SPATIAL].modes.push_back("depth_test_disabled");
 
+	shader_modes[RS::SHADER_SPATIAL].modes.push_back("sss_mode_skin");
+
 	shader_modes[RS::SHADER_SPATIAL].modes.push_back("cull_back");
 	shader_modes[RS::SHADER_SPATIAL].modes.push_back("cull_front");
 	shader_modes[RS::SHADER_SPATIAL].modes.push_back("cull_disabled");
diff --git a/servers/rendering_server.cpp b/servers/rendering_server.cpp
index 55afd78fda..d417567a23 100644
--- a/servers/rendering_server.cpp
+++ b/servers/rendering_server.cpp
@@ -2362,6 +2362,13 @@ RenderingServer::RenderingServer() {
 
 	GLOBAL_DEF("rendering/quality/screen_space_reflection/roughness_quality", 1);
 	ProjectSettings::get_singleton()->set_custom_property_info("rendering/quality/screen_space_reflection/roughness_quality", PropertyInfo(Variant::INT, "rendering/quality/screen_space_reflection/roughness_quality", PROPERTY_HINT_ENUM, "Disabled (Fastest),Low, Medium, High (Slowest)"));
+
+	GLOBAL_DEF("rendering/quality/subsurface_scattering/subsurface_scattering_quality", 1);
+	ProjectSettings::get_singleton()->set_custom_property_info("rendering/quality/subsurface_scattering/subsurface_scattering_quality", PropertyInfo(Variant::INT, "rendering/quality/subsurface_scattering/subsurface_scattering_quality", PROPERTY_HINT_ENUM, "Disabled, Low (Fastest),Medium, High (Slowest)"));
+	GLOBAL_DEF("rendering/quality/subsurface_scattering/subsurface_scattering_scale", 0.05);
+	ProjectSettings::get_singleton()->set_custom_property_info("rendering/quality/subsurface_scattering/subsurface_scattering_scale", PropertyInfo(Variant::FLOAT, "rendering/quality/subsurface_scattering/subsurface_scattering_scale", PROPERTY_HINT_RANGE, "0.001,1,0.001"));
+	GLOBAL_DEF("rendering/quality/subsurface_scattering/subsurface_scattering_depth_scale", 0.01);
+	ProjectSettings::get_singleton()->set_custom_property_info("rendering/quality/subsurface_scattering/subsurface_scattering_depth_scale", PropertyInfo(Variant::FLOAT, "rendering/quality/subsurface_scattering/subsurface_scattering_depth_scale", PROPERTY_HINT_RANGE, "0.001,1,0.001"));
 }
 
 RenderingServer::~RenderingServer() {
diff --git a/servers/rendering_server.h b/servers/rendering_server.h
index 9726a71f0c..ac8e92be59 100644
--- a/servers/rendering_server.h
+++ b/servers/rendering_server.h
@@ -782,6 +782,16 @@ public:
 
 	virtual void screen_space_roughness_limiter_set_active(bool p_enable, float p_curve) = 0;
 
+	enum SubSurfaceScatteringQuality {
+		SUB_SURFACE_SCATTERING_QUALITY_DISABLED,
+		SUB_SURFACE_SCATTERING_QUALITY_LOW,
+		SUB_SURFACE_SCATTERING_QUALITY_MEDIUM,
+		SUB_SURFACE_SCATTERING_QUALITY_HIGH,
+	};
+
+	virtual void sub_surface_scattering_set_quality(SubSurfaceScatteringQuality p_quality) = 0;
+	virtual void sub_surface_scattering_set_scale(float p_scale, float p_depth_scale) = 0;
+
 	/* CAMERA EFFECTS */
 
 	virtual RID camera_effects_create() = 0;