//! Noise & Grain effects, batch 2: Match Grain (grain statistics measured from another layer //! and synthesised onto this one), Median (Legacy) and Curl Noise (divergence-free flow //! distortion from the curl of a fractal noise potential). use effectcraft_keyframe::Value; use effectcraft_project::ParamUi; use effectcraft_raster::Image; use rayon::prelude::*; use crate::noise::{GrainLook, draw_boxes, fbm, grain_params, median_image, preview_params, sampling_boxes, sampling_params}; use crate::util::{Plane, gauss_plane, layer_or_self, premul, unpremul}; use crate::{Buf, EffectCtx, EffectSpec, num, p, popup, slider}; fn spec(id: &'static str, name: &'static str, params: Vec, render: crate::RenderFn) -> EffectSpec { EffectSpec { id, name, category: "Noise & Grain", params, render, gpu: false, float: true } } fn pct() -> ParamUi { slider(0.0, 100.0, 0.0, 100.0, 1) } // ---------------------------------------------------------------- Match Grain /// Grain statistics of an image: per-channel standard deviation of the high-pass residual /// (over opaque pixels) and a grain size from the lag-1 autocorrelation. #[derive(Clone, Copy, Debug, Default, PartialEq)] pub(crate) struct GrainStats { pub std: [f32; 3], pub size: f32, } #[cfg(test)] pub(crate) fn grain_stats(img: &Image) -> GrainStats { grain_stats_masked(img, None) } /// [`grain_stats`] over the pixels `mask` selects (all opaque pixels without one). pub(crate) fn grain_stats_masked(img: &Image, mask: Option<&[bool]>) -> GrainStats { let sel = |i: usize| mask.is_none_or(|m| m[i]); let (w, h) = (img.width as usize, img.height as usize); if w < 3 || h < 3 { return GrainStats::default(); } let straight: Vec<([f32; 3], f32)> = img.data.par_iter().map(|&p| unpremul(p)).collect(); let mut std = [0.0f32; 3]; let mut rho_num = 0.0f64; let mut rho_den = 0.0f64; for (k, s) in std.iter_mut().enumerate() { let pl = Plane { w, h, data: straight.iter().map(|(c, _)| c[k]).collect() }; let low = gauss_plane(&pl, 1.5, 1.5); let hp: Vec = pl.data.iter().zip(low.data.iter()).map(|(a, b)| a - b).collect(); let (mut sum, mut sq, mut n) = (0.0f64, 0.0f64, 0.0f64); for (i, v) in hp.iter().enumerate() { if straight[i].1 > 0.5 && sel(i) { sum += *v as f64; sq += (*v as f64) * (*v as f64); n += 1.0; } } if n > 1.0 { let mean = sum / n; *s = ((sq / n - mean * mean).max(0.0)).sqrt() as f32; } for y in 0..h { for x in 0..w - 1 { let i = y * w + x; if straight[i].1 > 0.5 && straight[i + 1].1 > 0.5 && sel(i) && sel(i + 1) { rho_num += (hp[i] * hp[i + 1]) as f64; rho_den += (hp[i] * hp[i]) as f64; } } } } let rho = if rho_den > 1e-12 { (rho_num / rho_den).clamp(0.0, 0.95) } else { 0.0 }; GrainStats { std, size: (1.0 / (1.0 - rho as f32)).clamp(0.5, 20.0) * 0.6 } } /// Match Grain's Viewing Mode options. const MATCH_GRAIN_VIEWS: [&str; 5] = ["Preview", "Noise Samples", "Compensation Samples", "Blending Matte", "Final Output"]; fn match_grain(ctx: &EffectCtx, mut b: Buf) -> Buf { if ctx.params.get("noiseSourceLayer").and_then(Value::as_layer).is_none() { return b; } // MATCH_GRAIN_VIEWS order. let view = ctx.params.e("viewingMode"); let source = layer_or_self(ctx, &b, "noiseSourceLayer", false, false); let src_boxes = sampling_boxes(ctx, &source, b.scale); let own_boxes = sampling_boxes(ctx, &b.img, b.scale); let box_col = ctx.params.get("sampling/sampleBoxColor").map(|v| v.as_color()).unwrap_or([1.0, 1.0, 0.0, 1.0]); if view == 1 || view == 2 { // Noise Samples (on the source) / Compensation Samples (on this layer). let (mut img, boxes) = if view == 1 { (source, src_boxes) } else { (b.img.clone(), own_boxes) }; draw_boxes(&mut img, &boxes, box_col); b.img = img; return b; } let mask_of = |img: &Image, boxes: &[(usize, usize, usize)]| -> Option> { if boxes.is_empty() { return None; } let w = img.width as usize; let mut m = vec![false; img.data.len()]; for &(x, y, s) in boxes { for yy in y..y + s { for xx in x..x + s { m[yy * w + xx] = true; } } } Some(m) }; let src_stats = grain_stats_masked(&source, mask_of(&source, &src_boxes).as_deref()); let comp = (ctx.params.f("compensateForExistingNoise") / 100.0) as f32; let own = if comp > 0.0 { grain_stats_masked(&b.img, mask_of(&b.img, &own_boxes).as_deref()) } else { GrainStats::default() }; let orig = b.img.clone(); let look = GrainLook::from_params(ctx); let amp: [f32; 3] = [0, 1, 2].map(|k| { let s = src_stats.std[k]; let e = own.std[k] * comp; (s * s - e * e).max(0.0).sqrt() * look.intensity * look.channel_intensity[k] }); let size = (src_stats.size.max(0.5) * look.size * b.scale as f32).max(0.05); let (w, h) = (b.img.width as usize, b.img.height as usize); let mut planes = look.planes(w, h, size, b.scale, 131); // Normalise each synthetic plane to unit standard deviation so the measured stats transfer. for pl in planes.iter_mut() { let n = pl.data.len().max(1) as f64; let mean = pl.data.iter().map(|&v| v as f64).sum::() / n; let var = pl.data.iter().map(|&v| (v as f64 - mean).powi(2)).sum::() / n; let k = if var > 1e-12 { 1.0 / var.sqrt() } else { 0.0 }; *pl = pl.map(|v| ((v as f64 - mean) * k) as f32); } b.img.data.par_iter_mut().enumerate().for_each(|(i, px)| { let (c, a) = unpremul(*px); let raw = if look.mono { [planes[0].data[i]; 3] } else { [planes[0].data[i], planes[1].data[i], planes[2].data[i]] }; let g = look.color(raw); let weight = look.weight(c); match view { 3 => *px = [weight.clamp(0.0, 1.0), weight.clamp(0.0, 1.0), weight.clamp(0.0, 1.0), 1.0], _ => { if a <= 0.0 { return; } let o = [0, 1, 2].map(|k| look.blend(c[k], g[k] * amp[k] * weight)); *px = premul(o, a); } } }); if view == 0 { crate::noise::preview_compose(ctx, &mut b, &orig); } b } // ---------------------------------------------------------------- Median (Legacy) fn median_legacy(ctx: &EffectCtx, mut b: Buf) -> Buf { let r = (ctx.params.f("radius") * b.scale).round().max(0.0) as usize; if r == 0 { return b; } if ctx.params.b("operateOnAlphaChannel") { b.img = median_image(&b.img, r); return b; } // Colour only: median of straight colour, alpha kept. let mut straight = b.img.clone(); straight.data.par_iter_mut().for_each(|px| { let (c, _) = unpremul(*px); *px = [c[0], c[1], c[2], 1.0]; }); let m = median_image(&straight, r); b.img.data.par_iter_mut().zip(m.data.par_iter()).for_each(|(px, q)| { *px = premul([q[0], q[1], q[2]], px[3]); }); b } // ---------------------------------------------------------------- Curl Noise fn curl_noise(ctx: &EffectCtx, mut b: Buf) -> Buf { let amount = ctx.params.f("displacementAmount") * b.scale; let view = ctx.params.e("view"); if amount.abs() < 1e-6 && view == 0 { return b; } let scale = (ctx.params.f("scale").max(1.0) * b.scale) as f32; let complexity = ctx.params.f("complexity") as f32; let rot = ctx.params.f("rotation").to_radians() as f32; let off = ctx.params.v2("offset"); let (ox, oy) = ((off[0] * b.scale + b.offset[0]) as f32, (off[1] * b.scale + b.offset[1]) as f32); let evo = (ctx.params.f("evolution") / 360.0) as f32; let seed = ctx.params.f("randomSeed") as i64 as u32 ^ 0x51_7cc1; let steps = ctx.params.f("steps").clamp(1.0, 16.0) as usize; let edge_wrap = ctx.params.e("edgeBehavior") == 1; let (w, h) = (b.img.width as usize, b.img.height as usize); let (cr, sr) = (rot.cos(), rot.sin()); // Potential field on the pixel grid. let mut pot = Plane::new(w, h); pot.data.par_chunks_mut(w.max(1)).enumerate().for_each(|(y, row)| { for (x, v) in row.iter_mut().enumerate() { let (dx, dy) = (x as f32 + 0.5 - ox, y as f32 + 0.5 - oy); let u = (dx * cr + dy * sr) / scale; let vv = (-dx * sr + dy * cr) / scale; *v = fbm(u, vv, evo * 4.0, seed, complexity); } }); // Curl: v = (dψ/dy, -dψ/dx), normalised so the typical speed is ~1 px per unit amount. let mut vx = Plane::new(w, h); let mut vy = Plane::new(w, h); vx.data.par_chunks_mut(w.max(1)).zip(vy.data.par_chunks_mut(w.max(1))).enumerate().for_each(|(y, (rx, ry))| { for x in 0..w { let (xi, yi) = (x as i64, y as i64); let gx = (pot.get_clamped(xi + 1, yi) - pot.get_clamped(xi - 1, yi)) * 0.5; let gy = (pot.get_clamped(xi, yi + 1) - pot.get_clamped(xi, yi - 1)) * 0.5; rx[x] = gy * scale; ry[x] = -gx * scale; } }); if view == 1 { b.img.data.par_iter_mut().enumerate().for_each(|(i, px)| { let (a, c) = (vx.data[i], vy.data[i]); *px = [(0.5 + a * 0.5).clamp(0.0, 1.0), (0.5 + c * 0.5).clamp(0.0, 1.0), 0.5, 1.0]; }); return b; } let src = b.img.clone(); let dt = amount as f32 / steps as f32; let (wf, hf) = (w as f64, h as f64); b.img.rows_mut().for_each(|(y, row)| { for (x, px) in row.iter_mut().enumerate() { let (mut sx, mut sy) = (x as f64 + 0.5, y as f64 + 0.5); for _ in 0..steps { // Trace the streamline backwards (semi-Lagrangian advection). let (u, v) = (vx.sample(sx, sy), vy.sample(sx, sy)); sx -= (u * dt) as f64; sy -= (v * dt) as f64; } *px = if edge_wrap { src.sample_bilinear_clamped(sx.rem_euclid(wf), sy.rem_euclid(hf)) } else { src.sample_bilinear_clamped(sx, sy) }; } }); b } pub fn specs() -> Vec { let (tweaking, color, application, animation) = grain_params(); vec![ spec( "ec.noise.matchgrain", "Match Grain", vec![ p("viewingMode", "Viewing Mode", Value::Enum(4), popup(&MATCH_GRAIN_VIEWS)), p("noiseSourceLayer", "Noise Source Layer", Value::Layer(None), ParamUi::Layer), p("noiseSourceLayerSource", "Noise Source Layer Source", Value::Enum(0), ParamUi::Hidden), p("compensateForExistingNoise", "Compensate for Existing Noise", num(0.0), pct()), ] .into_iter() .chain([preview_params(), sampling_params(), tweaking, color, application, animation].into_iter().flatten()) .collect(), match_grain, ), spec( "ec.noise.medianlegacy", "Median (Legacy)", vec![ p("radius", "Radius", num(1.0), slider(0.0, 255.0, 0.0, 20.0, 0)), p("operateOnAlphaChannel", "Operate On Alpha Channel", Value::Bool(false), ParamUi::Checkbox), ], median_legacy, ), spec( "ec.noise.curlnoise", "Curl Noise", vec![ p("view", "View", Value::Enum(0), popup(&["Final Output", "Flow Field"])), p("displacementAmount", "Displacement Amount", num(20.0), slider(-1000.0, 1000.0, -100.0, 100.0, 1)), p("scale", "Scale", num(100.0), slider(1.0, 10000.0, 10.0, 600.0, 1)), p("rotation", "Rotation", num(0.0), ParamUi::Angle), p("offset", "Offset Turbulence", Value::Vec2([0.5, 0.5]), ParamUi::Point), p("complexity", "Complexity", num(3.0), slider(1.0, 20.0, 1.0, 10.0, 1)), p("evolution", "Evolution", num(0.0), ParamUi::Angle), p("steps", "Flow Steps", num(4.0), slider(1.0, 16.0, 1.0, 16.0, 0)), p("edgeBehavior", "Edge Behavior", Value::Enum(0), popup(&["Clamp", "Wrap"])), p("randomSeed", "Random Seed", num(0.0), slider(0.0, 100000.0, 0.0, 1000.0, 0)), ], curl_noise, ), ] } #[cfg(test)] mod tests { use super::*; use crate::channel2::tests::FakeHost; use crate::{EffectEnv, run_fx}; fn noisy(w: u32, h: u32, amp: f32, seed: u32) -> Image { let mut img = Image::new(w, h); for y in 0..h { for x in 0..w { let n = (crate::util::hash1(x, y, seed) - 0.5) * 2.0 * amp; img.set(x, y, [0.5 + n, 0.5 + n, 0.5 + n, 1.0]); } } img } #[test] fn match_grain_transfers_grain_strength() { let flat = Image::filled(48, 48, [0.5, 0.5, 0.5, 1.0]); let none = run_fx("ec.noise.matchgrain", &[], flat.clone(), 0.0, EffectEnv::default()); assert_eq!(none.img.data, flat.data); let host = FakeHost(noisy(48, 48, 0.1, 3)); let env = EffectEnv { host: Some(&host), ..Default::default() }; let out = run_fx("ec.noise.matchgrain", &[("noiseSourceLayer", Value::Layer(Some(7)))], flat.clone(), 0.5, env); let target = grain_stats(&host.0).std[0]; let got = grain_stats(&out.img).std[0]; assert!(got > target * 0.4 && got < target * 2.5, "target {target} got {got}"); let again = run_fx("ec.noise.matchgrain", &[("noiseSourceLayer", Value::Layer(Some(7)))], flat, 0.5, env); assert_eq!(out.img.data, again.img.data); } #[test] fn median_legacy_radius_zero_identity_and_removes_speck() { let mut img = Image::filled(9, 9, [0.2, 0.2, 0.2, 1.0]); img.set(4, 4, [1.0, 1.0, 1.0, 1.0]); let id = run_fx("ec.noise.medianlegacy", &[("radius", num(0.0))], img.clone(), 0.0, EffectEnv::default()); assert_eq!(id.img.data, img.data); let out = run_fx("ec.noise.medianlegacy", &[("radius", num(1.0))], img.clone(), 0.0, EffectEnv::default()); assert!((out.img.get(4, 4)[0] - 0.2).abs() < 0.01); let again = run_fx("ec.noise.medianlegacy", &[("radius", num(1.0))], img, 0.0, EffectEnv::default()); assert_eq!(out.img.data, again.img.data); } #[test] fn curl_noise_zero_is_identity_and_flow_preserves_flat() { let img = crate::channel2::tests::sample(); let id = run_fx("ec.noise.curlnoise", &[("displacementAmount", num(0.0))], img.clone(), 0.0, EffectEnv::default()); assert_eq!(id.img.data, img.data); let flat = Image::filled(16, 16, [0.3, 0.6, 0.9, 1.0]); let out = run_fx("ec.noise.curlnoise", &[("displacementAmount", num(40.0)), ("scale", num(10.0))], flat.clone(), 0.0, EffectEnv::default()); assert!(out.img.data.iter().all(|p| (p[1] - 0.6).abs() < 1e-5)); let mut grad = Image::new(32, 32); for y in 0..32 { for x in 0..32 { grad.set(x, y, [x as f32 / 32.0, y as f32 / 32.0, 0.0, 1.0]); } } let a = run_fx("ec.noise.curlnoise", &[("displacementAmount", num(40.0)), ("scale", num(10.0))], grad.clone(), 0.0, EffectEnv::default()); let b = run_fx("ec.noise.curlnoise", &[("displacementAmount", num(40.0)), ("scale", num(10.0))], grad.clone(), 0.0, EffectEnv::default()); assert_eq!(a.img.data, b.img.data); assert!(a.img.data.iter().zip(grad.data.iter()).any(|(p, q)| (p[0] - q[0]).abs() > 0.01)); } }