fix(heal): eliminate boundary seams by windowing inpaint clusters to 1:1 scale and smoothing feather transition

docker/frontend/shared/utils/heal.ts

@@ -25,9 +25,9 @@ import type { HealSpot } from '../types';
 export const HEAL_DEFAULT_R = 0.012;
 
 // How much of a spot is painted at full strength, as a fraction of its radius:
-// out to 95% the mask is solid to completely eliminate defects without leaving
-// residual dirt/shadows, and the last 5% is a smooth Hermite fade to blend into the photo.
-export const HEAL_FEATHER = 0.95;
+// out to 75% the mask is solid to completely eliminate defects, and the outer 25%
+// is a smooth Hermite fade (smoothstep) to blend invisibly into the surrounding photo.
+export const HEAL_FEATHER = 0.75;
 
 // The finest of the fill's blurs, as a multiple of the widest brush in the
 // stroke. The light of the rim has to reach the middle of the hole, so the

docker/frontend/src/engine/exportEngine.ts

@@ -1326,20 +1326,43 @@ function clusterHealSpots(
       }
     }
   }
+
+  // Maximum dimension of a cluster's bounding box so that the inpaint crop
+  // runs at ~1:1 native resolution (avoiding heavy downscale/upscale blur & edge seams)
+  const isMob = typeof navigator !== 'undefined' && /iPad|iPhone|iPod|Android|Mobile/i.test(navigator.userAgent || '');
+  const maxSpan = isMob ? 240 : 380;
   const visited = new Uint8Array(n);
   const clusters: { x: number; y: number; r: number }[][] = [];
+
   for (let i = 0; i < n; i++) {
     if (visited[i]) continue;
     visited[i] = 1;
     const cluster: { x: number; y: number; r: number }[] = [spots[i]];
+    let minX = spots[i].x * width;
+    let maxX = minX;
+    let minY = spots[i].y * height;
+    let maxY = minY;
     const queue = [i];
+
     while (queue.length > 0) {
-      const curr = queue.pop()!;
+      const curr = queue.shift()!;
       for (const neighbor of adj[curr]) {
         if (!visited[neighbor]) {
-          visited[neighbor] = 1;
-          cluster.push(spots[neighbor]);
-          queue.push(neighbor);
+          const nx = spots[neighbor].x * width;
+          const ny = spots[neighbor].y * height;
+          const newMinX = Math.min(minX, nx);
+          const newMaxX = Math.max(maxX, nx);
+          const newMinY = Math.min(minY, ny);
+          const newMaxY = Math.max(maxY, ny);
+          if (newMaxX - newMinX <= maxSpan && newMaxY - newMinY <= maxSpan) {
+            visited[neighbor] = 1;
+            cluster.push(spots[neighbor]);
+            minX = newMinX;
+            maxX = newMaxX;
+            minY = newMinY;
+            maxY = newMaxY;
+            queue.push(neighbor);
+          }
         }
       }
     }
@@ -1503,7 +1526,7 @@ function clusterHealSpots(
         const my1 = Math.max(...cYs) + cWidest;
         const margin = Math.max(
           16,
-          Math.min(64, Math.round(Math.max(cWidest * 1.5, 0.25 * Math.max(mx1 - mx0, my1 - my0))))
+          Math.min(64, Math.round(cWidest * 1.5))
         );
         const srcLeft = Math.max(boxLeft, Math.floor(mx0 - margin));
         const srcTop = Math.max(boxTop, Math.floor(my0 - margin));

docker/frontend/src/engine/inpaint.ts

@@ -253,7 +253,7 @@ export function fastLocalInpaint(
   const fG = new Float32Array(w * h);
   const fB = new Float32Array(w * h);
 
-  const bStep = Math.max(1, Math.floor(bLen / 32));
+  const bStep = bLen <= 96 ? 1 : Math.max(1, Math.floor(bLen / 64));
 
   for (let y = 0; y < h; y++) {
     const row = y * w;
@@ -329,7 +329,8 @@ export function fastLocalInpaint(
   }
   noiseVariance /= Math.max(1, count);
   // High-frequency noise standard deviation (adjacent difference / sqrt(2))
-  const noiseAmp = Math.min(5, Math.max(0.6, Math.sqrt(noiseVariance * 0.5)));
+  const rawAmp = Math.sqrt(noiseVariance * 0.5);
+  const noiseAmp = rawAmp < 0.3 ? 0 : Math.min(4, rawAmp);
 
   let seed = (w * 31 + h * 17 + holeCount) | 0;
   function prng(): number {

docker/recipescam-images.tar.gz.sha256

@@ -1 +1 @@
-11bf4442bb348f4ac3ed55da345805502351bcadddbe11a1edb2b819a864a611  recipescam-images.tar.gz
+ca55c46c0374f76a007f5309759bf25e07fcee625483a87fe96cce5c34a93645  recipescam-images.tar.gz