@goodandready/dsh-image-gen 0.11.11 → 0.11.12

package/lib/vision-ocr-helpers.js

@@ -0,0 +1,119 @@
+// vision-ocr-helpers.js — OCR text localization and seamless replacement (#180).
+
+/**
+ * Parses bounding box representations into normalized [x1, y1, x2, y2] (0-1000 scale).
+ *
+ * @param {Array<number>|string} raw
+ * @param {number} [imgWidth=1000]
+ * @param {number} [imgHeight=1000]
+ * @returns {[number, number, number, number]|null}
+ */
+export function normalizeBbox(raw, imgWidth = 1000, imgHeight = 1000) {
+  if (!raw) return null
+  let coords = []
+  if (Array.isArray(raw)) {
+    coords = raw.map(Number)
+  } else if (typeof raw === 'string') {
+    coords = raw
+      .replace(/[\[\]]/g, '')
+      .split(/[,;\s]+/)
+      .map(Number)
+      .filter((n) => !Number.isNaN(n))
+  }
+
+  if (coords.length < 4) return null
+
+  let [x1, y1, x2, y2] = coords
+  // If coordinates exceed 1000, normalize from pixel dimensions
+  if (x1 > 1000 || x2 > 1000 || y1 > 1000 || y2 > 1000) {
+    x1 = Math.round((x1 / imgWidth) * 1000)
+    x2 = Math.round((x2 / imgWidth) * 1000)
+    y1 = Math.round((y1 / imgHeight) * 1000)
+    y2 = Math.round((y2 / imgHeight) * 1000)
+  }
+
+  const minX = Math.max(0, Math.min(x1, x2))
+  const maxX = Math.min(1000, Math.max(x1, x2))
+  const minY = Math.max(0, Math.min(y1, y2))
+  const maxY = Math.min(1000, Math.max(y1, y2))
+
+  return [minX, minY, maxX, maxY]
+}
+
+/**
+ * Builds an SVG inpaint mask from detected text bounding boxes.
+ * Solid black background (#000000) with solid white rectangles (#FFFFFF) for text regions to inpaint.
+ *
+ * @param {number} width Image pixel width
+ * @param {number} height Image pixel height
+ * @param {Array<[number, number, number, number]>} bboxes Normalized [x1, y1, x2, y2]
+ * @param {number} [paddingPx=8]
+ * @returns {string} SVG mask document string
+ */
+export function buildTextInpaintMaskSvg(width, height, bboxes, paddingPx = 8) {
+  const w = Math.max(1, width || 1024)
+  const h = Math.max(1, height || 1024)
+
+  const rects = (bboxes || []).map((box) => {
+    const norm = normalizeBbox(box, w, h)
+    if (!norm) return ''
+    const [x1n, y1n, x2n, y2n] = norm
+    const pxX1 = Math.max(0, Math.round((x1n / 1000) * w) - paddingPx)
+    const pxY1 = Math.max(0, Math.round((y1n / 1000) * h) - paddingPx)
+    const pxX2 = Math.min(w, Math.round((x2n / 1000) * w) + paddingPx)
+    const pxY2 = Math.min(h, Math.round((y2n / 1000) * h) + paddingPx)
+    const rectW = Math.max(1, pxX2 - pxX1)
+    const rectH = Math.max(1, pxY2 - pxY1)
+
+    return `<rect x="${pxX1}" y="${pxY1}" width="${rectW}" height="${rectH}" fill="#FFFFFF" rx="4" />`
+  }).filter(Boolean).join('\n    ')
+
+  return `<?xml version="1.0" encoding="UTF-8"?>
+<svg xmlns="http://www.w3.org/2000/svg" width="${w}" height="${h}" viewBox="0 0 ${w} ${h}">
+  <rect width="${w}" height="${h}" fill="#000000" />
+  ${rects}
+</svg>`
+}
+
+/**
+ * Builds an SVG typography overlay with replacement strings placed at original coordinates.
+ *
+ * @param {number} width
+ * @param {number} height
+ * @param {Array<{ text: string, bbox: [number, number, number, number], color?: string, fontSize?: number }>} items
+ * @param {string} [fontFamily='system-ui, -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, sans-serif']
+ * @returns {string} SVG overlay document string
+ */
+export function buildTextOverlaySvg(width, height, items, fontFamily = 'sans-serif') {
+  const w = Math.max(1, width || 1024)
+  const h = Math.max(1, height || 1024)
+
+  const textNodes = (items || []).map((item) => {
+    const norm = normalizeBbox(item.bbox, w, h)
+    if (!norm || !item.text) return ''
+    const [x1n, y1n, x2n, y2n] = norm
+    const pxX1 = Math.round((x1n / 1000) * w)
+    const pxY1 = Math.round((y1n / 1000) * h)
+    const boxW = Math.max(1, Math.round(((x2n - x1n) / 1000) * w))
+    const boxH = Math.max(1, Math.round(((y2n - y1n) / 1000) * h))
+
+    const computedFontSize = item.fontSize || Math.max(12, Math.round(boxH * 0.72))
+    const color = item.color || '#FFFFFF'
+    const centerX = pxX1 + Math.round(boxW / 2)
+    const baselineY = pxY1 + Math.round(boxH * 0.75)
+
+    // Escape XML entities
+    const safeText = String(item.text)
+      .replace(/&/g, '&amp;')
+      .replace(/</g, '&lt;')
+      .replace(/>/g, '&gt;')
+      .replace(/"/g, '&quot;')
+
+    return `<text x="${centerX}" y="${baselineY}" text-anchor="middle" font-family="${fontFamily}" font-size="${computedFontSize}" font-weight="700" fill="${color}" filter="drop-shadow(0 2px 4px rgba(0,0,0,0.6))">${safeText}</text>`
+  }).filter(Boolean).join('\n  ')
+
+  return `<?xml version="1.0" encoding="UTF-8"?>
+<svg xmlns="http://www.w3.org/2000/svg" width="${w}" height="${h}" viewBox="0 0 ${w} ${h}">
+  ${textNodes}
+</svg>`
+}

package/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@goodandready/dsh-image-gen",
-  "version": "0.11.11",
-  "description": "Image generation for DeepSeek Harness: a generate_image tool with pluggable providers \u2014 the FAL queue, any OpenAI-compatible images API, or a ChatGPT/Grok subscription with no API key at all. The picture is shown inline in the conversation; the model receives either a link (works with any chat model) or the image itself (needs dsh-vision-bridge or a vision-capable model).",
+  "version": "0.11.12",
+  "description": "Image generation for DeepSeek Harness: a generate_image tool with pluggable providers — the FAL queue, any OpenAI-compatible images API, or a ChatGPT/Grok subscription with no API key at all. The picture is shown inline in the conversation; the model receives either a link (works with any chat model) or the image itself (needs dsh-vision-bridge or a vision-capable model).",
   "keywords": [
     "deepseek-harness",
     "dsh",