feat: add reference images to local edit regions

app/api/canvas/local-edit/route.ts

@@ -30,6 +30,12 @@ type LocalEditRegionInput = {
   color?: string;
   prompt?: string;
   rect?: LocalEditRect;
+  referenceImages?: Array<{
+    id?: string;
+    url?: string;
+    name?: string;
+    source?: string;
+  }>;
 };
 
 type LocalEditRequest = {
@@ -47,6 +53,12 @@ type NormalizedLocalEditRegion = {
   color: string;
   colorName: string;
   prompt: string;
+  referenceImages: Array<{
+    id: string;
+    url: string;
+    name?: string;
+    source: "upload" | "canvas";
+  }>;
   rect: {
     x: number;
     y: number;
@@ -62,6 +74,7 @@ type NormalizedLocalEditRegion = {
 };
 
 const MAX_LOCAL_EDIT_REGIONS = 8;
+const MAX_REFERENCE_IMAGES_PER_REGION = 3;
 const MIN_REGION_EDGE_PX = 24;
 const DEFAULT_REGION_COLORS = [
   "#ff3b30",
@@ -100,6 +113,41 @@ function normalizeHexColor(input: unknown, fallback: string) {
   return /^#[0-9a-f]{6}$/i.test(expanded) ? expanded.toLowerCase() : fallback;
 }
 
+function normalizeHttpImageUrl(input: unknown) {
+  const raw = String(input || "").trim();
+  if (!/^https?:\/\//i.test(raw)) return "";
+  return raw;
+}
+
+function normalizeRegionReferenceImages(
+  region: LocalEditRegionInput,
+  itemIndex: number
+): NormalizedLocalEditRegion["referenceImages"] {
+  if (!Array.isArray(region.referenceImages)) return [];
+  return region.referenceImages
+    .map((reference, referenceIndex) => {
+      const url = normalizeHttpImageUrl(reference?.url);
+      if (!url) return null;
+      const name =
+        typeof reference?.name === "string" && reference.name.trim()
+          ? reference.name.trim().slice(0, 80)
+          : "";
+      return {
+        id:
+          typeof reference?.id === "string" && reference.id.trim()
+            ? reference.id.trim()
+            : `region-${itemIndex + 1}-reference-${referenceIndex + 1}`,
+        url,
+        ...(name ? { name } : {}),
+        source: reference?.source === "canvas" ? "canvas" as const : "upload" as const,
+      };
+    })
+    .filter((item): item is NormalizedLocalEditRegion["referenceImages"][number] =>
+      Boolean(item)
+    )
+    .slice(0, MAX_REFERENCE_IMAGES_PER_REGION);
+}
+
 function getColorName(hex: string) {
   const normalized = hex.toLowerCase();
   const names: Record<string, string> = {
@@ -203,6 +251,7 @@ function normalizeRegions(
       color,
       colorName: getColorName(color),
       prompt: prompt.slice(0, 300),
+      referenceImages: normalizeRegionReferenceImages(region, itemIndex),
       rect,
       pixelRect,
     };
@@ -278,18 +327,32 @@ function buildLocalEditPrompt(regions: NormalizedLocalEditRegion[]) {
       const top = Math.round(region.rect.y * 1000) / 10;
       const width = Math.round(region.rect.width * 1000) / 10;
       const height = Math.round(region.rect.height * 1000) / 10;
-      return `区域 ${region.index}（${region.colorName}，编号 ${region.index}，原图位置 x=${left}%，y=${top}%，w=${width}%，h=${height}%）：${region.prompt}`;
+      const referenceNote =
+        region.referenceImages.length > 0
+          ? `；该区域有 ${region.referenceImages.length} 张专属参考图，依次为后续输入图片中的“区域 ${region.index} 参考图”，只可用于该区域`
+          : "";
+      return `区域 ${region.index}（${region.colorName}，编号 ${region.index}，原图位置 x=${left}%，y=${top}%，w=${width}%，h=${height}%${referenceNote}）：${region.prompt}`;
     })
     .join("\n");
+  const referenceLines = regions
+    .flatMap((region) =>
+      region.referenceImages.map(
+        (reference, referenceIndex) =>
+          `区域 ${region.index} 参考图 ${referenceIndex + 1}：${reference.name || "未命名参考图"}，只能影响区域 ${region.index}`
+      )
+    )
+    .join("\n");
 
   return [
     "输入图片1是原始干净图，也是最终输出必须继承的唯一画面基底。",
     "输入图片2是同一张图的彩色标注说明图，只用于理解编辑区域位置和区域提示词，不是画面参考。",
+    "输入图片3及之后如存在，是各区域专属参考图；专属参考图只服务于对应区域，不得改变其它区域或整张图的主体身份。",
     "请不要复刻、保留、描摹或生成输入图片2里的任何彩色框、编号圆点、提示词气泡、边框线、半透明色块或标注文字。",
     "如果输入图片2的标注元素与输入图片1冲突，必须完全忽略标注元素，只保留输入图片1的真实画面内容。",
     "",
     "请严格根据每个区域下方的提示词，只在对应彩色框附近完成局部编辑：",
     regionLines,
+    referenceLines ? `\n区域专属参考图说明：\n${referenceLines}` : "",
     "",
     "最终输出必须是干净照片：不得出现任何彩色框、蓝框、绿框、红框、矩形边框、编号、提示词气泡、标注文字或 UI 元素。",
     "未被彩色框标注的区域必须尽量保持原图不变。",
@@ -368,6 +431,9 @@ export async function POST(request: NextRequest) {
       toArrayBuffer(annotatedBuffer)
     );
     const prompt = buildLocalEditPrompt(regions);
+    const regionReferenceImageUrls = regions.flatMap((region) =>
+      region.referenceImages.map((reference) => reference.url)
+    );
 
     const imageRequest: CanvasGenerateImageRequest = {
       aspectRatio:
@@ -378,6 +444,7 @@ export async function POST(request: NextRequest) {
       debugLabel: "canvas-local-edit",
       debugMeta: {
         annotatedReferenceUrl: annotatedUpload.url,
+        regionReferenceImageCount: regionReferenceImageUrls.length,
         regionCount: regions.length,
         sourceNodeId: body.sourceNodeId || null,
       },
@@ -394,7 +461,11 @@ export async function POST(request: NextRequest) {
         "彩色框、蓝色边框、绿色边框、红色边框、矩形框、编号圆点、提示词气泡、标注文字、UI 控件、半透明色块、选区边框",
       prompt,
       promptExtend: false,
-      referenceImages: [sourceImageUrl, annotatedUpload.url],
+      referenceImages: [
+        sourceImageUrl,
+        annotatedUpload.url,
+        ...regionReferenceImageUrls,
+      ],
       ownerKey: session.user.id,
     };
 

components/canvas/nodes/ImageNode.tsx

@@ -248,12 +248,21 @@ interface BoxCutoutPoint {
   y: number;
 }
 
+interface LocalEditReferenceImage {
+  id: string;
+  url: string;
+  name?: string;
+  source?: "upload" | "canvas";
+  createdAt: number;
+}
+
 interface LocalEditRegion {
   id: string;
   index: number;
   color: string;
   prompt: string;
   rect: NormalizedCropRect;
+  referenceImages?: LocalEditReferenceImage[];
   createdAt: number;
 }
 
@@ -261,6 +270,7 @@ const BOX_CUTOUT_DEFAULT_PADDING_RATIO = 0.08;
 const BOX_CUTOUT_MIN_EDGE_PX = 24;
 const LOCAL_EDIT_METADATA_KEY = "localEditRegions";
 const LOCAL_EDIT_MAX_REGIONS = 8;
+const LOCAL_EDIT_MAX_REFERENCE_IMAGES_PER_REGION = 3;
 const LOCAL_EDIT_MIN_EDGE_PX = BOX_CUTOUT_MIN_EDGE_PX;
 const LOCAL_EDIT_POLL_INTERVAL_MS = 2500;
 const LOCAL_EDIT_MAX_POLL_ATTEMPTS = 180;
@@ -875,6 +885,8 @@ const normalizeTextEditItems = (
 
 const createLocalEditRegionId = () =>
   `local-edit-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
+const createLocalEditReferenceImageId = () =>
+  `local-edit-ref-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
 
 const normalizeLocalEditRect = (value: unknown): NormalizedCropRect | null => {
   if (!value || typeof value !== "object") return null;
@@ -896,6 +908,40 @@ const normalizeLocalEditRect = (value: unknown): NormalizedCropRect | null => {
   return { x, y, width, height };
 };
 
+const normalizeLocalEditReferenceImages = (
+  value: unknown
+): LocalEditReferenceImage[] => {
+  if (!Array.isArray(value)) return [];
+  const references: LocalEditReferenceImage[] = [];
+  for (const item of value) {
+    if (!item || typeof item !== "object") continue;
+    const record = item as Record<string, unknown>;
+    const url = typeof record.url === "string" ? record.url.trim() : "";
+    if (!isHttpImageUrl(url)) continue;
+    const name =
+      typeof record.name === "string" && record.name.trim()
+        ? record.name.trim().slice(0, 80)
+        : "";
+    references.push({
+      id:
+        typeof record.id === "string" && record.id.trim()
+          ? record.id.trim()
+          : createLocalEditReferenceImageId(),
+      url,
+      ...(name ? { name } : {}),
+      source: record.source === "canvas" ? "canvas" : "upload",
+      createdAt:
+        typeof record.createdAt === "number" && Number.isFinite(record.createdAt)
+          ? record.createdAt
+          : Date.now(),
+    });
+    if (references.length >= LOCAL_EDIT_MAX_REFERENCE_IMAGES_PER_REGION) {
+      break;
+    }
+  }
+  return references;
+};
+
 const reindexLocalEditRegions = (
   regions: LocalEditRegion[]
 ): LocalEditRegion[] =>
@@ -907,38 +953,36 @@ const reindexLocalEditRegions = (
 
 const normalizeLocalEditRegions = (value: unknown): LocalEditRegion[] => {
   if (!Array.isArray(value)) return [];
-  return reindexLocalEditRegions(
+  const regions: LocalEditRegion[] = [];
-    value
+  value.forEach((item, index) => {
-      .map((item, index) => {
+    if (!item || typeof item !== "object") return;
-        if (!item || typeof item !== "object") return null;
+    const record = item as Record<string, unknown>;
-        const record = item as Record<string, unknown>;
+    const rect = normalizeLocalEditRect(record.rect);
-        const rect = normalizeLocalEditRect(record.rect);
+    if (!rect) return;
-        if (!rect) return null;
+    regions.push({
-        return {
+      id:
-          id:
+        typeof record.id === "string" && record.id.trim()
-            typeof record.id === "string" && record.id.trim()
+          ? record.id.trim()
-              ? record.id.trim()
+          : createLocalEditRegionId(),
-              : createLocalEditRegionId(),
+      index:
-          index:
+        typeof record.index === "number" && Number.isFinite(record.index)
-            typeof record.index === "number" && Number.isFinite(record.index)
+          ? record.index
-              ? record.index
+          : index + 1,
-              : index + 1,
+      color:
-          color:
+        typeof record.color === "string" && record.color.trim()
-            typeof record.color === "string" && record.color.trim()
+          ? record.color.trim()
-              ? record.color.trim()
+          : LOCAL_EDIT_COLORS[index % LOCAL_EDIT_COLORS.length],
-              : LOCAL_EDIT_COLORS[index % LOCAL_EDIT_COLORS.length],
+      prompt:
-          prompt:
+        typeof record.prompt === "string" ? record.prompt.trim() : "",
-            typeof record.prompt === "string" ? record.prompt.trim() : "",
+      rect,
-          rect,
+      referenceImages: normalizeLocalEditReferenceImages(record.referenceImages),
-          createdAt:
+      createdAt:
-            typeof record.createdAt === "number" && Number.isFinite(record.createdAt)
+        typeof record.createdAt === "number" && Number.isFinite(record.createdAt)
-              ? record.createdAt
+          ? record.createdAt
-              : Date.now(),
+          : Date.now(),
-        } satisfies LocalEditRegion;
+    });
-      })
+  });
-      .filter((item): item is LocalEditRegion => !!item)
+  return reindexLocalEditRegions(regions.slice(0, LOCAL_EDIT_MAX_REGIONS));
-      .slice(0, LOCAL_EDIT_MAX_REGIONS)
-  );
 };
 
 const getLocalEditRegionsFromMetadata = (
@@ -1838,6 +1882,10 @@ export function ImageNode({
   const [editingLocalEditRegionId, setEditingLocalEditRegionId] =
     useState<string | null>(null);
   const [isSubmittingLocalEdit, setIsSubmittingLocalEdit] = useState(false);
+  const [
+    uploadingLocalEditReferenceRegionId,
+    setUploadingLocalEditReferenceRegionId,
+  ] = useState<string | null>(null);
   const anglePreviewDragRef = useRef<AnglePreviewDragState | null>(null);
   const activeImageAdjustments = imageAdjustPanelOpen
     ? imageAdjustmentsDraft
@@ -2384,9 +2432,13 @@ export function ImageNode({
       const width = Number(style.width || 0);
       const height = Number(style.height || 0);
       const imageBottom = boxCutoutMetrics.offsetY + boxCutoutMetrics.renderedHeight;
-      const placeAbove = top + height + 58 > imageBottom;
+      const referenceCount = region.referenceImages?.length || 0;
-      const promptTop = placeAbove ? top - 46 : top + height + 8;
+      const promptStackHeight = referenceCount > 0 ? 82 : 42;
-      const promptWidth = Math.min(Math.max(width, 176), 300);
+      const placeAbove = top + height + promptStackHeight + 14 > imageBottom;
+      const promptTop = placeAbove
+        ? top - promptStackHeight - 8
+        : top + height + 8;
+      const promptWidth = Math.min(Math.max(width, 236), 360);
 
       return {
         region,
@@ -2467,6 +2519,7 @@ export function ImageNode({
     setLocalEditDragEnd(null);
     setLocalEditRegionInteraction(null);
     setEditingLocalEditRegionId(null);
+    setUploadingLocalEditReferenceRegionId(null);
   }, []);
 
   const focusLocalEditPromptInput = useCallback((regionId: string) => {
@@ -3402,6 +3455,7 @@ export function ImageNode({
             LOCAL_EDIT_COLORS[localEditRegions.length % LOCAL_EDIT_COLORS.length],
           prompt: "",
           rect: nextRect,
+          referenceImages: [],
           createdAt: Date.now(),
         };
         commitLocalEditRegions([...localEditRegions, nextRegion]);
@@ -3618,6 +3672,95 @@ export function ImageNode({
     [commitLocalEditRegions, editingLocalEditRegionId]
   );
 
+  const handleLocalEditReferenceUpload = useCallback(
+    async (regionId: string, fileList: FileList | null) => {
+      const files = Array.from(fileList || []).filter((file) =>
+        file.type.startsWith("image/")
+      );
+      if (files.length === 0) return;
+
+      const currentRegion = localEditRegionsRef.current.find(
+        (region) => region.id === regionId
+      );
+      if (!currentRegion) return;
+
+      const currentReferences = currentRegion.referenceImages || [];
+      const remainingSlots =
+        LOCAL_EDIT_MAX_REFERENCE_IMAGES_PER_REGION - currentReferences.length;
+      if (remainingSlots <= 0) {
+        toast.info(
+          `每个局部编辑区域最多添加 ${LOCAL_EDIT_MAX_REFERENCE_IMAGES_PER_REGION} 张参考图`
+        );
+        return;
+      }
+
+      const selectedFiles = files.slice(0, remainingSlots);
+      if (files.length > selectedFiles.length) {
+        toast.info(
+          `已自动保留前 ${LOCAL_EDIT_MAX_REFERENCE_IMAGES_PER_REGION} 张参考图`
+        );
+      }
+
+      setUploadingLocalEditReferenceRegionId(regionId);
+      try {
+        const uploadedReferences = await Promise.all(
+          selectedFiles.map(async (file) => {
+            const uploaded = await uploadChatAttachmentBlob(file, file.name);
+            return {
+              id: createLocalEditReferenceImageId(),
+              url: uploaded.url,
+              name: file.name,
+              source: "upload" as const,
+              createdAt: Date.now(),
+            } satisfies LocalEditReferenceImage;
+          })
+        );
+
+        commitLocalEditRegions(
+          localEditRegionsRef.current.map((region) =>
+            region.id === regionId
+              ? {
+                  ...region,
+                  referenceImages: [
+                    ...(region.referenceImages || []),
+                    ...uploadedReferences,
+                  ].slice(0, LOCAL_EDIT_MAX_REFERENCE_IMAGES_PER_REGION),
+                }
+              : region
+          )
+        );
+        toast.success(
+          uploadedReferences.length > 1
+            ? `已添加 ${uploadedReferences.length} 张区域参考图`
+            : "已添加区域参考图"
+        );
+      } catch (error) {
+        toast.error(error instanceof Error ? error.message : "参考图上传失败");
+      } finally {
+        setUploadingLocalEditReferenceRegionId(null);
+      }
+    },
+    [commitLocalEditRegions]
+  );
+
+  const removeLocalEditReferenceImage = useCallback(
+    (regionId: string, referenceId: string) => {
+      commitLocalEditRegions(
+        localEditRegionsRef.current.map((region) =>
+          region.id === regionId
+            ? {
+                ...region,
+                referenceImages: (region.referenceImages || []).filter(
+                  (reference) => reference.id !== referenceId
+                ),
+              }
+            : region
+        )
+      );
+    },
+    [commitLocalEditRegions]
+  );
+
   const clearLocalEditRegions = useCallback(() => {
     localEditPromptDraftRefs.current = {};
     commitLocalEditRegions([]);
@@ -3696,6 +3839,12 @@ export function ImageNode({
           color: region.color,
           prompt: region.prompt,
           rect: region.rect,
+          referenceImages: (region.referenceImages || []).map((reference) => ({
+            id: reference.id,
+            url: reference.url,
+            name: reference.name,
+            source: reference.source,
+          })),
         })),
       });
 
@@ -8366,7 +8515,7 @@ export function ImageNode({
         localEditPromptPortalOverlays.map(({ region, isEditing, style }) =>
           createPortal(
             <div
-              className="nodrag nopan nowheel fixed z-[10000] flex items-center gap-1.5"
+              className="nodrag nopan nowheel fixed z-[10000] flex flex-col items-stretch gap-1.5"
               data-canvas-no-drag="true"
               data-local-edit-control="true"
               key={region.id}
@@ -8378,110 +8527,190 @@ export function ImageNode({
               onPointerDownCapture={(event) => event.stopPropagation()}
               style={style}
             >
-              {isEditing ? (
+              <div className="flex items-center gap-1.5">
-                <input
+                {isEditing ? (
-                  autoFocus={editingLocalEditRegionId === region.id}
+                  <input
-                  className="nodrag nopan nowheel h-9 min-w-0 flex-1 rounded-full border border-slate-300 bg-white/95 px-3 text-[12px] font-medium text-slate-950 caret-slate-950 shadow-[0_10px_24px_rgba(15,23,42,0.14)] outline-none ring-2 ring-transparent transition placeholder:text-slate-400 focus:border-slate-950 focus:ring-slate-950/15 dark:border-white/20 dark:bg-slate-950/88 dark:text-white dark:caret-white dark:placeholder:text-white/42 dark:focus:border-white/40 dark:focus:ring-white/18"
+                    autoFocus={editingLocalEditRegionId === region.id}
-                  data-canvas-no-drag="true"
+                    className="nodrag nopan nowheel h-9 min-w-0 flex-1 rounded-full border border-slate-300 bg-white/95 px-3 text-[12px] font-medium text-slate-950 caret-slate-950 shadow-[0_10px_24px_rgba(15,23,42,0.14)] outline-none ring-2 ring-transparent transition placeholder:text-slate-400 focus:border-slate-950 focus:ring-slate-950/15 dark:border-white/20 dark:bg-slate-950/88 dark:text-white dark:caret-white dark:placeholder:text-white/42 dark:focus:border-white/40 dark:focus:ring-white/18"
-                  data-local-edit-control="true"
+                    data-canvas-no-drag="true"
-                  defaultValue={
+                    data-local-edit-control="true"
-                    localEditPromptDraftRefs.current[region.id] ?? region.prompt
+                    defaultValue={
-                  }
+                      localEditPromptDraftRefs.current[region.id] ?? region.prompt
-                  onBlur={(event) => {
-                    const nextPrompt = event.currentTarget.value;
-                    localEditPromptDraftRefs.current[region.id] = nextPrompt;
-                    persistLocalEditRegionPrompt(region.id, nextPrompt);
-                    if (isLocalEditControlTarget(event.relatedTarget)) {
-                      return;
                     }
-                    window.setTimeout(() => {
+                    onBlur={(event) => {
-                      if (isLocalEditControlTarget(document.activeElement)) {
+                      const nextPrompt = event.currentTarget.value;
+                      localEditPromptDraftRefs.current[region.id] = nextPrompt;
+                      persistLocalEditRegionPrompt(region.id, nextPrompt);
+                      if (isLocalEditControlTarget(event.relatedTarget)) {
                         return;
                       }
-                      setEditingLocalEditRegionId((current) => {
+                      window.setTimeout(() => {
-                        if (current === region.id) {
+                        if (isLocalEditControlTarget(document.activeElement)) {
-                          focusLocalEditPromptInput(region.id);
+                          return;
                         }
-                        return current;
+                        setEditingLocalEditRegionId((current) => {
-                      });
+                          if (current === region.id) {
-                    }, 80);
+                            focusLocalEditPromptInput(region.id);
-                  }}
+                          }
-                  onClick={(event) => {
+                          return current;
-                    event.nativeEvent.stopImmediatePropagation?.();
+                        });
-                    event.stopPropagation();
+                      }, 80);
-                    event.currentTarget.focus({ preventScroll: true });
+                    }}
-                  }}
+                    onClick={(event) => {
-                  onFocus={(event) => {
+                      event.nativeEvent.stopImmediatePropagation?.();
-                    event.stopPropagation();
+                      event.stopPropagation();
-                    setEditingLocalEditRegionId((current) =>
+                      event.currentTarget.focus({ preventScroll: true });
-                      current === region.id ? current : region.id
+                    }}
-                    );
+                    onFocus={(event) => {
-                  }}
+                      event.stopPropagation();
-                  onInput={(event) => {
+                      setEditingLocalEditRegionId((current) =>
-                    localEditPromptDraftRefs.current[region.id] =
+                        current === region.id ? current : region.id
-                      event.currentTarget.value;
+                      );
-                  }}
+                    }}
+                    onInput={(event) => {
+                      localEditPromptDraftRefs.current[region.id] =
+                        event.currentTarget.value;
+                    }}
+                    onChange={(event) => {
+                      localEditPromptDraftRefs.current[region.id] =
+                        event.currentTarget.value;
+                    }}
+                    onKeyDown={(event) => {
+                      event.nativeEvent.stopImmediatePropagation?.();
+                      event.stopPropagation();
+                      if (event.key === "Enter") {
+                        event.preventDefault();
+                        persistLocalEditRegionPrompt(
+                          region.id,
+                          event.currentTarget.value
+                        );
+                        setEditingLocalEditRegionId(null);
+                      }
+                      if (event.key === "Escape") {
+                        event.preventDefault();
+                        persistLocalEditRegionPrompt(
+                          region.id,
+                          event.currentTarget.value
+                        );
+                        setEditingLocalEditRegionId(null);
+                      }
+                    }}
+                    onKeyUp={(event) => {
+                      event.nativeEvent.stopImmediatePropagation?.();
+                      event.stopPropagation();
+                    }}
+                    onMouseDown={(event) => {
+                      event.nativeEvent.stopImmediatePropagation?.();
+                      event.stopPropagation();
+                    }}
+                    onMouseDownCapture={(event) => {
+                      event.nativeEvent.stopImmediatePropagation?.();
+                      event.stopPropagation();
+                      event.currentTarget.focus({ preventScroll: true });
+                    }}
+                    onPointerDown={(event) => {
+                      event.nativeEvent.stopImmediatePropagation?.();
+                      event.stopPropagation();
+                    }}
+                    onPointerDownCapture={(event) => {
+                      event.nativeEvent.stopImmediatePropagation?.();
+                      event.stopPropagation();
+                      event.currentTarget.focus({ preventScroll: true });
+                    }}
+                    placeholder={`区域 ${region.index} 提示词`}
+                    ref={(input) => {
+                      localEditPromptInputRefs.current[region.id] = input;
+                    }}
+                  />
+                ) : (
+                  <button
+                    className="nodrag nopan nowheel h-9 min-w-0 flex-1 truncate rounded-full border border-slate-200 bg-white/95 px-3 text-left text-[12px] font-semibold text-slate-900 shadow-[0_10px_24px_rgba(15,23,42,0.14)] hover:bg-slate-50 dark:border-white/18 dark:bg-slate-950/82 dark:text-white dark:shadow-[0_10px_24px_rgba(0,0,0,0.24)] dark:hover:bg-slate-900"
+                    data-canvas-no-drag="true"
+                    data-local-edit-control="true"
+                    onClick={(event) => {
+                      event.preventDefault();
+                      event.nativeEvent.stopImmediatePropagation?.();
+                      event.stopPropagation();
+                      setEditingLocalEditRegionId(region.id);
+                    }}
+                    onMouseDown={(event) => {
+                      event.preventDefault();
+                      event.nativeEvent.stopImmediatePropagation?.();
+                      event.stopPropagation();
+                    }}
+                    onPointerDown={(event) => {
+                      event.preventDefault();
+                      event.nativeEvent.stopImmediatePropagation?.();
+                      event.stopPropagation();
+                      setEditingLocalEditRegionId(region.id);
+                    }}
+                    title={region.prompt}
+                    type="button"
+                  >
+                    {region.index}. {region.prompt}
+                  </button>
+                )}
+                <input
+                  accept="image/jpeg,image/png,image/webp"
+                  className="hidden"
+                  data-canvas-no-drag="true"
+                  data-local-edit-control="true"
+                  id={`local-edit-reference-upload-${region.id}`}
+                  multiple
                   onChange={(event) => {
-                    localEditPromptDraftRefs.current[region.id] =
+                    void handleLocalEditReferenceUpload(
-                      event.currentTarget.value;
+                      region.id,
-                  }}
+                      event.currentTarget.files
-                  onKeyDown={(event) => {
+                    );
-                    event.nativeEvent.stopImmediatePropagation?.();
+                    event.currentTarget.value = "";
-                    event.stopPropagation();
-                    if (event.key === "Enter") {
-                      event.preventDefault();
-                      persistLocalEditRegionPrompt(
-                        region.id,
-                        event.currentTarget.value
-                      );
-                      setEditingLocalEditRegionId(null);
-                    }
-                    if (event.key === "Escape") {
-                      event.preventDefault();
-                      persistLocalEditRegionPrompt(
-                        region.id,
-                        event.currentTarget.value
-                      );
-                      setEditingLocalEditRegionId(null);
-                    }
-                  }}
-                  onKeyUp={(event) => {
-                    event.nativeEvent.stopImmediatePropagation?.();
-                    event.stopPropagation();
-                  }}
-                  onMouseDown={(event) => {
-                    event.nativeEvent.stopImmediatePropagation?.();
-                    event.stopPropagation();
-                  }}
-                  onMouseDownCapture={(event) => {
-                    event.nativeEvent.stopImmediatePropagation?.();
-                    event.stopPropagation();
-                    event.currentTarget.focus({ preventScroll: true });
-                  }}
-                  onPointerDown={(event) => {
-                    event.nativeEvent.stopImmediatePropagation?.();
-                    event.stopPropagation();
-                  }}
-                  onPointerDownCapture={(event) => {
-                    event.nativeEvent.stopImmediatePropagation?.();
-                    event.stopPropagation();
-                    event.currentTarget.focus({ preventScroll: true });
-                  }}
-                  placeholder={`区域 ${region.index} 提示词`}
-                  ref={(input) => {
-                    localEditPromptInputRefs.current[region.id] = input;
                   }}
+                  type="file"
                 />
-              ) : (
                 <button
-                  className="nodrag nopan nowheel h-9 min-w-0 flex-1 truncate rounded-full border border-slate-200 bg-white/95 px-3 text-left text-[12px] font-semibold text-slate-900 shadow-[0_10px_24px_rgba(15,23,42,0.14)] hover:bg-slate-50 dark:border-white/18 dark:bg-slate-950/82 dark:text-white dark:shadow-[0_10px_24px_rgba(0,0,0,0.24)] dark:hover:bg-slate-900"
+                  className="nodrag nopan nowheel flex h-8 w-8 shrink-0 items-center justify-center rounded-full border border-slate-200 bg-white/95 text-slate-600 shadow-[0_10px_24px_rgba(15,23,42,0.12)] hover:bg-slate-950 hover:text-white disabled:cursor-not-allowed disabled:opacity-50 dark:border-white/18 dark:bg-slate-950/82 dark:text-white/72 dark:shadow-[0_10px_24px_rgba(0,0,0,0.24)] dark:hover:bg-white dark:hover:text-slate-950"
                   data-canvas-no-drag="true"
                   data-local-edit-control="true"
                   onClick={(event) => {
                     event.preventDefault();
                     event.nativeEvent.stopImmediatePropagation?.();
                     event.stopPropagation();
-                    setEditingLocalEditRegionId(region.id);
+                    document
+                      .getElementById(`local-edit-reference-upload-${region.id}`)
+                      ?.click();
+                  }}
+                  disabled={
+                    uploadingLocalEditReferenceRegionId === region.id ||
+                    (region.referenceImages?.length || 0) >=
+                      LOCAL_EDIT_MAX_REFERENCE_IMAGES_PER_REGION
+                  }
+                  onMouseDown={(event) => {
+                    event.preventDefault();
+                    event.nativeEvent.stopImmediatePropagation?.();
+                    event.stopPropagation();
+                  }}
+                  onPointerDown={(event) => {
+                    event.preventDefault();
+                    event.nativeEvent.stopImmediatePropagation?.();
+                    event.stopPropagation();
+                  }}
+                  title="添加区域参考图"
+                  type="button"
+                >
+                  {uploadingLocalEditReferenceRegionId === region.id ? (
+                    <Loader2 className="h-3.5 w-3.5 animate-spin" />
+                  ) : (
+                    <ImagePlus className="h-3.5 w-3.5" />
+                  )}
+                </button>
+                <button
+                  className="nodrag nopan nowheel flex h-8 w-8 shrink-0 items-center justify-center rounded-full border border-slate-200 bg-white/95 text-slate-500 shadow-[0_10px_24px_rgba(15,23,42,0.12)] hover:bg-red-500 hover:text-white dark:border-white/18 dark:bg-slate-950/82 dark:text-white/72 dark:shadow-[0_10px_24px_rgba(0,0,0,0.24)]"
+                  data-canvas-no-drag="true"
+                  data-local-edit-control="true"
+                  onClick={(event) => {
+                    event.preventDefault();
+                    event.nativeEvent.stopImmediatePropagation?.();
+                    event.stopPropagation();
+                    deleteLocalEditRegion(region.id);
                   }}
                   onMouseDown={(event) => {
                     event.preventDefault();
@@ -8492,40 +8721,50 @@ export function ImageNode({
                     event.preventDefault();
                     event.nativeEvent.stopImmediatePropagation?.();
                     event.stopPropagation();
-                    setEditingLocalEditRegionId(region.id);
+                    deleteLocalEditRegion(region.id);
                   }}
-                  title={region.prompt}
+                  title="删除区域"
                   type="button"
                 >
-                  {region.index}. {region.prompt}
+                  <X className="h-3.5 w-3.5" />
                 </button>
-              )}
+              </div>
-              <button
+              {region.referenceImages?.length ? (
-                className="nodrag nopan nowheel flex h-8 w-8 shrink-0 items-center justify-center rounded-full border border-slate-200 bg-white/95 text-slate-500 shadow-[0_10px_24px_rgba(15,23,42,0.12)] hover:bg-red-500 hover:text-white dark:border-white/18 dark:bg-slate-950/82 dark:text-white/72 dark:shadow-[0_10px_24px_rgba(0,0,0,0.24)]"
+                <div className="flex max-w-full items-center gap-1 overflow-hidden rounded-2xl border border-slate-200 bg-white/92 px-1.5 py-1 shadow-[0_8px_20px_rgba(15,23,42,0.1)] dark:border-white/12 dark:bg-slate-950/80">
-                data-canvas-no-drag="true"
+                  <span className="shrink-0 px-1 text-[10px] font-semibold text-slate-500 dark:text-white/52">
-                data-local-edit-control="true"
+                    参考
-                onClick={(event) => {
+                  </span>
-                  event.preventDefault();
+                  {region.referenceImages.map((reference) => (
-                  event.nativeEvent.stopImmediatePropagation?.();
+                    <div
-                  event.stopPropagation();
+                      className="group relative h-8 w-8 shrink-0 overflow-hidden rounded-lg border border-slate-200 bg-slate-100 dark:border-white/12 dark:bg-slate-800"
-                  deleteLocalEditRegion(region.id);
+                      key={reference.id}
-                }}
+                      title={reference.name || "区域参考图"}
-                onMouseDown={(event) => {
+                    >
-                  event.preventDefault();
+                      <img
-                  event.nativeEvent.stopImmediatePropagation?.();
+                        alt={reference.name || "区域参考图"}
-                  event.stopPropagation();
+                        className="h-full w-full object-cover"
-                }}
+                        draggable={false}
-                onPointerDown={(event) => {
+                        src={reference.url}
-                  event.preventDefault();
+                      />
-                  event.nativeEvent.stopImmediatePropagation?.();
+                      <button
-                  event.stopPropagation();
+                        className="absolute inset-0 hidden items-center justify-center bg-black/52 text-white group-hover:flex"
-                  deleteLocalEditRegion(region.id);
+                        data-canvas-no-drag="true"
-                }}
+                        data-local-edit-control="true"
-                title="删除区域"
+                        onClick={(event) => {
-                type="button"
+                          event.preventDefault();
-              >
+                          event.nativeEvent.stopImmediatePropagation?.();
-                <X className="h-3.5 w-3.5" />
+                          event.stopPropagation();
-              </button>
+                          removeLocalEditReferenceImage(region.id, reference.id);
+                        }}
+                        title="移除参考图"
+                        type="button"
+                      >
+                        <X className="h-3.5 w-3.5" />
+                      </button>
+                    </div>
+                  ))}
+                </div>
+              ) : null}
             </div>,
             document.body
           )

services/canvas/index.ts

@@ -307,6 +307,12 @@ export const imageGenerationService = {
       color: string;
       prompt: string;
       rect: NormalizedCropRect;
+      referenceImages?: Array<{
+        id?: string;
+        url: string;
+        name?: string;
+        source?: "upload" | "canvas";
+      }>;
     }>;
     model?: string;
     imageSize?: string;