diff --git a/app/(chat)/api/chat/route.ts b/app/(chat)/api/chat/route.ts
index d8ca94ef2..6865d963c 100644
--- a/app/(chat)/api/chat/route.ts
+++ b/app/(chat)/api/chat/route.ts
@@ -7688,6 +7688,8 @@ export async function POST(request: Request) {
           ? presetModeConfig.sku.includeModelSlot
           : false;
       const replacementReferences = normalizeAgentModeReferences(agentModePayload);
+      const detailReplacementStrategy =
+        normalizeAgentModeDetailReplacementStrategy(agentModePayload);
       const isMatchingReplacementPayload =
         (isDetailReplacementPreset && agentModePayload?.modeType === "detail_replacement") ||
         (isSkuReplacementPreset && agentModePayload?.modeType === "sku_replacement");
diff --git a/components/canvas/nodes/ImageNode.tsx b/components/canvas/nodes/ImageNode.tsx
index d8a79e892..b637c96b2 100644
--- a/components/canvas/nodes/ImageNode.tsx
+++ b/components/canvas/nodes/ImageNode.tsx
@@ -1981,6 +1981,9 @@ export function ImageNode({
   const cropOverlayRef = useRef<HTMLDivElement>(null);
   const toolbarRef = useRef<HTMLDivElement>(null);
   const boxCutoutToolbarRef = useRef<HTMLDivElement>(null);
+  const localEditPromptInputRefs = useRef<
+    Record<string, HTMLInputElement | null>
+  >({});
 
   const updateNode = useCanvasStore((state) => state.updateNode);
   const deleteNode = useCanvasStore((state) => state.deleteNode);
@@ -2410,6 +2413,20 @@ export function ImageNode({
     setEditingLocalEditRegionId(null);
   }, []);
 
+  const focusLocalEditPromptInput = useCallback((regionId: string) => {
+    if (typeof window === "undefined") return;
+    window.requestAnimationFrame(() => {
+      const input = localEditPromptInputRefs.current[regionId];
+      if (!input) return;
+      input.focus({ preventScroll: true });
+    });
+  }, []);
+
+  useEffect(() => {
+    if (!localEditMode || !editingLocalEditRegionId) return;
+    focusLocalEditPromptInput(editingLocalEditRegionId);
+  }, [editingLocalEditRegionId, focusLocalEditPromptInput, localEditMode]);
+
   const commitLocalEditRegions = useCallback(
     (regions: LocalEditRegion[]) => {
       const nextRegions = reindexLocalEditRegions(
@@ -7528,8 +7545,11 @@ export function ImageNode({
                         className="absolute z-[175] flex items-center gap-1.5"
                         data-canvas-no-drag="true"
                         onClick={(event) => event.stopPropagation()}
+                        onClickCapture={(event) => event.stopPropagation()}
                         onMouseDown={(event) => event.stopPropagation()}
+                        onMouseDownCapture={(event) => event.stopPropagation()}
                         onPointerDown={(event) => event.stopPropagation()}
+                        onPointerDownCapture={(event) => event.stopPropagation()}
                         style={{
                           left,
                           top: Math.max(4, promptTop),
@@ -7539,17 +7559,29 @@ export function ImageNode({
                         {isEditing ? (
                           <input
                             autoFocus={editingLocalEditRegionId === region.id}
-                            className="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 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:placeholder:text-white/42 dark:focus:border-white/40 dark:focus:ring-white/18"
+                            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-canvas-no-drag="true"
                             onBlur={(event) => {
-                              const nextPrompt = event.currentTarget.value;
-                              persistLocalEditRegionPrompt(region.id, nextPrompt);
-                              if (nextPrompt.trim()) {
-                                setEditingLocalEditRegionId(null);
+                              const relatedTarget =
+                                event.relatedTarget instanceof HTMLElement
+                                  ? event.relatedTarget
+                                  : null;
+                              const nextPrompt = event.currentTarget.value.trim();
+                              if (
+                                !nextPrompt &&
+                                editingLocalEditRegionId === region.id &&
+                                !relatedTarget?.closest("[data-local-edit-control='true']")
+                              ) {
+                                focusLocalEditPromptInput(region.id);
                               }
                             }}
                             onChange={(event) =>
                               updateLocalEditRegionPrompt(region.id, event.target.value)
                             }
+                            onClick={(event) => {
+                              event.stopPropagation();
+                              focusLocalEditPromptInput(region.id);
+                            }}
                             onFocus={() => setEditingLocalEditRegionId(region.id)}
                             onKeyDown={(event) => {
                               event.stopPropagation();
@@ -7571,7 +7603,17 @@ export function ImageNode({
                               }
                             }}
                             onKeyUp={(event) => event.stopPropagation()}
+                            onMouseDown={(event) => event.stopPropagation()}
+                            onMouseDownCapture={(event) => event.stopPropagation()}
+                            onPointerDown={(event) => event.stopPropagation()}
+                            onPointerDownCapture={(event) => {
+                              event.stopPropagation();
+                              focusLocalEditPromptInput(region.id);
+                            }}
                             placeholder={`区域 ${region.index} 提示词`}
+                            ref={(input) => {
+                              localEditPromptInputRefs.current[region.id] = input;
+                            }}
                             value={region.prompt}
                           />
                         ) : (
@@ -7586,6 +7628,7 @@ export function ImageNode({
                         )}
                         <button
                           className="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-local-edit-control="true"
                           onClick={() => deleteLocalEditRegion(region.id)}
                           title="删除区域"
                           type="button"
@@ -8167,6 +8210,7 @@ export function ImageNode({
         createPortal(
           <div
             className="fixed z-[9999] flex items-center gap-2 rounded-[24px] border border-slate-200 bg-white px-2 py-2 text-gray-800 shadow-[0_14px_38px_rgba(15,23,42,0.16)] dark:border-slate-700 dark:bg-slate-900 dark:text-gray-100"
+            data-local-edit-control="true"
             onClick={(event) => event.stopPropagation()}
             onMouseDown={(event) => event.stopPropagation()}
             onPointerDown={(event) => event.stopPropagation()}