diff --git a/components/canvas/CanvasChatPanel.tsx b/components/canvas/CanvasChatPanel.tsx
index 8ce64f5a2..deefd87b2 100644
--- a/components/canvas/CanvasChatPanel.tsx
+++ b/components/canvas/CanvasChatPanel.tsx
@@ -6630,7 +6630,9 @@ export function CanvasChatPanel({
     return lines.join("\n");
   }, [nodes, selectedImageNodes]);
 
-  const buildContextPayload = useCallback(() => {
+  const buildContextPayload = useCallback((textOverride?: string) => {
+    const contextText =
+      typeof textOverride === "string" ? textOverride : input;
     const pendingAnnotationTargetNode = resolvePendingCanvasAgentTargetNode({
       nodes,
       pendingTarget: pendingCanvasAgentAnnotationEditTargetRef.current,
@@ -6660,7 +6662,7 @@ export function CanvasChatPanel({
     const shouldIsolateDirectModeContext = effectiveComposerMode !== "agent";
     const shouldPreserveCanvasControlContext =
       shouldBypassCanvasImagePointsPrecheckForAgentControl({
-        text: input,
+        text: contextText,
         activeComposerMode: effectiveComposerMode,
         attachmentCount: attachments.length,
         hasSkillTag: Boolean(selectedSkillTag),
@@ -6673,7 +6675,7 @@ export function CanvasChatPanel({
     ).length;
     const shouldPreserveSelectedSlotContext =
       shouldPreserveCanvasAgentSlotOnlyContext({
-        text: input,
+        text: contextText,
         activeComposerMode: effectiveComposerMode,
         selectedAgentSlotCount: selectedAgentSlotCountForContext,
         attachmentCount: attachments.length,
@@ -6697,7 +6699,7 @@ export function CanvasChatPanel({
 
     if (shouldIsolateDirectModeContext) {
       selectedImageNodesForContext = resolveMultiImageReferenceContext({
-        text: input,
+        text: contextText,
         selectedItems: selectedImageNodesForContext,
         candidateItems: imageNodes,
         getItemId: (node) => node.id,
@@ -9699,7 +9701,7 @@ export function CanvasChatPanel({
     const autoContext = buildAutoContext();
     autoContextRef.current =
       clampChatRequestText(autoContext, MAX_CHAT_AUTO_CONTEXT_LENGTH) || null;
-    contextPayloadRef.current = buildContextPayload() || null;
+    contextPayloadRef.current = buildContextPayload(textContent) || null;
     
     if (textContent) {
       parts.push({ type: "text" as const, text: textContent });
diff --git a/lib/ai/providers/snapgen-reference-order.ts b/lib/ai/providers/snapgen-reference-order.ts
index daa03f2bd..9525d97af 100644
--- a/lib/ai/providers/snapgen-reference-order.ts
+++ b/lib/ai/providers/snapgen-reference-order.ts
@@ -5,7 +5,7 @@ export function buildSnapgenReferenceOrderPrompt(prompt: string, imageUrls: stri
 
   return [
     `参考图顺序说明：本次请求包含 ${referenceCount} 张参考图，顺序与上传/选中顺序一致。`,
-    "用户提示中的“第一张图/图1”对应第 1 张参考图，“第二张图/图2”对应第 2 张参考图，依次类推。",
+    "用户提示中的“第一张图/图1/图一/图片1/图片一”对应第 1 张参考图，“第二张图/图2/图二/图片2/图片二”对应第 2 张参考图，依次类推。",
     "本说明只用于明确图片编号，不改变用户对主体、风格、构图、权重或融合方式的要求；若用户指定主参考或融合方式，以用户提示为准。",
     normalizedPrompt,
   ]
diff --git a/lib/canvas/multi-image-reference-context.ts b/lib/canvas/multi-image-reference-context.ts
index 04c86e111..b24b9ca6e 100644
--- a/lib/canvas/multi-image-reference-context.ts
+++ b/lib/canvas/multi-image-reference-context.ts
@@ -1,8 +1,8 @@
 const ORDINAL_REFERENCE_PATTERNS = [
-  /(?:第一|第\s*1|图\s*1|图片\s*1|照片\s*1|first\s+(?:image|photo|reference)|1st\s+(?:image|photo|reference))[\s\S]{0,120}(?:第二|第\s*2|图\s*2|图片\s*2|照片\s*2|second\s+(?:image|photo|reference)|2nd\s+(?:image|photo|reference))/i,
-  /(?:第二|第\s*2|图\s*2|图片\s*2|照片\s*2|second\s+(?:image|photo|reference)|2nd\s+(?:image|photo|reference))[\s\S]{0,120}(?:第一|第\s*1|图\s*1|图片\s*1|照片\s*1|first\s+(?:image|photo|reference)|1st\s+(?:image|photo|reference))/i,
-  /(?:第三|第\s*3|图\s*3|图片\s*3|照片\s*3|third\s+(?:image|photo|reference)|3rd\s+(?:image|photo|reference))/i,
-  /(?:第四|第\s*4|图\s*4|图片\s*4|照片\s*4|fourth\s+(?:image|photo|reference)|4th\s+(?:image|photo|reference))/i,
+  /(?:第一|第\s*1|图片\s*(?:1|一)|照片\s*(?:1|一)|图\s*(?:1|一)|first\s+(?:image|photo|reference)|1st\s+(?:image|photo|reference))[\s\S]{0,120}(?:第二|第\s*(?:2|二|两)|图片\s*(?:2|二|两)|照片\s*(?:2|二|两)|图\s*(?:2|二|两)|second\s+(?:image|photo|reference)|2nd\s+(?:image|photo|reference))/i,
+  /(?:第二|第\s*(?:2|二|两)|图片\s*(?:2|二|两)|照片\s*(?:2|二|两)|图\s*(?:2|二|两)|second\s+(?:image|photo|reference)|2nd\s+(?:image|photo|reference))[\s\S]{0,120}(?:第一|第\s*1|图片\s*(?:1|一)|照片\s*(?:1|一)|图\s*(?:1|一)|first\s+(?:image|photo|reference)|1st\s+(?:image|photo|reference))/i,
+  /(?:第三|第\s*(?:3|三)|图片\s*(?:3|三)|照片\s*(?:3|三)|图\s*(?:3|三)|third\s+(?:image|photo|reference)|3rd\s+(?:image|photo|reference))/i,
+  /(?:第四|第\s*(?:4|四)|图片\s*(?:4|四)|照片\s*(?:4|四)|图\s*(?:4|四)|fourth\s+(?:image|photo|reference)|4th\s+(?:image|photo|reference))/i,
 ];
 
 const MULTI_IMAGE_REFERENCE_PATTERNS = [
@@ -11,7 +11,7 @@ const MULTI_IMAGE_REFERENCE_PATTERNS = [
   /(?:这|这些|这几|上述|以上|下面|上传|给的|参考|选中|画布|原图|素材)[\s\S]{0,18}(?:两|俩|二|三|四|五|六|七|八|2|3|4|5|6|7|8|多|几)[\s\S]{0,12}(?:张|个)?(?:图|图片|照片|参考图|素材)/i,
   /(?:两|俩|二|三|四|五|六|七|八|2|3|4|5|6|7|8|多|几)[\s\S]{0,8}(?:张|个)?(?:参考图|素材图|输入图|源图|上传图|原图)/i,
   /(?:两|俩|二|三|四|五|六|七|八|2|3|4|5|6|7|8|多|几)[\s\S]{0,8}(?:张|个)?(?:图|图片|照片)[\s\S]{0,28}(?:合成|融合|拼接|对比|混合|组合|参考|编辑|处理|改)/i,
-  /(?:合成|融合|拼接|对比|混合|组合|参考|编辑|处理)[\s\S]{0,28}(?:两|俩|二|三|四|五|六|七|八|2|3|4|5|6|7|8|多|几|第一|第二|第三|第四|图\s*1|图\s*2|图\s*3|图\s*4)[\s\S]{0,12}(?:图|图片|照片|参考|素材)?/i,
+  /(?:合成|融合|拼接|对比|混合|组合|参考|编辑|增加|添加|放入|换成)[\s\S]{0,28}(?:两|俩|二|三|四|五|六|七|八|2|3|4|5|6|7|8|多|几|第二|第三|第四|图片\s*(?:2|3|4|二|三|四)|照片\s*(?:2|3|4|二|三|四)|图\s*(?:2|3|4|二|三|四))[\s\S]{0,12}(?:图|图片|照片|参考|素材)?/i,
   /(?:multi|multiple|two|several|many)[\s\S]{0,20}(?:reference|input|source)[\s\S]{0,16}(?:image|photo)s?/i,
   /(?:reference|use|combine|merge|blend|edit)[\s\S]{0,24}(?:first|1st)[\s\S]{0,80}(?:second|2nd)[\s\S]{0,32}(?:image|photo|reference)?/i,
 ];
@@ -19,31 +19,31 @@ const MULTI_IMAGE_REFERENCE_PATTERNS = [
 const REQUESTED_COUNT_PATTERNS: Array<{ count: number; pattern: RegExp }> = [
   {
     count: 8,
-    pattern: /(?:第\s*(?:八|8)\s*(?:张|个)?(?:图|图片|照片)|(?:图|图片|照片|image)\s*8\b|(?:八|8)\s*(?:张|个)?(?:参考图|素材图|输入图|源图|上传图|原图))/i,
+    pattern: /(?:第\s*(?:八|8)\s*(?:张|个)?(?:图|图片|照片)|(?:图片|照片|图|image)\s*(?:8|八)\b|(?:八|8)\s*(?:张|个)?(?:参考图|素材图|输入图|源图|上传图|原图))/i,
   },
   {
     count: 7,
-    pattern: /(?:第\s*(?:七|7)\s*(?:张|个)?(?:图|图片|照片)|(?:图|图片|照片|image)\s*7\b|(?:七|7)\s*(?:张|个)?(?:参考图|素材图|输入图|源图|上传图|原图))/i,
+    pattern: /(?:第\s*(?:七|7)\s*(?:张|个)?(?:图|图片|照片)|(?:图片|照片|图|image)\s*(?:7|七)\b|(?:七|7)\s*(?:张|个)?(?:参考图|素材图|输入图|源图|上传图|原图))/i,
   },
   {
     count: 6,
-    pattern: /(?:第\s*(?:六|6)\s*(?:张|个)?(?:图|图片|照片)|(?:图|图片|照片|image)\s*6\b|(?:六|6)\s*(?:张|个)?(?:参考图|素材图|输入图|源图|上传图|原图))/i,
+    pattern: /(?:第\s*(?:六|6)\s*(?:张|个)?(?:图|图片|照片)|(?:图片|照片|图|image)\s*(?:6|六)\b|(?:六|6)\s*(?:张|个)?(?:参考图|素材图|输入图|源图|上传图|原图))/i,
   },
   {
     count: 5,
-    pattern: /(?:第\s*(?:五|5)\s*(?:张|个)?(?:图|图片|照片)|(?:图|图片|照片|image)\s*5\b|(?:五|5)\s*(?:张|个)?(?:参考图|素材图|输入图|源图|上传图|原图))/i,
+    pattern: /(?:第\s*(?:五|5)\s*(?:张|个)?(?:图|图片|照片)|(?:图片|照片|图|image)\s*(?:5|五)\b|(?:五|5)\s*(?:张|个)?(?:参考图|素材图|输入图|源图|上传图|原图))/i,
   },
   {
     count: 4,
-    pattern: /(?:第\s*(?:四|4)\s*(?:张|个)?(?:图|图片|照片)|(?:图|图片|照片|image)\s*4\b|(?:四|4)\s*(?:张|个)?(?:图|图片|照片|参考图|素材图|输入图|源图|上传图|原图))/i,
+    pattern: /(?:第\s*(?:四|4)\s*(?:张|个)?(?:图|图片|照片)|(?:图片|照片|图|image)\s*(?:4|四)\b|(?:四|4)\s*(?:张|个)?(?:图|图片|照片|参考图|素材图|输入图|源图|上传图|原图))/i,
   },
   {
     count: 3,
-    pattern: /(?:第\s*(?:三|3)\s*(?:张|个)?(?:图|图片|照片)|(?:图|图片|照片|image)\s*3\b|(?:三|3)\s*(?:张|个)?(?:图|图片|照片|参考图|素材图|输入图|源图|上传图|原图))/i,
+    pattern: /(?:第\s*(?:三|3)\s*(?:张|个)?(?:图|图片|照片)|(?:图片|照片|图|image)\s*(?:3|三)\b|(?:三|3)\s*(?:张|个)?(?:图|图片|照片|参考图|素材图|输入图|源图|上传图|原图))/i,
   },
   {
     count: 2,
-    pattern: /(?:第\s*(?:二|两|2)\s*(?:张|个)?(?:图|图片|照片)|(?:图|图片|照片|image)\s*2\b|(?:两|俩|二|2)\s*(?:张|个)?(?:图|图片|照片|参考图|素材图|输入图|源图|上传图|原图))/i,
+    pattern: /(?:第\s*(?:二|两|2)\s*(?:张|个)?(?:图|图片|照片)|(?:图片|照片|图|image)\s*(?:2|二|两)\b|(?:两|俩|二|2)\s*(?:张|个)?(?:图|图片|照片|参考图|素材图|输入图|源图|上传图|原图))/i,
   },
 ];