diff --git a/components/multimodal-input.tsx b/components/multimodal-input.tsx
index b7c90c2f0..2992989f1 100644
--- a/components/multimodal-input.tsx
+++ b/components/multimodal-input.tsx
@@ -52,6 +52,14 @@ const VIDEO_PROVIDER_TYPES = new Set([
   "fal-video",
 ]);
 
+function isVideoProviderType(providerType?: string | null) {
+  return VIDEO_PROVIDER_TYPES.has(String(providerType || ""));
+}
+
+function isImageGenerationModel(model: DynamicModel) {
+  return model.supportsImageGen === true && !isVideoProviderType(model.providerType);
+}
+
 function isNanoBananaProModel(model: { displayName?: string | null; modelId?: string | null }) {
   const displayName = String(model.displayName || "").toLowerCase();
   const modelId = String(model.modelId || "").toLowerCase();
@@ -237,21 +245,20 @@ function PureMultimodalInput({
     const models = modelCatalog?.models || [];
     const nanoBananaImage = models.find(
       (model) =>
-        model.supportsImageGen &&
-        !VIDEO_PROVIDER_TYPES.has(model.providerType || "") &&
+        isImageGenerationModel(model) &&
         isNanoBananaProModel(model)
     );
     const imageMain = models.find(
-      (model) => model.supportsImageGen && !VIDEO_PROVIDER_TYPES.has(model.providerType || "") && !model.isSubModel
+      (model) => isImageGenerationModel(model) && !model.isSubModel
     );
     const imageFallback = models.find(
-      (model) => model.supportsImageGen && !VIDEO_PROVIDER_TYPES.has(model.providerType || "")
+      (model) => isImageGenerationModel(model)
     );
     const videoMain = models.find(
-      (model) => VIDEO_PROVIDER_TYPES.has(model.providerType || "") && !model.isSubModel
+      (model) => isVideoProviderType(model.providerType) && !model.isSubModel
     );
     const videoFallback = models.find((model) =>
-      VIDEO_PROVIDER_TYPES.has(model.providerType || "")
+      isVideoProviderType(model.providerType)
     );
     return {
       image: nanoBananaImage || imageMain || imageFallback || null,
@@ -260,14 +267,14 @@ function PureMultimodalInput({
   }, [modelCatalog?.models]);
 
   const activeQuickMode = useMemo<"image" | "video" | null>(() => {
-    if (isVideoModel || VIDEO_PROVIDER_TYPES.has(currentSelectedModel?.providerType || "")) {
+    if (isVideoModel || isVideoProviderType(currentSelectedModel?.providerType)) {
       return "video";
     }
-    if (isImageModel || currentSelectedModel?.supportsImageGen) {
+    if (isImageModel || (currentSelectedModel && isImageGenerationModel(currentSelectedModel))) {
       return "image";
     }
     return null;
-  }, [currentSelectedModel?.providerType, currentSelectedModel?.supportsImageGen, isImageModel, isVideoModel]);
+  }, [currentSelectedModel, isImageModel, isVideoModel]);
 
   const switchComposerMode = useCallback(
     (targetMode: "image" | "video") => {
@@ -1255,6 +1262,7 @@ function PureMultimodalInput({
                   selectedModelId={selectedModelId}
                   attachments={attachments}
                   avatarOnly={true}
+                  modeFilter="video"
                 />
               </>
             ) : (
@@ -1269,6 +1277,7 @@ function PureMultimodalInput({
                     onModelChange={onModelChange}
                     selectedModelId={selectedModelId}
                     attachments={attachments}
+                    modeFilter="image"
                   />
                 ) : null}
                 {isImageModel && onImageParamsChange && (
@@ -1417,12 +1426,14 @@ function PureModelSelectorCompact({
   attachments,
   avatarOnly = false,
   readOnly = false,
+  modeFilter = "all",
 }: {
   selectedModelId: string;
   onModelChange?: (modelId: string) => void;
   attachments: Attachment[];
   avatarOnly?: boolean;
   readOnly?: boolean;
+  modeFilter?: "all" | "image" | "video";
 }) {
   const [optimisticModelId, setOptimisticModelId] = useState(selectedModelId);
 
@@ -1496,9 +1507,20 @@ function PureModelSelectorCompact({
     providerName: "System",
   };
 
+  const modeFilteredModels = useMemo(() => {
+    if (modeFilter === "video") {
+      return dynamicModels.filter((model) => isVideoProviderType(model.providerType));
+    }
+    if (modeFilter === "image") {
+      return dynamicModels.filter((model) => isImageGenerationModel(model));
+    }
+    return dynamicModels;
+  }, [dynamicModels, modeFilter]);
+
   // 合并模型列表（根据设置决定是否包含自动模型）
   const showAutoModel = siteSettings?.showAutoModel ?? true;
-  const displayModels = showAutoModel ? [autoModel, ...dynamicModels] : dynamicModels;
+  const canShowAutoModel = showAutoModel && modeFilter === "all";
+  const displayModels = canShowAutoModel ? [autoModel, ...modeFilteredModels] : modeFilteredModels;
 
   useEffect(() => {
     setOptimisticModelId(selectedModelId);
@@ -1510,10 +1532,12 @@ function PureModelSelectorCompact({
     const currentExists = displayModels.some((m) => m.id === selectedModelId);
     if (currentExists) return;
 
+    const fallbackImageModel =
+      modeFilteredModels.find((model) => isNanoBananaProModel(model)) ||
+      modeFilteredModels.find((model) => model.isDefault) ||
+      modeFilteredModels[0];
     const fallbackModelId =
-      dynamicModels.find((m) => m.isDefault)?.id ??
-      dynamicModels[0]?.id ??
-      (showAutoModel ? "auto" : "");
+      fallbackImageModel?.id ?? (canShowAutoModel ? "auto" : "");
 
     if (!fallbackModelId) return;
 
@@ -1522,7 +1546,7 @@ function PureModelSelectorCompact({
     startTransition(() => {
       saveChatModelAsCookie(fallbackModelId);
     });
-  }, [displayModels, dynamicModels, onModelChange, selectedModelId, showAutoModel]);
+  }, [canShowAutoModel, displayModels, modeFilteredModels, onModelChange, selectedModelId]);
 
   // 自动选择模型的逻辑 - 必须在所有条件 return 之前定义
   const selectBestModel = useCallback(() => {
@@ -1760,6 +1784,9 @@ const ModelSelectorCompact = memo(
     if (prevProps.readOnly !== nextProps.readOnly) {
       return false;
     }
+    if (prevProps.modeFilter !== nextProps.modeFilter) {
+      return false;
+    }
     if (!equal(prevProps.attachments, nextProps.attachments)) {
       return false;
     }