diff --git a/src/api/adVideoPlanClient.ts b/src/api/adVideoPlanClient.ts
index ae700ea..2bae189 100644
--- a/src/api/adVideoPlanClient.ts
+++ b/src/api/adVideoPlanClient.ts
@@ -1,7 +1,8 @@
 import { buildApiUrl, buildAuthHeaders } from "./serverConnection";
 
 const TEXT_MODEL = "qwen-max";
-const VISION_MODEL = "qwen3.6-plus";
+const VISION_MODEL = "qwen3.7-plus";
+const VISION_FALLBACK_MODEL = "qwen-vl-plus";
 
 export interface AdVideoUserConfig {
   platform: string;
@@ -107,36 +108,63 @@ interface ChatMessage {
   content: string;
 }
 
+const MAX_RETRIES = 3;
+const RETRY_BASE_MS = 2000;
+const CHAT_TIMEOUT_MS = 120_000; // 2 minutes per AI call
+
+function isTransientError(err: unknown): boolean {
+  if (!(err instanceof Error)) return false;
+  const msg = err.message.toLowerCase();
+  return /\b429\b/.test(msg) || msg.includes("signal timed out") || msg.includes("aborted") || msg.includes("timeout");
+}
+
+async function retryOnTransient<T>(fn: () => Promise<T>, signal?: AbortSignal): Promise<T> {
+  for (let attempt = 0; attempt <= MAX_RETRIES; attempt++) {
+    try {
+      return await fn();
+    } catch (err) {
+      if (signal?.aborted) throw err;
+      if (attempt === MAX_RETRIES) throw err;
+      if (!isTransientError(err)) throw err;
+      const delay = RETRY_BASE_MS * 2 ** attempt + Math.random() * 1000;
+      await new Promise((r) => setTimeout(r, delay));
+    }
+  }
+  throw new Error("unreachable");
+}
+
 async function chat(
   systemPrompt: string,
   userContent: string,
   options?: { model?: string; signal?: AbortSignal },
 ): Promise<string> {
-  const messages: ChatMessage[] = [
-    { role: "system", content: systemPrompt },
-    { role: "user", content: userContent },
-  ];
-  const timeoutSignal = AbortSignal.timeout(60000);
-  const combinedSignal = options?.signal
-    ? AbortSignal.any([options.signal, timeoutSignal])
-    : timeoutSignal;
-  const res = await fetch(buildApiUrl("ai/chat"), {
-    method: "POST",
-    headers: buildAuthHeaders(),
-    body: JSON.stringify({
-      model: options?.model ?? TEXT_MODEL,
-      messages,
-      stream: false,
-      temperature: 0.4,
-    }),
-    signal: combinedSignal,
-  });
-  if (!res.ok) throw new Error(`AI 调用失败 (${res.status})`);
-  const payload = await res.json();
-  const content: string =
-    payload?.content ?? payload?.choices?.[0]?.message?.content ?? payload?.text ?? "";
-  if (!content) throw new Error("模型未返回有效内容");
-  return content;
+  return retryOnTransient(async () => {
+    const messages: ChatMessage[] = [
+      { role: "system", content: systemPrompt },
+      { role: "user", content: userContent },
+    ];
+    const timeoutSignal = AbortSignal.timeout(CHAT_TIMEOUT_MS);
+    const combinedSignal = options?.signal
+      ? AbortSignal.any([options.signal, timeoutSignal])
+      : timeoutSignal;
+    const res = await fetch(buildApiUrl("ai/chat"), {
+      method: "POST",
+      headers: buildAuthHeaders(),
+      body: JSON.stringify({
+        model: options?.model ?? TEXT_MODEL,
+        messages,
+        stream: false,
+        temperature: 0.4,
+      }),
+      signal: combinedSignal,
+    });
+    if (!res.ok) throw new Error(`AI 调用失败 (${res.status})`);
+    const payload = await res.json();
+    const content: string =
+      payload?.content ?? payload?.choices?.[0]?.message?.content ?? payload?.text ?? "";
+    if (!content) throw new Error("模型未返回有效内容");
+    return content;
+  }, options?.signal);
 }
 
 async function visionChat(
@@ -149,30 +177,43 @@ async function visionChat(
     ...imageUrls.map((url) => ({ type: "image_url", image_url: { url } })),
     { type: "text", text },
   ];
-  const timeoutSignal = AbortSignal.timeout(60000);
-  const combinedSignal = signal
-    ? AbortSignal.any([signal, timeoutSignal])
-    : timeoutSignal;
-  const res = await fetch(buildApiUrl("ai/chat"), {
-    method: "POST",
-    headers: buildAuthHeaders(),
-    body: JSON.stringify({
-      model: VISION_MODEL,
-      messages: [
-        { role: "system", content: systemPrompt },
-        { role: "user", content },
-      ],
-      stream: false,
-      temperature: 0.3,
-    }),
-    signal: combinedSignal,
-  });
-  if (!res.ok) throw new Error(`图片理解调用失败 (${res.status})`);
-  const payload = await res.json();
-  const out: string =
-    payload?.content ?? payload?.choices?.[0]?.message?.content ?? payload?.text ?? "";
-  if (!out) throw new Error("图片理解未返回有效内容");
-  return out;
+  const messages = [
+    { role: "system", content: systemPrompt },
+    { role: "user", content },
+  ];
+
+  for (const model of [VISION_MODEL, VISION_FALLBACK_MODEL]) {
+    const timeoutSignal = AbortSignal.timeout(CHAT_TIMEOUT_MS);
+    const combinedSignal = signal
+      ? AbortSignal.any([signal, timeoutSignal])
+      : timeoutSignal;
+    try {
+      const out = await retryOnTransient(async () => {
+        const res = await fetch(buildApiUrl("ai/chat"), {
+          method: "POST",
+          headers: buildAuthHeaders(),
+          body: JSON.stringify({ model, messages, stream: false, temperature: 0.3 }),
+          signal: combinedSignal,
+        });
+        if (!res.ok) {
+          const errBody = await res.text().catch(() => "");
+          if (model === VISION_MODEL && errBody.includes("image format")) throw new Error("IMAGE_FORMAT_FALLBACK");
+          throw new Error(`图片理解调用失败 (${res.status})`);
+        }
+        const payload = await res.json();
+        const result: string =
+          payload?.content ?? payload?.choices?.[0]?.message?.content ?? payload?.text ?? "";
+        if (!result) throw new Error("图片理解未返回有效内容");
+        return result;
+      }, signal);
+      return out;
+    } catch (err) {
+      if (err instanceof Error && err.message === "IMAGE_FORMAT_FALLBACK") continue;
+      if (model === VISION_MODEL && err instanceof Error && err.message?.includes("图片理解调用失败")) continue;
+      throw err;
+    }
+  }
+  throw new Error("图片理解调用失败，所有模型均不可用");
 }
 
 const IMAGE_UNDERSTANDING_PROMPT = `你是电商产品图片分析专家。请分析用户提供的产品图片，识别产品主体、外观、颜色、材质、形状、尺寸感、品牌标识、关键部件、可视化卖点和适合展示的镜头角度。请用简洁的中文段落描述，不要编造图片中看不到的信息。`;
diff --git a/src/api/aiGenerationClient.ts b/src/api/aiGenerationClient.ts
index f580b5a..1335847 100644
--- a/src/api/aiGenerationClient.ts
+++ b/src/api/aiGenerationClient.ts
@@ -403,6 +403,24 @@ export const aiGenerationClient = {
     return readJsonResponse<{ url: string; ossKey?: string }>(res, "Asset upload response failed");
   },
 
+  async uploadAssetBinary(blob: Blob, options?: { name?: string; mimeType?: string; scope?: string }): Promise<{ url: string; signedUrl?: string; ossKey?: string }> {
+    const form = new FormData();
+    form.append("file", blob, options?.name || "upload.png");
+    if (options?.scope) form.append("scope", options.scope);
+    if (options?.mimeType) form.append("mimeType", options.mimeType);
+    // Exclude Content-Type so browser auto-sets multipart/form-data with boundary
+    const { "Content-Type": _ct, ...authHeaders } = buildAuthHeaders();
+    const res = await fetch(buildApiUrl("oss/upload-binary"), {
+      method: "POST",
+      headers: authHeaders,
+      body: form,
+    });
+    if (!res.ok) {
+      await throwResponseError(res, "Binary asset upload failed");
+    }
+    return readJsonResponse<{ url: string; ossKey?: string }>(res, "Binary asset upload response failed");
+  },
+
   async uploadAssetByUrl(input: UploadAssetByUrlInput): Promise<{ url: string; signedUrl?: string; ossKey?: string }> {
     const res = await fetch(buildApiUrl("oss/upload-by-url"), {
       method: "POST",
diff --git a/src/features/ecommerce/EcommercePage.tsx b/src/features/ecommerce/EcommercePage.tsx
index 1f561ba..3c500ba 100644
--- a/src/features/ecommerce/EcommercePage.tsx
+++ b/src/features/ecommerce/EcommercePage.tsx
@@ -1321,18 +1321,15 @@ function ProductClonePage(_props: ProductClonePageProps = {}) {
   });
 
   const uploadProductImages = async (): Promise<string[]> => {
+    const SUPPORTED_IMAGE_TYPES = new Set(["image/jpeg", "image/png", "image/webp", "image/gif"]);
     const urls: string[] = [];
     for (const item of productImages) {
       try {
         const resp = await fetch(item.src);
-        const blob = await resp.blob();
-        const dataUrl = await new Promise<string>((resolve, reject) => {
-          const reader = new FileReader();
-          reader.onload = () => resolve(String(reader.result));
-          reader.onerror = () => reject(reader.error);
-          reader.readAsDataURL(blob);
-        });
-        const { url } = await aiGenerationClient.uploadAsset({ dataUrl, name: item.name, mimeType: blob.type });
+        const rawBlob = await resp.blob();
+        const mimeType = SUPPORTED_IMAGE_TYPES.has(rawBlob.type) ? rawBlob.type : "image/png";
+        const blob = rawBlob.type === mimeType ? rawBlob : new Blob([rawBlob], { type: mimeType });
+        const { url } = await aiGenerationClient.uploadAssetBinary(blob, { name: item.name, mimeType, scope: "ecommerce-product" });
         urls.push(url);
       } catch {
         // skip images that fail to upload
diff --git a/src/features/ecommerce/ecommerceVideoService.ts b/src/features/ecommerce/ecommerceVideoService.ts
index f1f7639..5f92e86 100644
--- a/src/features/ecommerce/ecommerceVideoService.ts
+++ b/src/features/ecommerce/ecommerceVideoService.ts
@@ -9,7 +9,6 @@ import {
   type AdVideoUserConfig,
 } from "../../api/adVideoPlanClient";
 import { aiGenerationClient } from "../../api/aiGenerationClient";
-import { uploadAssetWithProgress } from "../../api/uploadWithProgress";
 import { waitForTask } from "../../api/taskSubscription";
 import { resolveVideoRequestModel } from "../../utils/resolveVideoModel";
 import type {
@@ -34,12 +33,18 @@ export async function runVideoPlan(
 
   onStepStart("upload");
   const imageUrls: string[] = [];
-  for (const dataUrl of imageDataUrls) {
-    const result = await uploadAssetWithProgress(
-      { dataUrl, scope: "ecommerce-product", mimeType: "image/png" },
-      { signal },
-    );
-    imageUrls.push(result.url);
+  const SUPPORTED_IMAGE_TYPES = new Set(["image/jpeg", "image/png", "image/webp", "image/gif"]);
+  for (const srcUrl of imageDataUrls) {
+    try {
+      const resp = await fetch(srcUrl);
+      const rawBlob = await resp.blob();
+      const mimeType = SUPPORTED_IMAGE_TYPES.has(rawBlob.type) ? rawBlob.type : "image/png";
+      const blob = rawBlob.type === mimeType ? rawBlob : new Blob([rawBlob], { type: mimeType });
+      const result = await aiGenerationClient.uploadAssetBinary(blob, { mimeType, scope: "ecommerce-product" });
+      imageUrls.push(result.url);
+    } catch {
+      // skip images that fail to upload
+    }
   }
   onStepDone("upload");