diff --git a/components/home/lovart-home.tsx b/components/home/lovart-home.tsx
index e39d340ca..770c06d2c 100644
--- a/components/home/lovart-home.tsx
+++ b/components/home/lovart-home.tsx
@@ -766,23 +766,6 @@ export function LovartHome({
       .filter(Boolean);
   }, [composerFixedPrefix, composerPreviewTexts]);
 
-  const lovartPromptTagVisibility = useMemo<LovartPromptTagVisibility>(() => {
-    const raw = siteSettingsData?.lovartPromptTagVisibility || {};
-    return {
-      nanoBananaPro: raw.nanoBananaPro !== false,
-      design: raw.design !== false,
-      branding: raw.branding !== false,
-      illustration: raw.illustration !== false,
-      ecommerce: raw.ecommerce !== false,
-      video: raw.video !== false,
-    };
-  }, [siteSettingsData?.lovartPromptTagVisibility]);
-
-  const visiblePromptTags = useMemo(
-    () => PROMPT_TAGS.filter((tag) => lovartPromptTagVisibility[tag.key]),
-    [lovartPromptTagVisibility]
-  );
-
   const contactConfig = useMemo(
     () => ({
       phone: String(siteSettingsData?.contactConfig?.phone || "").trim(),
@@ -1826,7 +1809,6 @@ export function LovartHome({
             shouldInvertLogo={shouldInvertLogo}
             siteName={siteName}
             selectedModelId={selectedComposerModel.id}
-            visiblePromptTags={visiblePromptTags}
           />
         )}
 
diff --git a/components/home/lovart/lovart-hero-composer.tsx b/components/home/lovart/lovart-hero-composer.tsx
index 60721643f..c5f491c59 100644
--- a/components/home/lovart/lovart-hero-composer.tsx
+++ b/components/home/lovart/lovart-hero-composer.tsx
@@ -11,9 +11,6 @@ import {
 import { Textarea } from "@/components/ui/textarea";
 import { cn } from "@/lib/utils";
 import type { ModelOption, UploadFileItem } from "./types";
-import type { PROMPT_TAGS } from "./utils";
-
-type PromptTag = (typeof PROMPT_TAGS)[number];
 
 type LovartHeroComposerProps = {
   composerPlaceholder: string;
@@ -43,7 +40,6 @@ type LovartHeroComposerProps = {
   onModelChange: (modelId: string) => void;
   onStartAgentMode: () => void;
   selectedModelId: string;
-  visiblePromptTags: PromptTag[];
 };
 
 export function LovartHeroComposer({
@@ -70,7 +66,6 @@ export function LovartHeroComposer({
   onModelChange,
   onStartAgentMode,
   selectedModelId,
-  visiblePromptTags,
 }: LovartHeroComposerProps) {
   return (
     <section className="mt-8 text-center">
@@ -111,7 +106,7 @@ export function LovartHeroComposer({
         {heroSubtitleText || "懂你的设计代理，帮你搞定一切"}
       </p>
 
-      <div className="mx-auto mt-[34px] w-full max-w-[990px] rounded-[22px] border border-[#e1e4e8] bg-white px-[24px] pb-[12px] pt-[18px] shadow-[0_10px_26px_rgba(0,0,0,0.06)] dark:border-white/10 dark:bg-[#141416] dark:shadow-[0_18px_50px_rgba(0,0,0,0.35)]">
+      <div className="mx-auto mt-[34px] w-full max-w-[900px] rounded-[22px] border border-[#e1e4e8] bg-white px-[24px] pb-[12px] pt-[18px] shadow-[0_10px_26px_rgba(0,0,0,0.06)] dark:border-white/10 dark:bg-[#141416] dark:shadow-[0_18px_50px_rgba(0,0,0,0.35)]">
         <Textarea
           value={prompt}
           onChange={(event) => setPrompt(event.target.value)}
@@ -222,23 +217,6 @@ export function LovartHeroComposer({
         </div>
       </div>
 
-      <div className="mt-[22px] flex flex-wrap items-center justify-center gap-x-[12px] gap-y-[12px]">
-        {visiblePromptTags.map((tag) => (
-          <button
-            key={tag.label}
-            type="button"
-            className={`inline-flex h-[36px] items-center gap-1.5 rounded-full border px-[13px] text-[14px] font-normal ${
-              tag.highlighted
-                ? "border-[#ffb15f] bg-[#fff4e7] text-[#ff8c1f] dark:border-[#8b6330] dark:bg-[#38270f] dark:text-[#ffbe6b]"
-                : "border-black/15 bg-white text-black/70 dark:border-white/10 dark:bg-[#141416] dark:text-white/70"
-            }`}
-          >
-            <tag.icon className="h-3.5 w-3.5" />
-            {tag.label}
-          </button>
-        ))}
-      </div>
-
       {homeAnimationData?.enabled && homeAnimationData?.htmlContent ? (
         <div className="mt-6 w-full">
           <HomeAnimationEmbed htmlContent={homeAnimationData.htmlContent} />
diff --git a/components/home/lovart/lovart-projects-section.tsx b/components/home/lovart/lovart-projects-section.tsx
index 656c01309..36ab66f87 100644
--- a/components/home/lovart/lovart-projects-section.tsx
+++ b/components/home/lovart/lovart-projects-section.tsx
@@ -29,9 +29,9 @@ export function LovartProjectsSection({
   return (
     <section
       ref={projectSectionRef}
-      className={isProjectsOnlyView ? "mt-6" : "mt-[86px]"}
+      className={isProjectsOnlyView ? "mt-6" : "mt-[156px]"}
     >
-      <div className="mb-[56px] flex items-center justify-between">
+      <div className="mb-[32px] flex items-center justify-between">
         <h2 className="text-[26px] font-semibold leading-none tracking-[-0.01em] text-[#1f1f1f] dark:text-white">
           {isProjectsOnlyView ? "所有项目" : "最近项目"}
         </h2>
diff --git a/lib/ai/grsai-gpt-image-sizes.ts b/lib/ai/grsai-gpt-image-sizes.ts
index a7820d10b..444fd5ffe 100644
--- a/lib/ai/grsai-gpt-image-sizes.ts
+++ b/lib/ai/grsai-gpt-image-sizes.ts
@@ -7,33 +7,36 @@ type GrsaiGptImageSizeOption = {
 };
 
 const GRSAI_GPT_IMAGE_SIZE_OPTIONS: GrsaiGptImageSizeOption[] = [
-  { bucket: "1K", width: 1024, height: 1024 },
-  { bucket: "1K", width: 1774, height: 887 },
-  { bucket: "1K", width: 887, height: 1774 },
-  { bucket: "1K", width: 1536, height: 1024 },
-  { bucket: "1K", width: 1024, height: 1536 },
+  { bucket: "1K", width: 1280, height: 1280 },
+  { bucket: "1K", width: 848, height: 1280 },
+  { bucket: "1K", width: 1280, height: 848 },
+  { bucket: "1K", width: 960, height: 1280 },
+  { bucket: "1K", width: 1280, height: 960 },
+  { bucket: "1K", width: 1024, height: 1280 },
+  { bucket: "1K", width: 1280, height: 1024 },
+  { bucket: "1K", width: 720, height: 1280 },
+  { bucket: "1K", width: 1280, height: 720 },
+  { bucket: "1K", width: 1280, height: 544 },
   { bucket: "2K", width: 2048, height: 2048 },
-  { bucket: "2K", width: 2048, height: 1152 },
-  { bucket: "2K", width: 1152, height: 2048 },
-  { bucket: "2K", width: 2048, height: 1360 },
   { bucket: "2K", width: 1360, height: 2048 },
-  { bucket: "2K", width: 2048, height: 880 },
-  { bucket: "2K", width: 880, height: 2048 },
-  { bucket: "2K", width: 2048, height: 688 },
-  { bucket: "2K", width: 688, height: 2048 },
-  { bucket: "2K", width: 2048, height: 1024 },
-  { bucket: "2K", width: 1024, height: 2048 },
+  { bucket: "2K", width: 2048, height: 1360 },
+  { bucket: "2K", width: 1536, height: 2048 },
+  { bucket: "2K", width: 2048, height: 1536 },
+  { bucket: "2K", width: 1632, height: 2048 },
+  { bucket: "2K", width: 2048, height: 1632 },
+  { bucket: "2K", width: 1152, height: 2048 },
+  { bucket: "2K", width: 2048, height: 1152 },
+  { bucket: "2K", width: 2048, height: 864 },
   { bucket: "4K", width: 2880, height: 2880 },
-  { bucket: "4K", width: 3840, height: 2160 },
+  { bucket: "4K", width: 2336, height: 3520 },
+  { bucket: "4K", width: 3520, height: 2336 },
+  { bucket: "4K", width: 2480, height: 3312 },
+  { bucket: "4K", width: 3312, height: 2480 },
+  { bucket: "4K", width: 2560, height: 3216 },
+  { bucket: "4K", width: 3216, height: 2560 },
   { bucket: "4K", width: 2160, height: 3840 },
-  { bucket: "4K", width: 3504, height: 2336 },
-  { bucket: "4K", width: 2336, height: 3504 },
-  { bucket: "4K", width: 3840, height: 1648 },
-  { bucket: "4K", width: 1648, height: 3840 },
-  { bucket: "4K", width: 3840, height: 1280 },
-  { bucket: "4K", width: 1280, height: 3840 },
-  { bucket: "4K", width: 3840, height: 1920 },
-  { bucket: "4K", width: 1920, height: 3840 },
+  { bucket: "4K", width: 3840, height: 2160 },
+  { bucket: "4K", width: 3840, height: 1632 },
 ];
 
 const GRSAI_GPT_IMAGE_SIZE_VALUES = new Set(
diff --git a/lib/ai/providers/alibaba-image.ts b/lib/ai/providers/alibaba-image.ts
index 003ca12de..a3d89154f 100644
--- a/lib/ai/providers/alibaba-image.ts
+++ b/lib/ai/providers/alibaba-image.ts
@@ -50,6 +50,9 @@ interface AlibabaImageResponse {
 const ALIBABA_IMAGE_FETCH_TIMEOUT_MS = 20000;
 const ALIBABA_MIN_SIZE = 512;
 const ALIBABA_MAX_SIZE = 2048;
+const ALIBABA_MAX_INPUT_SIDE = 4096;
+const ALIBABA_DEFAULT_MAX_INPUT_BYTES = 18 * 1024 * 1024;
+const ALIBABA_QWEN_MAX_INPUT_BYTES = 9.5 * 1024 * 1024;
 
 function normalizeImageSizeLabel(imageSize?: string): "1K" | "2K" | "4K" {
   const normalized = String(imageSize || "").trim().toUpperCase();
@@ -57,6 +60,68 @@ function normalizeImageSizeLabel(imageSize?: string): "1K" | "2K" | "4K" {
   return "2K";
 }
 
+function getAlibabaMaxInputBytes(modelId: string) {
+  const normalized = modelId.trim().toLowerCase();
+  if (normalized.includes("qwen-image")) {
+    return ALIBABA_QWEN_MAX_INPUT_BYTES;
+  }
+  return ALIBABA_DEFAULT_MAX_INPUT_BYTES;
+}
+
+function bufferToArrayBuffer(buffer: Buffer): ArrayBuffer {
+  return buffer.buffer.slice(
+    buffer.byteOffset,
+    buffer.byteOffset + buffer.byteLength
+  ) as ArrayBuffer;
+}
+
+function createAbortSignal(timeoutMs: number) {
+  const controller = new AbortController();
+  const timeoutId = setTimeout(() => controller.abort(), timeoutMs);
+  return {
+    signal: controller.signal,
+    clear: () => clearTimeout(timeoutId),
+  };
+}
+
+async function fetchImageBufferWithRetry(
+  imageUrl: string,
+  timeoutMs = ALIBABA_IMAGE_FETCH_TIMEOUT_MS,
+  attempts = 2
+): Promise<{ buffer: Buffer; contentType: string | null }> {
+  let lastError: unknown = null;
+  for (let attempt = 1; attempt <= attempts; attempt += 1) {
+    const ctx = createAbortSignal(timeoutMs);
+    try {
+      const response = await fetch(imageUrl, {
+        cache: "no-store",
+        redirect: "follow",
+        signal: ctx.signal,
+      });
+      if (!response.ok) {
+        throw new Error(`Failed to download image: ${response.status}`);
+      }
+      const buffer = Buffer.from(await response.arrayBuffer());
+      if (!buffer.length) {
+        throw new Error("Downloaded image is empty");
+      }
+      return {
+        buffer,
+        contentType: response.headers.get("content-type"),
+      };
+    } catch (error) {
+      lastError = error;
+      if (attempt < attempts) {
+        await new Promise((resolve) => setTimeout(resolve, attempt * 500));
+      }
+    } finally {
+      ctx.clear();
+    }
+  }
+
+  throw lastError instanceof Error ? lastError : new Error(String(lastError || "下载图片失败"));
+}
+
 function calculateAlibabaSizeByAspect(aspectValue: number, baseSize: number): string {
   const safeAspect = Number.isFinite(aspectValue) && aspectValue > 0 ? aspectValue : 1;
   const totalPixels = baseSize * baseSize;
@@ -97,16 +162,8 @@ function calculateAlibabaSizeByAspect(aspectValue: number, baseSize: number): st
 async function inferAspectRatioFromImageUrl(imageUrl: string): Promise<number | null> {
   if (!/^https?:\/\//i.test(imageUrl)) return null;
 
-  const controller = new AbortController();
-  const timeoutId = setTimeout(() => controller.abort(), ALIBABA_IMAGE_FETCH_TIMEOUT_MS);
   try {
-    const response = await fetch(imageUrl, { signal: controller.signal });
-    if (!response.ok) {
-      console.warn("[Alibaba Image] Failed to fetch reference image for auto size:", response.status);
-      return null;
-    }
-
-    const buffer = Buffer.from(await response.arrayBuffer());
+    const { buffer } = await fetchImageBufferWithRetry(imageUrl);
     if (!buffer.length) return null;
 
     const metadata = await sharp(buffer).metadata();
@@ -115,8 +172,6 @@ async function inferAspectRatioFromImageUrl(imageUrl: string): Promise<number |
   } catch (error) {
     console.warn("[Alibaba Image] Failed to infer aspect ratio from reference image:", error);
     return null;
-  } finally {
-    clearTimeout(timeoutId);
   }
 }
 
@@ -125,16 +180,8 @@ async function inferImageDimensionsFromImageUrl(
 ): Promise<{ width: number; height: number; aspectRatio: number } | null> {
   if (!/^https?:\/\//i.test(imageUrl)) return null;
 
-  const controller = new AbortController();
-  const timeoutId = setTimeout(() => controller.abort(), ALIBABA_IMAGE_FETCH_TIMEOUT_MS);
   try {
-    const response = await fetch(imageUrl, { signal: controller.signal });
-    if (!response.ok) {
-      console.warn("[Alibaba Image] Failed to fetch reference image dimensions:", response.status);
-      return null;
-    }
-
-    const buffer = Buffer.from(await response.arrayBuffer());
+    const { buffer } = await fetchImageBufferWithRetry(imageUrl);
     if (!buffer.length) return null;
 
     const metadata = await sharp(buffer).metadata();
@@ -148,8 +195,6 @@ async function inferImageDimensionsFromImageUrl(
   } catch (error) {
     console.warn("[Alibaba Image] Failed to infer image dimensions from reference image:", error);
     return null;
-  } finally {
-    clearTimeout(timeoutId);
   }
 }
 
@@ -219,6 +264,124 @@ function generateFilename(prefix: string, extension: string): string {
   return `${prefix}-${timestamp}-${random}.${extension}`;
 }
 
+async function uploadPreparedAlibabaInputImage(
+  buffer: Buffer,
+  extension: "png" | "jpg"
+): Promise<string> {
+  const filename = generateFilename("alibaba-input", extension);
+  const result = await uploadFile(filename, bufferToArrayBuffer(buffer));
+  console.log(`[Alibaba Image] Prepared input image uploaded to: ${result.url}`);
+  return result.url;
+}
+
+async function prepareAlibabaInputImageUrl(
+  imageUrl: string,
+  modelId: string
+): Promise<string> {
+  const normalizedUrl = String(imageUrl || "").trim();
+  if (!/^https?:\/\//i.test(normalizedUrl)) return normalizedUrl;
+
+  const maxInputBytes = getAlibabaMaxInputBytes(modelId);
+  const { buffer, contentType } = await fetchImageBufferWithRetry(normalizedUrl);
+  let metadata: sharp.Metadata;
+  try {
+    metadata = await sharp(buffer, { failOn: "none" }).metadata();
+  } catch (error) {
+    throw new Error(
+      `输入图片无法识别，请转换为 JPG/PNG 后重试: ${
+        error instanceof Error ? error.message : String(error)
+      }`
+    );
+  }
+
+  const format = String(metadata.format || "").toLowerCase();
+  const type = String(contentType || "").toLowerCase();
+  const cleanUrl = normalizedUrl.split("?")[0]?.toLowerCase() || "";
+  const isGif = format === "gif" || type.includes("gif") || cleanUrl.endsWith(".gif");
+  const isTooLarge = buffer.length > maxInputBytes;
+  const isTooWide = Number(metadata.width || 0) > ALIBABA_MAX_INPUT_SIDE;
+  const isTooTall = Number(metadata.height || 0) > ALIBABA_MAX_INPUT_SIDE;
+
+  if (!isGif && !isTooLarge && !isTooWide && !isTooTall) {
+    return normalizedUrl;
+  }
+
+  console.warn("[Alibaba Image] Normalizing input image before provider request:", {
+    bytes: buffer.length,
+    contentType,
+    format,
+    height: metadata.height,
+    isGif,
+    isTooLarge,
+    maxInputBytes,
+    modelId,
+    sourceHost: (() => {
+      try {
+        return new URL(normalizedUrl).host;
+      } catch {
+        return "";
+      }
+    })(),
+    width: metadata.width,
+  });
+
+  const basePipeline = sharp(buffer, {
+    animated: false,
+    failOn: "none",
+  })
+    .rotate()
+    .resize(ALIBABA_MAX_INPUT_SIDE, ALIBABA_MAX_INPUT_SIDE, {
+      fit: "inside",
+      withoutEnlargement: true,
+    });
+
+  let prepared = await basePipeline.png({ compressionLevel: 8 }).toBuffer();
+  let extension: "png" | "jpg" = "png";
+
+  if (prepared.length > maxInputBytes) {
+    prepared = await sharp(buffer, { animated: false, failOn: "none" })
+      .rotate()
+      .resize(ALIBABA_MAX_INPUT_SIDE, ALIBABA_MAX_INPUT_SIDE, {
+        fit: "inside",
+        withoutEnlargement: true,
+      })
+      .jpeg({ quality: 88, mozjpeg: true })
+      .toBuffer();
+    extension = "jpg";
+  }
+
+  if (prepared.length > maxInputBytes) {
+    prepared = await sharp(buffer, { animated: false, failOn: "none" })
+      .rotate()
+      .resize(3072, 3072, {
+        fit: "inside",
+        withoutEnlargement: true,
+      })
+      .jpeg({ quality: 82, mozjpeg: true })
+      .toBuffer();
+    extension = "jpg";
+  }
+
+  if (prepared.length > maxInputBytes) {
+    prepared = await sharp(buffer, { animated: false, failOn: "none" })
+      .rotate()
+      .resize(2048, 2048, {
+        fit: "inside",
+        withoutEnlargement: true,
+      })
+      .jpeg({ quality: 76, mozjpeg: true })
+      .toBuffer();
+    extension = "jpg";
+  }
+
+  if (prepared.length > maxInputBytes) {
+    const limitMb = Math.floor(maxInputBytes / 1024 / 1024);
+    throw new Error(`输入图片压缩后仍超过 ${limitMb}MB，请先压缩图片后重试`);
+  }
+
+  return uploadPreparedAlibabaInputImage(prepared, extension);
+}
+
 /**
  * 下载图片并上传到 S3
  */
@@ -229,12 +392,7 @@ async function downloadAndUploadToS3(
   try {
     console.log("[Alibaba Image] Downloading image from:", imageUrl.substring(0, 100) + "...");
     
-    const response = await fetch(imageUrl);
-    if (!response.ok) {
-      throw new Error(`Failed to download image: ${response.status}`);
-    }
-    
-    const buffer = Buffer.from(await response.arrayBuffer());
+    const { buffer } = await fetchImageBufferWithRetry(imageUrl, 30000, 3);
     console.log(`[Alibaba Image] Downloaded image, size: ${buffer.length} bytes`);
     
     // 创建缩略图
@@ -242,7 +400,7 @@ async function downloadAndUploadToS3(
     
     // 上传原图（Buffer 转 ArrayBuffer）
     const originalFilename = generateFilename(`${prefix}-orig`, "png");
-    const originalResult = await uploadFile(originalFilename, buffer.buffer.slice(buffer.byteOffset, buffer.byteOffset + buffer.byteLength));
+    const originalResult = await uploadFile(originalFilename, bufferToArrayBuffer(buffer));
     console.log(`[Alibaba Image] Uploaded original to: ${originalResult.url}`);
     
     // 上传缩略图
@@ -271,7 +429,7 @@ async function uploadBase64ToS3(base64Data: string): Promise<string> {
     const buffer = Buffer.from(base64Clean, "base64");
     
     const filename = generateFilename("alibaba-ref", "png");
-    const result = await uploadFile(filename, buffer.buffer.slice(buffer.byteOffset, buffer.byteOffset + buffer.byteLength));
+    const result = await uploadFile(filename, bufferToArrayBuffer(buffer));
     console.log(`[Alibaba Image] Uploaded reference image to: ${result.url}`);
     return result.url;
   } catch (error) {
@@ -513,8 +671,13 @@ export function createAlibabaImage({ apiKey, baseURL }: AlibabaImageConfig) {
             throw new Error("No prompt found in message");
           }
           
+          const preparedImageUrls: string[] = [];
+          for (const imageUrl of Array.from(new Set(imageUrls)).slice(0, 3)) {
+            preparedImageUrls.push(await prepareAlibabaInputImageUrl(imageUrl, activeModelId));
+          }
+
           // 验证：qwen-image-edit 系列模型必须有输入图片
-          if (imageUrls.length === 0) {
+          if (preparedImageUrls.length === 0) {
             const modelName = options.modelDisplayName || modelId;
             throw new Error(`「${modelName}」是图片编辑模型，需要至少上传一张图片才能进行编辑。如需纯文生图，请使用其他生图模型。`);
           }
@@ -574,7 +737,7 @@ export function createAlibabaImage({ apiKey, baseURL }: AlibabaImageConfig) {
             if (aspectRatio === "auto") {
               // 阿里文档在 auto 下默认接近 1024*1024，这会让 4K 选择失效。
               // 这里改为根据参考图比例计算 size，确保 1K/2K/4K 档位真正生效。
-              const referenceImageUrl = imageUrls[0];
+              const referenceImageUrl = preparedImageUrls[0];
               const inferredDimensions = await inferImageDimensionsFromImageUrl(referenceImageUrl);
               if (inferredDimensions) {
                 params.size = calculateAlibabaSizeFromDimensions(
@@ -613,10 +776,10 @@ export function createAlibabaImage({ apiKey, baseURL }: AlibabaImageConfig) {
           }
           
           // 根据图片数量确定编辑模式
-          const editMode = imageUrls.length === 1 ? "单图编辑" : `多图融合 (${imageUrls.length}张)`;
+          const editMode = preparedImageUrls.length === 1 ? "单图编辑" : `多图融合 (${preparedImageUrls.length}张)`;
           console.log("[Alibaba Image] 编辑模式:", editMode);
           console.log("[Alibaba Image] Prompt:", prompt.substring(0, 100) + (prompt.length > 100 ? "..." : ""));
-          console.log("[Alibaba Image] Input images:", imageUrls.length);
+          console.log("[Alibaba Image] Input images:", preparedImageUrls.length);
           console.log("[Alibaba Image] Final params:", params);
           
           // 发送请求
@@ -625,7 +788,7 @@ export function createAlibabaImage({ apiKey, baseURL }: AlibabaImageConfig) {
             apiKey,
             activeModelId,
             prompt,
-            imageUrls,
+            preparedImageUrls,
             params
           );
           
diff --git a/lib/ai/providers/fal-image.ts b/lib/ai/providers/fal-image.ts
index 2bf11349b..bee405a58 100644
--- a/lib/ai/providers/fal-image.ts
+++ b/lib/ai/providers/fal-image.ts
@@ -42,7 +42,7 @@ const DEFAULT_ASPECT_RATIO = "auto";
 const DEFAULT_RESOLUTION = "1K";
 const DEFAULT_OUTPUT_FORMAT = "png";
 const DEFAULT_SAFETY_TOLERANCE = "4";
-const DEFAULT_TIMEOUT_MS = 180000;
+const DEFAULT_TIMEOUT_MS = 300000;
 const FAL_NANO_BANANA_MODEL_IDS = new Set([
   "fal-ai/nano-banana",
   "fal-ai/nano-banana-2",
@@ -70,7 +70,10 @@ const FAL_ASPECT_RATIO_VALUES = new Set([
   "3:4",
   "2:3",
   "9:16",
-  "9:21",
+  "4:1",
+  "1:4",
+  "8:1",
+  "1:8",
 ]);
 const FAL_IMAGE_SIZE_PRESETS = new Set([
   "square_hd",
@@ -588,6 +591,34 @@ function summarizeFalError(error: any) {
   };
 }
 
+function stringifyFalErrorDetail(value: unknown) {
+  if (!value) return "";
+  if (typeof value === "string") return value;
+  try {
+    return JSON.stringify(value);
+  } catch {
+    return String(value);
+  }
+}
+
+function createDetailedFalError(error: unknown) {
+  const summary = summarizeFalError(error);
+  const detail = stringifyFalErrorDetail(summary.body);
+  const message = [
+    "FAL request failed",
+    summary.status ? `status=${summary.status}` : "",
+    summary.message ? `message=${summary.message}` : "",
+    detail ? `body=${detail}` : "",
+  ]
+    .filter(Boolean)
+    .join("; ")
+    .slice(0, 1800);
+  const detailedError = new Error(message);
+  detailedError.name = "FalImageProviderError";
+  (detailedError as Error & { cause?: unknown }).cause = error;
+  return detailedError;
+}
+
 function createFalClientFromApiKey(apiKey: string) {
   return import("@fal-ai/client").then(({ createFalClient }) =>
     createFalClient({
@@ -663,7 +694,9 @@ function buildFalRequestPayload(params: {
     };
 
     if (isFalNanoBananaModelId(falEndpointId)) {
-      input.aspect_ratio = aspectRatio;
+      input.aspect_ratio = FAL_ASPECT_RATIO_VALUES.has(aspectRatio)
+        ? aspectRatio
+        : DEFAULT_ASPECT_RATIO;
       input.resolution = resolution;
       input.limit_generations = true;
     } else if (isFalFluxKontextDevModelId(falEndpointId) && imageUrls.length > 0) {
@@ -698,6 +731,10 @@ function buildFalRequestPayload(params: {
     if (typeof params.imageParams?.enableWebSearch === "boolean") {
       input.enable_web_search = params.imageParams.enableWebSearch;
     }
+
+    if (typeof params.imageParams?.syncMode === "boolean") {
+      input.sync_mode = params.imageParams.syncMode;
+    }
   }
 
   return {
@@ -862,7 +899,7 @@ export function createFalImage({ apiKey, baseURL }: FalImageConfig) {
           });
         } catch (error) {
           console.error("[Fal Image] Request failed:", summarizeFalError(error));
-          throw error;
+          throw createDetailedFalError(error);
         }
 
         const text = await normalizeFalResultToMarkdown({
diff --git a/lib/ai/providers/gemini-image.ts b/lib/ai/providers/gemini-image.ts
index 426b79ce9..58030389f 100644
--- a/lib/ai/providers/gemini-image.ts
+++ b/lib/ai/providers/gemini-image.ts
@@ -798,39 +798,19 @@ async function sendGeminiRequest(
         ? await inferImageAspectRatioFromUrl(processedUrls[0])
         : null;
     const resolvedAspectRatio = inferredAspectRatio || requestedAspectRatio;
-    const normalizedRequestedAspectRatio = normalizeAspectRatio(requestedAspectRatio);
     const resolvedSize = resolveGrsaiGptImageSize(resolvedAspectRatio, imageParams?.imageSize);
-    const shouldPassAspectRatio =
-      Boolean(normalizedRequestedAspectRatio && normalizedRequestedAspectRatio !== "1:1");
 
-    if (normalizedRequestedAspectRatio === "1:1") {
-      logger.warn(
-        `${apiLogPrefix} Received 1:1 aspect ratio for GPT image on ${baseURL}. Skip aspect ratio params for compatibility.`,
-      );
-    }
-
-    // GPT-image（如 gpt-image-2-vip）在不同网关中参数差异较大：
-    // - 有的网关只识别 size；
-    // - 有的网关同时支持 aspectRatio/aspect_ratio；
-    // - 有的会读取 imageSize/image_size；
-    // 兼容不同网关：优先使用 OpenAI 原生 size；对容易兼容失败的比例（如 1:1）不再附带 aspect 变体参数。
+    // Grsai GPT-image-2-vip 只接受固定像素 size；继续传比例或档位字段会触发
+    // “does not support 1:1/3:4 parameter, please use pixel value input”。
     requestBody.size = resolvedSize;
-    requestBody.imageSize = imageParams?.imageSize;
-    requestBody.image_size = imageParams?.imageSize;
-
-    if (shouldPassAspectRatio) {
-      requestBody.aspectRatio = normalizedRequestedAspectRatio;
-      requestBody.aspect_ratio = normalizedRequestedAspectRatio;
-      requestBody.aspectRatioRaw = requestedAspectRatio;
-    }
     requestBody.shutProgress = false;
 
     if (processedUrls.length > 0) {
       requestBody.image = processedUrls[0];
-      requestBody.imageUrls = [processedUrls[0]];
+      requestBody.imageUrls = processedUrls;
     }
     logger.debug(
-      `${apiLogPrefix} GPT image payload mapping: requestedAspectRatio=${requestedAspectRatio}, normalizedRequestedAspectRatio=${normalizedRequestedAspectRatio || "n/a"}, resolvedAspectRatio=${resolvedAspectRatio}, imageSize=${imageParams?.imageSize || "2K"}, resolvedSize=${resolvedSize}`
+      `${apiLogPrefix} GPT image payload mapping: requestedAspectRatio=${requestedAspectRatio}, resolvedAspectRatio=${resolvedAspectRatio}, imageSize=${imageParams?.imageSize || "2K"}, resolvedSize=${resolvedSize}`
     );
     if (inferredAspectRatio) {
       logger.debug(`${apiLogPrefix} Inferred reference aspect ratio: ${inferredAspectRatio}`);
diff --git a/lib/ai/runtime/model-health.ts b/lib/ai/runtime/model-health.ts
index 73bb58feb..28a3a7a7a 100644
--- a/lib/ai/runtime/model-health.ts
+++ b/lib/ai/runtime/model-health.ts
@@ -90,7 +90,8 @@ export function classifyAiModelHealthFailure(error: unknown): {
     normalized.includes("model not exist") ||
     normalized.includes("model_not_found") ||
     normalized.includes("model not found") ||
-    normalized.includes("unknown model")
+    normalized.includes("unknown model") ||
+    normalized.includes("model name") && normalized.includes("not supported")
   ) {
     return {
       eventType: "invalid_model",
@@ -102,7 +103,11 @@ export function classifyAiModelHealthFailure(error: unknown): {
 
   if (
     normalized.includes("insufficient balance") ||
+    normalized.includes("credits insufficient") ||
     normalized.includes("quota") ||
+    normalized.includes("余额") ||
+    normalized.includes("额度") ||
+    normalized.includes("预扣费额度") ||
     normalized.includes("exceeded_current_quota") ||
     normalized.includes("pre_consume_token_quota_failed") ||
     normalized.includes("token quota is not enough")
@@ -117,6 +122,8 @@ export function classifyAiModelHealthFailure(error: unknown): {
 
   if (
     normalized.includes("timeout") ||
+    normalized.includes("超时") ||
+    normalized.includes("terminated") ||
     normalized.includes("timed out") ||
     normalized.includes("body timeout") ||
     normalized.includes("headers timeout")
@@ -140,7 +147,8 @@ export function classifyAiModelHealthFailure(error: unknown): {
     normalized.includes("no available channel") ||
     normalized.includes("no supported channel") ||
     normalized.includes("service unavailable") ||
-    normalized.includes("fetch failed")
+    normalized.includes("fetch failed") ||
+    normalized.includes("econnreset")
   ) {
     return {
       eventType: "unavailable",
diff --git a/lib/canvas/generate-image.ts b/lib/canvas/generate-image.ts
index d77c59cf3..e05ec6726 100644
--- a/lib/canvas/generate-image.ts
+++ b/lib/canvas/generate-image.ts
@@ -466,7 +466,10 @@ export async function generateCanvasImage(
   ).slice(0, 12);
   const { kept: reachableReferenceImages, dropped: unreachableReferenceImages } =
     await filterReachableRemoteUrls(normalizedReferenceImages, { timeoutMs: 3000 });
-  const effectiveReferenceImages = normalizedReferenceImages;
+  const effectiveReferenceImages =
+    reachableReferenceImages.length > 0
+      ? reachableReferenceImages
+      : normalizedReferenceImages;
   if (unreachableReferenceImages.length > 0) {
     console.warn("[Canvas Generate Image] Local reachability check failed for some reference images:", {
       count: unreachableReferenceImages.length,
@@ -857,6 +860,9 @@ export async function generateCanvasImage(
       imageSize,
       imageCount,
       requestedReferenceImageCount: normalizedReferenceImages.length,
+      reachableReferenceImageCount: reachableReferenceImages.length,
+      effectiveReferenceImageCount: effectiveReferenceImages.length,
+      droppedReferenceImageCount: unreachableReferenceImages.length,
       ...debugMeta,
     },
   }).catch((healthError) => {
diff --git a/lib/db/migrations/0086_external_image_log_model_delete_set_null.sql b/lib/db/migrations/0086_external_image_log_model_delete_set_null.sql
new file mode 100644
index 000000000..fce3da302
--- /dev/null
+++ b/lib/db/migrations/0086_external_image_log_model_delete_set_null.sql
@@ -0,0 +1,13 @@
+DO $$
+BEGIN
+  ALTER TABLE "ExternalImageApiRequestLog"
+    DROP CONSTRAINT IF EXISTS "ExternalImageApiRequestLog_modelDbId_AiModel_id_fk";
+
+  ALTER TABLE "ExternalImageApiRequestLog"
+    ADD CONSTRAINT "ExternalImageApiRequestLog_modelDbId_AiModel_id_fk"
+    FOREIGN KEY ("modelDbId")
+    REFERENCES "public"."AiModel"("id")
+    ON DELETE SET NULL
+    ON UPDATE NO ACTION;
+END
+$$;
diff --git a/lib/db/migrations/meta/_journal.json b/lib/db/migrations/meta/_journal.json
index 22ee22139..236089b9d 100644
--- a/lib/db/migrations/meta/_journal.json
+++ b/lib/db/migrations/meta/_journal.json
@@ -505,6 +505,13 @@
       "when": 1780075200000,
       "tag": "0085_add_system_setting_snapshots",
       "breakpoints": true
+    },
+    {
+      "idx": 72,
+      "version": "7",
+      "when": 1781174400000,
+      "tag": "0086_external_image_log_model_delete_set_null",
+      "breakpoints": true
     }
   ]
 }
diff --git a/lib/db/queries.ts b/lib/db/queries.ts
index ddd37cf44..c2e03fcd5 100644
--- a/lib/db/queries.ts
+++ b/lib/db/queries.ts
@@ -74,6 +74,7 @@ import {
   type MediaAsset,
   mediaAssetRelation,
   type MediaAssetRelation,
+  externalImageApiRequestLog,
   type AgentPreset,
   type AgentPresetCategory,
   type AgentPresetExecutionSnapshot,
@@ -4354,8 +4355,54 @@ export async function updateAiModel({
 
 export async function deleteAiModel({ id }: { id: string }) {
   try {
-    return await db.delete(aiModel).where(eq(aiModel.id, id)).returning();
+    return await db.transaction(async (tx) => {
+      const [modelToDelete] = await tx
+        .select({
+          id: aiModel.id,
+          modelId: aiModel.modelId,
+        })
+        .from(aiModel)
+        .where(eq(aiModel.id, id))
+        .limit(1);
+
+      if (!modelToDelete) {
+        return [];
+      }
+
+      await tx
+        .update(externalImageApiRequestLog)
+        .set({ modelDbId: null })
+        .where(eq(externalImageApiRequestLog.modelDbId, id));
+
+      await tx
+        .update(aiModel)
+        .set({ parentModelId: null })
+        .where(eq(aiModel.parentModelId, id));
+
+      await tx
+        .update(aiModel)
+        .set({ fallbackModelId1: null })
+        .where(
+          inArray(aiModel.fallbackModelId1, [
+            modelToDelete.id,
+            modelToDelete.modelId,
+          ])
+        );
+
+      await tx
+        .update(aiModel)
+        .set({ fallbackModelId2: null })
+        .where(
+          inArray(aiModel.fallbackModelId2, [
+            modelToDelete.id,
+            modelToDelete.modelId,
+          ])
+        );
+
+      return await tx.delete(aiModel).where(eq(aiModel.id, id)).returning();
+    });
   } catch (_error) {
+    console.error("deleteAiModel error:", _error);
     throw new ChatSDKError(
       "bad_request:database",
       "Failed to delete AI model"
diff --git a/lib/db/schema.ts b/lib/db/schema.ts
index 406520003..2acecf1d8 100644
--- a/lib/db/schema.ts
+++ b/lib/db/schema.ts
@@ -600,7 +600,9 @@ export const externalImageApiRequestLog = pgTable("ExternalImageApiRequestLog",
   apiKeyId: varchar("apiKeyId", { length: 64 }),
   apiKeyName: varchar("apiKeyName", { length: 128 }),
   ownerUserId: uuid("ownerUserId").references(() => user.id),
-  modelDbId: uuid("modelDbId").references(() => aiModel.id),
+  modelDbId: uuid("modelDbId").references(() => aiModel.id, {
+    onDelete: "set null",
+  }),
   modelRuntimeId: varchar("modelRuntimeId", { length: 128 }),
   modelDisplayName: varchar("modelDisplayName", { length: 128 }),
   prompt: text("prompt"),