diff --git a/app/api/v1/admin.py b/app/api/v1/admin.py
index 5163485..22c2b69 100644
--- a/app/api/v1/admin.py
+++ b/app/api/v1/admin.py
@@ -1,6 +1,6 @@
 from fastapi import APIRouter, Depends, HTTPException, Request, Query
 from fastapi.responses import HTMLResponse, StreamingResponse
-from app.core.auth import verify_api_key, verify_app_key
+from app.core.auth import verify_api_key, verify_app_key, get_admin_api_key
 from app.core.config import config, get_config
 from app.core.batch_tasks import create_task, get_task, expire_task
 from app.core.storage import get_storage, LocalStorage, RedisStorage, SQLStorage
@@ -33,7 +33,7 @@ def _sse_event(payload: dict) -> str:
 
 
 def _verify_stream_api_key(request: Request) -> None:
-    api_key = get_config("app.api_key", "")
+    api_key = get_admin_api_key()
     if not api_key:
         return
     key = request.query_params.get("api_key")
@@ -110,7 +110,7 @@ async def admin_token_page():
 @router.post("/api/v1/admin/login", dependencies=[Depends(verify_app_key)])
 async def admin_login_api():
     """管理后台登录验证（使用 app_key）"""
-    return {"status": "success", "api_key": get_config("app.api_key", "")}
+    return {"status": "success", "api_key": get_admin_api_key()}
 
 
 @router.get("/api/v1/admin/config", dependencies=[Depends(verify_api_key)])
diff --git a/app/api/v1/chat.py b/app/api/v1/chat.py
index 4866663..69db305 100644
--- a/app/api/v1/chat.py
+++ b/app/api/v1/chat.py
@@ -48,8 +48,8 @@ class VideoConfig(BaseModel):
     aspect_ratio: Optional[str] = Field(
         "3:2", description="视频比例: 3:2, 16:9, 1:1 等"
     )
-    video_length: Optional[int] = Field(6, description="视频时长(秒): 5-15")
-    resolution: Optional[str] = Field("SD", description="视频分辨率: SD, HD")
+    video_length: Optional[int] = Field(6, description="视频时长(秒): 6 或 10")
+    resolution_name: Optional[str] = Field("480p", description="视频分辨率: 480p, 720p")
     preset: Optional[str] = Field("custom", description="风格预设: fun, normal, spicy")
 
     @field_validator("aspect_ratio")
@@ -68,22 +68,22 @@ class VideoConfig(BaseModel):
     @classmethod
     def validate_video_length(cls, v):
         if v is not None:
-            if v < 5 or v > 15:
+            if v not in (6, 10):
                 raise ValidationException(
-                    message="video_length must be between 5 and 15 seconds",
+                    message="video_length must be 6 or 10 seconds",
                     param="video_config.video_length",
                     code="invalid_video_length",
                 )
         return v
 
-    @field_validator("resolution")
+    @field_validator("resolution_name")
     @classmethod
     def validate_resolution(cls, v):
-        allowed = ["SD", "HD"]
+        allowed = ["480p", "720p"]
         if v and v not in allowed:
             raise ValidationException(
-                message=f"resolution must be one of {allowed}",
-                param="video_config.resolution",
+                message=f"resolution_name must be one of {allowed}",
+                param="video_config.resolution_name",
                 code="invalid_resolution",
             )
         return v
@@ -272,7 +272,7 @@ async def chat_completions(request: ChatCompletionRequest):
             thinking=request.thinking,
             aspect_ratio=v_conf.aspect_ratio,
             video_length=v_conf.video_length,
-            resolution=v_conf.resolution,
+            resolution=v_conf.resolution_name,
             preset=v_conf.preset,
         )
     else:
diff --git a/app/core/auth.py b/app/core/auth.py
index 7c96460..a4c8735 100644
--- a/app/core/auth.py
+++ b/app/core/auth.py
@@ -16,15 +16,27 @@ security = HTTPBearer(
 )
 
 
+def get_admin_api_key() -> str:
+    """
+    获取后台 API Key。
+
+    为空时表示不启用后台接口认证。
+    """
+    api_key = get_config("app.api_key", "")
+    return api_key or ""
+
+
 async def verify_api_key(
     auth: Optional[HTTPAuthorizationCredentials] = Security(security),
 ) -> Optional[str]:
     """
     验证 Bearer Token
 
-    如果 config.toml 中未配置 api_key，则使用默认密钥 "grok2api"。
+    如果 config.toml 中未配置 api_key，则不启用认证。
     """
-    api_key = get_config("app.api_key", "grok2api")
+    api_key = get_admin_api_key()
+    if not api_key:
+        return None
 
     if not auth:
         raise HTTPException(
diff --git a/app/services/grok/media.py b/app/services/grok/media.py
index 8c5159e..e6c6e1f 100644
--- a/app/services/grok/media.py
+++ b/app/services/grok/media.py
@@ -177,7 +177,7 @@ class VideoService:
         post_id: str,
         aspect_ratio: str = "3:2",
         video_length: int = 6,
-        resolution: str = "SD",
+        resolution_name: str = "480p",
         preset: str = "normal",
     ) -> dict:
         """构建视频生成载荷"""
@@ -190,7 +190,12 @@ class VideoService:
             mode_flag = "--mode=extremely-spicy-or-crazy"
 
         full_prompt = f"{prompt} {mode_flag}"
-
+        video_gen_config = {
+            "parentPostId": post_id,
+            "aspectRatio": aspect_ratio,
+            "videoLength": video_length,
+            "resolutionName": resolution_name,
+        }
         return {
             "temporary": True,
             "modelName": "grok-3",
@@ -201,12 +206,7 @@ class VideoService:
                 "experiments": [],
                 "modelConfigOverride": {
                     "modelMap": {
-                        "videoGenModelConfig": {
-                            "parentPostId": post_id,
-                            "aspectRatio": aspect_ratio,
-                            "videoLength": video_length,
-                            "videoResolution": resolution,
-                        }
+                        "videoGenModelConfig": video_gen_config
                     }
                 },
             },
@@ -218,7 +218,7 @@ class VideoService:
         prompt: str,
         aspect_ratio: str = "3:2",
         video_length: int = 6,
-        resolution: str = "SD",
+        resolution_name: str = "480p",
         stream: bool = True,
         preset: str = "normal",
     ) -> AsyncGenerator[bytes, None]:
@@ -230,7 +230,7 @@ class VideoService:
             prompt: 视频描述
             aspect_ratio: 宽高比
             video_length: 视频时长
-            resolution: 分辨率
+            resolution_name: 分辨率
             stream: 是否流式
             preset: 预设
 
@@ -249,7 +249,7 @@ class VideoService:
                 # Step 2: 建立连接
                 headers = self._build_headers(token)
                 payload = self._build_payload(
-                    prompt, post_id, aspect_ratio, video_length, resolution, preset
+                    prompt, post_id, aspect_ratio, video_length, resolution_name, preset
                 )
 
                 session = AsyncSession(impersonate=BROWSER)
@@ -302,7 +302,7 @@ class VideoService:
         image_url: str,
         aspect_ratio: str = "3:2",
         video_length: int = 6,
-        resolution: str = "SD",
+        resolution: str = "480p",
         stream: bool = True,
         preset: str = "normal",
     ) -> AsyncGenerator[bytes, None]:
@@ -418,7 +418,7 @@ class VideoService:
         thinking: str = None,
         aspect_ratio: str = "3:2",
         video_length: int = 6,
-        resolution: str = "SD",
+        resolution: str = "480p",
         preset: str = "normal",
     ):
         """
diff --git a/app/services/grok/model.py b/app/services/grok/model.py
index 9f5475c..76a9664 100644
--- a/app/services/grok/model.py
+++ b/app/services/grok/model.py
@@ -85,11 +85,18 @@ class ModelService:
             display_name="Grok 4 Heavy",
         ),
         ModelInfo(
-            model_id="grok-4.1",
+            model_id="grok-4.1-fast",
             grok_model="grok-4-1-thinking-1129",
-            model_mode="MODEL_MODE_AUTO",
+            model_mode="MODEL_MODE_FAST",
             cost=Cost.LOW,
-            display_name="Grok 4.1",
+            display_name="Grok 4.1 Fast",
+        ),
+        ModelInfo(
+            model_id="grok-4.1-expert",
+            grok_model="grok-4-1-thinking-1129",
+            model_mode="MODEL_MODE_EXPERT",
+            cost=Cost.LOW,
+            display_name="Grok 4.1 Expert",
         ),
         ModelInfo(
             model_id="grok-4.1-thinking",
diff --git a/app/services/grok/nsfw.py b/app/services/grok/nsfw.py
index 2f6a625..e3ac3b5 100644
--- a/app/services/grok/nsfw.py
+++ b/app/services/grok/nsfw.py
@@ -62,14 +62,24 @@ class NSFWService:
     @staticmethod
     def _build_payload() -> bytes:
         """构造请求 payload"""
-        # protobuf: enable_unhinged=true (field1=1, field2=1)
-        protobuf = bytes([0x08, 0x01, 0x10, 0x01])
+        # protobuf (match captured HAR):
+        # 0a 02 10 01                   -> field 1 (len=2) with inner bool=true
+        # 12 1a                         -> field 2, length 26
+        #   0a 18 <name>                -> nested message with name string
+        name = b"always_show_nsfw_content"
+        inner = b"\x0a" + bytes([len(name)]) + name
+        protobuf = b"\x0a\x02\x10\x01\x12" + bytes([len(inner)]) + inner
         return encode_grpc_web_payload(protobuf)
 
     async def enable(self, token: str) -> NSFWResult:
         """为单个 token 开启 NSFW 模式"""
         headers = self._build_headers(token)
         payload = self._build_payload()
+        logger.debug(
+            "NSFW payload: len={} hex={}",
+            len(payload),
+            payload.hex(),
+        )
         proxies = {"http": self.proxy, "https": self.proxy} if self.proxy else None
 
         try:
@@ -96,6 +106,13 @@ class NSFWService:
                 )
 
                 grpc_status = get_grpc_status(trailers)
+                logger.debug(
+                    "NSFW response: http={} grpc={} msg={} trailers={}",
+                    response.status_code,
+                    grpc_status.code,
+                    grpc_status.message,
+                    trailers,
+                )
 
                 # HTTP 200 且无 grpc-status（空响应）或 grpc-status=0 都算成功
                 success = grpc_status.code == -1 or grpc_status.ok
diff --git a/app/services/grok/processor.py b/app/services/grok/processor.py
index 7c47c69..8f0847b 100644
--- a/app/services/grok/processor.py
+++ b/app/services/grok/processor.py
@@ -498,7 +498,7 @@ class VideoStreamProcessor(BaseProcessor):
         self.response_id: Optional[str] = None
         self.think_opened: bool = False
         self.role_sent: bool = False
-        self.video_format = get_config("app.video_format", "url")
+        self.video_format = str(get_config("app.video_format", "html")).lower()
 
         if think is None:
             self.show_think = get_config("grok.thinking", False)
@@ -568,10 +568,13 @@ class VideoStreamProcessor(BaseProcessor):
                                     thumbnail_url, "image"
                                 )
 
-                            video_html = self._build_video_html(
-                                final_video_url, final_thumbnail_url
-                            )
-                            yield self._sse(video_html)
+                            if self.video_format == "url":
+                                yield self._sse(final_video_url)
+                            else:
+                                video_html = self._build_video_html(
+                                    final_video_url, final_thumbnail_url
+                                )
+                                yield self._sse(video_html)
 
                             logger.info(f"Video generated: {video_url}")
                     continue
@@ -626,7 +629,7 @@ class VideoCollectProcessor(BaseProcessor):
 
     def __init__(self, model: str, token: str = ""):
         super().__init__(model, token)
-        self.video_format = get_config("app.video_format", "url")
+        self.video_format = str(get_config("app.video_format", "html")).lower()
 
     def _build_video_html(self, video_url: str, thumbnail_url: str = "") -> str:
         poster_attr = f' poster="{thumbnail_url}"' if thumbnail_url else ""
@@ -666,9 +669,12 @@ class VideoCollectProcessor(BaseProcessor):
                                     thumbnail_url, "image"
                                 )
 
-                            content = self._build_video_html(
-                                final_video_url, final_thumbnail_url
-                            )
+                            if self.video_format == "url":
+                                content = final_video_url
+                            else:
+                                content = self._build_video_html(
+                                    final_video_url, final_thumbnail_url
+                                )
                             logger.info(f"Video generated: {video_url}")
 
         except asyncio.CancelledError:
diff --git a/app/static/config/config.js b/app/static/config/config.js
index 56097cf..bef6667 100644
--- a/app/static/config/config.js
+++ b/app/static/config/config.js
@@ -4,6 +4,10 @@ const byId = (id) => document.getElementById(id);
 const NUMERIC_FIELDS = new Set([
   'timeout',
   'max_retry',
+  'retry_backoff_base',
+  'retry_backoff_factor',
+  'retry_backoff_max',
+  'retry_budget',
   'refresh_interval_hours',
   'fail_threshold',
   'limit_mb',
@@ -17,6 +21,8 @@ const NUMERIC_FIELDS = new Set([
   'usage_batch_size',
   'usage_max_tokens',
   'reload_interval_sec',
+  'stream_idle_timeout',
+  'video_idle_timeout',
   'nsfw_max_concurrent',
   'nsfw_batch_size',
   'nsfw_max_tokens'
@@ -29,7 +35,7 @@ const LOCALE_MAP = {
     "app_key": { title: "后台密码", desc: "登录 Grok2API 服务管理后台的密码，请妥善保管。" },
     "app_url": { title: "应用地址", desc: "当前 Grok2API 服务的外部访问 URL，用于文件链接访问。" },
     "image_format": { title: "图片格式", desc: "生成的图片格式（url 或 base64）。" },
-    "video_format": { title: "视频格式", desc: "生成的视频格式（仅支持 url）。" }
+    "video_format": { title: "视频格式", desc: "生成的视频格式（html 或 url，url 为处理后的链接）。" }
   },
   "grok": {
     "label": "Grok 设置",
@@ -43,7 +49,13 @@ const LOCALE_MAP = {
     "asset_proxy_url": { title: "资源代理 URL", desc: "代理请求到 Grok 官网的静态资源（图片/视频）地址。" },
     "cf_clearance": { title: "CF Clearance", desc: "Cloudflare 验证 Cookie，用于验证 Cloudflare 的验证。" },
     "max_retry": { title: "最大重试", desc: "请求 Grok 服务失败时的最大重试次数。" },
-    "retry_status_codes": { title: "重试状态码", desc: "触发重试的 HTTP 状态码列表。" }
+    "retry_status_codes": { title: "重试状态码", desc: "触发重试的 HTTP 状态码列表。" },
+    "retry_backoff_base": { title: "退避基数", desc: "重试退避的基础延迟（秒）。" },
+    "retry_backoff_factor": { title: "退避倍率", desc: "重试退避的指数放大系数。" },
+    "retry_backoff_max": { title: "退避上限", desc: "单次重试等待的最大延迟（秒）。" },
+    "retry_budget": { title: "退避预算", desc: "单次请求的最大重试总耗时（秒）。" },
+    "stream_idle_timeout": { title: "流空闲超时", desc: "流式响应空闲超时（秒），超过将断开。" },
+    "video_idle_timeout": { title: "视频空闲超时", desc: "视频生成空闲超时（秒），超过将断开。" }
   },
   "token": {
     "label": "Token 池设置",
@@ -268,7 +280,8 @@ function renderConfig(data) {
         ]);
       }
       else if (key === 'video_format') {
-        built = buildSelectInput(section, key, 'url', [
+        built = buildSelectInput(section, key, val, [
+          { val: 'html', text: 'HTML' },
           { val: 'url', text: 'URL' }
         ]);
       }
diff --git a/config.defaults.toml b/config.defaults.toml
index a6ddbea..79a9e27 100644
--- a/config.defaults.toml
+++ b/config.defaults.toml
@@ -24,7 +24,7 @@ app_url = "http://127.0.0.1:8000"
 app_key = "grok2api"
 api_key = ""
 image_format = "url"
-video_format = "url"
+video_format = "html"
 
 [token]
 auto_refresh = true
diff --git a/data/config.toml b/data/config.toml
index da8be8e..82eee63 100644
--- a/data/config.toml
+++ b/data/config.toml
@@ -10,6 +10,12 @@ asset_proxy_url = ""
 cf_clearance = ""
 max_retry = 3
 retry_status_codes = [401,429,403]
+retry_backoff_base = 0.5
+retry_backoff_factor = 2.0
+retry_backoff_max = 30.0
+retry_budget = 90.0
+stream_idle_timeout = 45.0
+video_idle_timeout = 90.0
 
 [app]
 app_url = "http://127.0.0.1:8000"
diff --git a/docs/README.en.md b/docs/README.en.md
index f5cb201..39d2ff3 100644
--- a/docs/README.en.md
+++ b/docs/README.en.md
@@ -99,8 +99,8 @@ curl http://localhost:8000/v1/chat/completions \
 | `thinking` | string | Thinking mode | `enabled`, `disabled`, `null` |
 | `video_config` | object | **Video model only** | - |
 | └─ `aspect_ratio` | string | Video aspect ratio | `16:9`, `9:16`, `1:1`, `2:3`, `3:2` |
-| └─ `video_length` | integer | Video length (seconds) | `5` - `15` |
-| └─ `resolution` | string | Resolution | `SD`, `HD` |
+| └─ `video_length` | integer | Video length (seconds) | `6`, `10` |
+| └─ `resolution_name` | string | Resolution | `480p`, `720p` |
 | └─ `preset` | string | Style preset | `fun`, `normal`, `spicy` |
 
 Note: any other parameters will be discarded and ignored.
@@ -157,7 +157,7 @@ Config file: `data/config.toml`
 | | `app_key` | Admin password | Password for the Grok2API admin panel. | `grok2api` |
 | | `api_key` | API key | Bearer token required to call Grok2API. | `""` |
 | | `image_format` | Image format | Output image format (`url` or `base64`). | `url` |
-| | `video_format` | Video format | Output video format (url only). | `url` |
+| | `video_format` | Video format | Output video format (html tag or processed url). | `html` |
 | **grok** | `temporary` | Temporary chat | Enable temporary conversation mode. | `true` |
 | | `stream` | Streaming | Enable streaming by default. | `true` |
 | | `thinking` | Thinking chain | Enable model thinking output. | `true` |
@@ -169,6 +169,12 @@ Config file: `data/config.toml`
 | | `cf_clearance` | CF Clearance | Cloudflare clearance cookie for verification. | `""` |
 | | `max_retry` | Max retries | Max retries on Grok request failure. | `3` |
 | | `retry_status_codes` | Retry status codes | HTTP status codes that trigger retry. | `[401, 429, 403]` |
+| | `retry_backoff_base` | Backoff base | Base delay for retry backoff (seconds). | `0.5` |
+| | `retry_backoff_factor` | Backoff factor | Exponential multiplier for retry backoff. | `2.0` |
+| | `retry_backoff_max` | Backoff max | Max wait per retry (seconds). | `30.0` |
+| | `retry_budget` | Backoff budget | Max total retry time per request (seconds). | `90.0` |
+| | `stream_idle_timeout` | Stream idle timeout | Idle timeout for streaming responses (seconds). | `45.0` |
+| | `video_idle_timeout` | Video idle timeout | Idle timeout for video generation (seconds). | `90.0` |
 | **token** | `auto_refresh` | Auto refresh | Enable automatic token refresh. | `true` |
 | | `refresh_interval_hours` | Refresh interval | Token refresh interval (hours). | `8` |
 | | `fail_threshold` | Failure threshold | Consecutive failures before a token is disabled. | `5` |
diff --git a/readme.md b/readme.md
index 976ffa2..71b00ea 100644
--- a/readme.md
+++ b/readme.md
@@ -112,8 +112,8 @@ curl http://localhost:8000/v1/chat/completions \
 | `thinking`           | string  | 思维链模式                     | `enabled`, `disabled`, `null`                      |
 | `video_config`       | object  | **视频模型专用配置对象**       | -                                                  |
 | └─`aspect_ratio`     | string  | 视频宽高比                     | `16:9`, `9:16`, `1:1`, `2:3`, `3:2`                |
-| └─`video_length`     | integer | 视频时长 (秒)                  | `5` - `15`                                         |
-| └─`resolution`       | string  | 分辨率                         | `SD`, `HD`                                         |
+| └─`video_length`     | integer | 视频时长 (秒)                  | `6`, `10`                                          |
+| └─`resolution_name`  | string  | 分辨率                         | `480p`, `720p`                                     |
 | └─`preset`           | string  | 风格预设                       | `fun`, `normal`, `spicy`, `custom`                 |
 
 **消息格式 (messages)**：
@@ -190,7 +190,7 @@ curl http://localhost:8000/v1/images/generations \
 |                       | `app_key`                  | 后台密码     | 登录 Grok2API 服务管理后台的密码，请妥善保管。       | `grok2api`                                              |
 |                       | `api_key`                  | API 密钥     | 调用 Grok2API 服务所需的 Bearer Token，请妥善保管。  | `grok2api`                                              |
 |                       | `image_format`             | 图片格式     | 生成的图片格式（url 或 base64）。                    | `url`                                                   |
-|                       | `video_format`             | 视频格式     | 生成的视频格式（仅支持 url）。                       | `url`                                                   |
+|                       | `video_format`             | 视频格式     | 生成的视频格式（html 或 url，url 为处理后的链接）。  | `html`                                                  |
 | **grok**        | `temporary`                | 临时对话     | 是否启用临时对话模式。                               | `true`                                                  |
 |                       | `stream`                   | 流式响应     | 是否默认启用流式输出。                               | `true`                                                  |
 |                       | `thinking`                 | 思维链       | 是否启用模型思维链输出。                             | `true`                                                  |
@@ -202,6 +202,12 @@ curl http://localhost:8000/v1/images/generations \
 |                       | `cf_clearance`             | CF Clearance | Cloudflare 验证 Cookie，用于验证 Cloudflare 的验证。 | `""`                                                    |
 |                       | `max_retry`                | 最大重试     | 请求 Grok 服务失败时的最大重试次数。                 | `3`                                                     |
 |                       | `retry_status_codes`       | 重试状态码   | 触发重试的 HTTP 状态码列表。                         | `[401, 429, 403]`                                       |
+|                       | `retry_backoff_base`       | 退避基数     | 重试退避的基础延迟（秒）。                           | `0.5`                                                   |
+|                       | `retry_backoff_factor`     | 退避倍率     | 重试退避的指数放大系数。                             | `2.0`                                                   |
+|                       | `retry_backoff_max`        | 退避上限     | 单次重试等待的最大延迟（秒）。                       | `30.0`                                                  |
+|                       | `retry_budget`             | 退避预算     | 单次请求的最大重试总耗时（秒）。                     | `90.0`                                                  |
+|                       | `stream_idle_timeout`      | 流空闲超时   | 流式响应空闲超时（秒），超过将断开。                 | `45.0`                                                  |
+|                       | `video_idle_timeout`       | 视频空闲超时 | 视频生成空闲超时（秒），超过将断开。                 | `90.0`                                                  |
 | **token**       | `auto_refresh`             | 自动刷新     | 是否开启 Token 自动刷新机制。                        | `true`                                                  |
 |                       | `refresh_interval_hours`   | 刷新间隔     | Token 刷新的时间间隔（小时）。                       | `8`                                                     |
 |                       | `fail_threshold`           | 失败阈值     | 单个 Token 连续失败多少次后被标记为不可用。          | `5`                                                     |