diff --git a/astrbot/core/agent/message.py b/astrbot/core/agent/message.py
index 4292f4c04e6..94e457ca69c 100644
--- a/astrbot/core/agent/message.py
+++ b/astrbot/core/agent/message.py
@@ -21,7 +21,7 @@ class ContentPart(BaseModel):
 
     __content_part_registry: ClassVar[dict[str, type["ContentPart"]]] = {}
 
-    type: Literal["text", "think", "image_url", "audio_url"]
+    type: Literal["text", "think", "image_url", "audio_url", "video_url"]
     _no_save: bool = PrivateAttr(default=False)
 
     def __init_subclass__(cls, **kwargs: Any) -> None:
@@ -141,6 +141,22 @@ class AudioURL(BaseModel):
     audio_url: AudioURL
 
 
+class VideoURLPart(ContentPart):
+    """
+    >>> VideoURLPart(video_url=VideoURLPart.VideoURL(url="https://example.com/clip.mp4")).model_dump()
+    {'type': 'video_url', 'video_url': {'url': 'https://example.com/clip.mp4', 'id': None}}
+    """
+
+    class VideoURL(BaseModel):
+        url: str
+        """The URL of the video, can be data URI scheme like `data:video/mp4;base64,...`."""
+        id: str | None = None
+        """The ID of the video, to allow LLMs to distinguish different videos."""
+
+    type: str = "video_url"
+    video_url: VideoURL
+
+
 class ToolCall(BaseModel):
     """
     A tool call requested by the assistant.
diff --git a/astrbot/core/agent/runners/tool_loop_agent_runner.py b/astrbot/core/agent/runners/tool_loop_agent_runner.py
index f8418fe4205..0da391d6d9d 100644
--- a/astrbot/core/agent/runners/tool_loop_agent_runner.py
+++ b/astrbot/core/agent/runners/tool_loop_agent_runner.py
@@ -42,6 +42,7 @@
     ToolCallsResult,
 )
 from astrbot.core.provider.modalities import (
+    has_video_blocks,
     log_context_sanitize_stats,
     sanitize_contexts_by_modalities,
 )
@@ -317,6 +318,7 @@ async def reset(
             request.prompt is not None
             or request.image_urls
             or request.audio_urls
+            or request.video_urls
             or request.extra_user_content_parts
         ):
             m = await self._assemble_request_context_for_provider(request)
@@ -342,17 +344,26 @@ async def _assemble_request_context_for_provider(
     ) -> dict[str, T.Any]:
         modalities = self.provider.provider_config.get("modalities", None)
         if not modalities:  # Unconfigured (None or empty list) defaults to support all modalities for backward compatibility
-            return await request.assemble_context()
+            # Video is excluded from that default. It postdates the "supports
+            # everything" convention and is not universally supported, so a
+            # provider that never opted in would fail the whole request rather
+            # than degrade. Video is therefore strictly opt-in via the
+            # `modalities` list.
+            if not request.video_urls:
+                return await request.assemble_context()
+            return await replace(request, video_urls=[]).assemble_context()
 
         supports_image = "image" in modalities
         supports_audio = "audio" in modalities
-        if supports_image and supports_audio:
+        supports_video = "video" in modalities
+        if supports_image and supports_audio and supports_video:
             return await request.assemble_context()
 
         adjusted_request = replace(
             request,
             image_urls=request.image_urls if supports_image else [],
             audio_urls=request.audio_urls if supports_audio else [],
+            video_urls=request.video_urls if supports_video else [],
         )
         context = await adjusted_request.assemble_context()
         content = context.get("content")
@@ -369,6 +380,9 @@ async def _assemble_request_context_for_provider(
         if not supports_audio:
             for _ in request.audio_urls:
                 content_blocks.append({"type": "text", "text": "[Audio]"})
+        if not supports_video:
+            for _ in request.video_urls:
+                content_blocks.append({"type": "text", "text": "[Video]"})
 
         return {"role": "user", "content": content_blocks}
 
@@ -663,13 +677,17 @@ def _sanitize_contexts_for_provider(
         contexts: list[Message] | list[dict[str, T.Any]],
     ) -> list[Message] | list[dict[str, T.Any]]:
         modalities = self.provider.provider_config.get("modalities", None)
-        if (
-            not modalities
-        ):  # Unconfigured (None or empty list) defaults to support all modalities
+        if not modalities and not has_video_blocks(contexts):
+            # Unconfigured (None or empty list) defaults to supporting every
+            # modality, so the context objects are handed back untouched. Video
+            # is the exception: it is strictly opt-in (see
+            # `_assemble_request_context_for_provider`), and this path also runs
+            # when a request falls back to another provider, whose context was
+            # assembled for the primary one and may already carry a `video_url`.
             return contexts
         sanitized_contexts, stats = sanitize_contexts_by_modalities(
             contexts,
-            self.provider.provider_config.get("modalities", None),
+            modalities,
         )
         log_context_sanitize_stats(stats)
         return sanitized_contexts
diff --git a/astrbot/core/astr_main_agent.py b/astrbot/core/astr_main_agent.py
index 59183b8a613..a5390e387ca 100644
--- a/astrbot/core/astr_main_agent.py
+++ b/astrbot/core/astr_main_agent.py
@@ -877,6 +877,10 @@ async def _append_video_attachment(
     else:
         text = f"[Video Attachment: name {video_name}, path {video_path}]"
 
+    # Collect the resolved path so providers that declare the `video` modality
+    # receive the video itself. The text part above is kept for every other
+    # provider, and for the persisted history where the video block is dropped.
+    req.video_urls.append(video_path)
     req.extra_user_content_parts.append(TextPart(text=text))
 
 
@@ -1821,6 +1825,7 @@ async def build_main_agent(
         )
     req.image_urls = normalize_and_dedupe_strings(req.image_urls)
     req.audio_urls = normalize_and_dedupe_strings(req.audio_urls)
+    req.video_urls = normalize_and_dedupe_strings(req.video_urls)
 
     if config.file_extract_enabled:
         try:
diff --git a/astrbot/core/config/default.py b/astrbot/core/config/default.py
index 150804573a8..2a8abdad8f9 100644
--- a/astrbot/core/config/default.py
+++ b/astrbot/core/config/default.py
@@ -2248,8 +2248,8 @@ def get_local_permission_defaults(system: str | None = None) -> dict:
                         "description": "模型能力",
                         "type": "list",
                         "items": {"type": "string"},
-                        "options": ["text", "image", "audio", "tool_use"],
-                        "labels": ["文本", "图像", "音频", "工具使用"],
+                        "options": ["text", "image", "audio", "video", "tool_use"],
+                        "labels": ["文本", "图像", "音频", "视频", "工具使用"],
                         "render_type": "checkbox",
                         "hint": "模型支持的模态及能力。",
                     },
diff --git a/astrbot/core/provider/entities.py b/astrbot/core/provider/entities.py
index 382932b8579..2b7fd42cd85 100644
--- a/astrbot/core/provider/entities.py
+++ b/astrbot/core/provider/entities.py
@@ -101,6 +101,8 @@ class ProviderRequest:
     """图片 URL 列表"""
     audio_urls: list[str] = field(default_factory=list)
     """音频 URL 列表，也支持本地路径"""
+    video_urls: list[str] = field(default_factory=list)
+    """Video URL list. Local paths are supported as well."""
     extra_user_content_parts: list[ContentPart] = field(default_factory=list)
     """额外的用户消息内容部分列表，用于在用户消息后添加额外的内容块（如系统提醒、指令等）。支持 dict 或 ContentPart 对象"""
     func_tool: ToolSet | None = None
@@ -124,6 +126,7 @@ def __repr__(self) -> str:
             f"ProviderRequest(prompt={self.prompt}, session_id={self.session_id}, "
             f"image_count={len(self.image_urls or [])}, "
             f"audio_count={len(self.audio_urls or [])}, "
+            f"video_count={len(self.video_urls or [])}, "
             f"func_tool={self.func_tool}, "
             f"contexts={self._print_friendly_context()}, "
             f"system_prompt={self.system_prompt}, "
@@ -146,7 +149,8 @@ def _print_friendly_context(self):
         if not self.contexts:
             return (
                 f"prompt: {self.prompt}, image_count: {len(self.image_urls or [])}, "
-                f"audio_count: {len(self.audio_urls or [])}"
+                f"audio_count: {len(self.audio_urls or [])}, "
+                f"video_count: {len(self.video_urls or [])}"
             )
 
         result_parts = []
@@ -163,6 +167,7 @@ def _print_friendly_context(self):
                 msg_parts = []
                 image_count = 0
                 audio_count = 0
+                video_count = 0
 
                 for item in content:
                     item_type = item.get("type", "")
@@ -173,6 +178,8 @@ def _print_friendly_context(self):
                         image_count += 1
                     elif item_type == "audio_url":
                         audio_count += 1
+                    elif item_type == "video_url":
+                        video_count += 1
 
                 if image_count > 0:
                     if msg_parts:
@@ -184,6 +191,11 @@ def _print_friendly_context(self):
                         msg_parts.append(f"[+{audio_count} audios]")
                     else:
                         msg_parts.append(f"[{audio_count} audios]")
+                if video_count > 0:
+                    if msg_parts:
+                        msg_parts.append(f"[+{video_count} videos]")
+                    else:
+                        msg_parts.append(f"[{video_count} videos]")
 
                 result_parts.append(f"{role}: {''.join(msg_parts)}")
 
@@ -198,6 +210,7 @@ async def assemble_context(self) -> dict:
         # 构建内容块列表
         content_blocks = []
         image_capture_failed = False
+        transient_video_added = False
 
         # 1. 用户原始发言（OpenAI 建议：用户发言在前）
         if self.prompt and self.prompt.strip():
@@ -208,6 +221,9 @@ async def assemble_context(self) -> dict:
         elif self.audio_urls:
             # 如果没有文本但有音频，添加占位文本
             content_blocks.append({"type": "text", "text": "[音频]"})
+        elif self.video_urls:
+            # If there is no text but there is video, add a placeholder text.
+            content_blocks.append({"type": "text", "text": "[Video]"})
 
         # 2. 额外的内容块（系统提醒、指令等）
         if self.extra_user_content_parts:
@@ -245,6 +261,40 @@ async def assemble_context(self) -> dict:
                                 "url": resolved.to_data_url(),
                             },
                         }
+                # A video supplied through `extra_user_content_parts` is handled
+                # like `video_urls` below: resolve it to a portable data URI and
+                # flag it transient. Persisting it would replay the whole payload
+                # on every later request, and a local path may be gone by then.
+                if isinstance(dumped, dict) and dumped.get("type") == "video_url":
+                    video_url = dumped.get("video_url")
+                    url = video_url.get("url") if isinstance(video_url, dict) else None
+                    if isinstance(url, str) and url:
+                        try:
+                            resolved_video = await MediaResolver(
+                                url,
+                                media_type="video",
+                                default_suffix=".mp4",
+                            ).to_base64_data(strict=True)
+                        except Exception as exc:
+                            logger.warning(
+                                "Video preprocessing failed; skipping it. Error: %s",
+                                exc,
+                            )
+                            continue
+                        if not resolved_video:
+                            logger.warning(
+                                "Video preprocessing returned no data; skipping it."
+                            )
+                            continue
+                        dumped = {
+                            **dumped,
+                            "video_url": {
+                                **video_url,
+                                "url": resolved_video.to_data_url(),
+                            },
+                            "_no_save": True,
+                        }
+                        transient_video_added = True
                 content_blocks.append(dumped)
 
         # 3. Read image references without resizing or transcoding.
@@ -299,12 +349,49 @@ async def assemble_context(self) -> dict:
                     },
                 )
 
+        # 5. Video content. The block is flagged `_no_save` so that the
+        # multi-megabyte data URI is dropped when the message is written back to
+        # the conversation history; otherwise it would be replayed on every later
+        # request in the session. Providers still receive it, because they
+        # serialize messages with `model_dump()`, which does not emit `_no_save`.
+        if self.video_urls:
+            for video_url in self.video_urls:
+                try:
+                    video_data = await MediaResolver(
+                        video_url,
+                        media_type="video",
+                        default_suffix=".mp4",
+                    ).to_base64_data(strict=True)
+                except Exception as exc:
+                    logger.warning(
+                        "Video preprocessing failed; skipping it. Error: %s", exc
+                    )
+                    continue
+                if not video_data:
+                    logger.warning("Video preprocessing returned no data; skipping it.")
+                    continue
+                content_blocks.append(
+                    {
+                        "type": "video_url",
+                        "video_url": {"url": video_data.to_data_url()},
+                        "_no_save": True,
+                    },
+                )
+
         if image_capture_failed and not any(
             block.get("type") != "text" or block.get("text", "").strip()
             for block in content_blocks
         ):
             content_blocks = [{"type": "text", "text": "[Image unavailable]"}]
 
+        # A message made only of transient blocks would be persisted with an
+        # empty content list and replayed that way on every later request, so
+        # keep a text placeholder for the attachment.
+        if transient_video_added and not any(
+            not block.get("_no_save") for block in content_blocks
+        ):
+            content_blocks.insert(0, {"type": "text", "text": "[Video]"})
+
         # 只有当只有一个来自 prompt 的文本块且没有额外内容块时，才降级为简单格式以保持向后兼容
         if (
             len(content_blocks) == 1
@@ -312,6 +399,7 @@ async def assemble_context(self) -> dict:
             and not self.extra_user_content_parts
             and not self.image_urls
             and not self.audio_urls
+            and not self.video_urls
         ):
             return {"role": "user", "content": content_blocks[0]["text"]}
 
diff --git a/astrbot/core/provider/modalities.py b/astrbot/core/provider/modalities.py
index 66ac74e9b72..b028d7541bd 100644
--- a/astrbot/core/provider/modalities.py
+++ b/astrbot/core/provider/modalities.py
@@ -13,6 +13,7 @@
 class ContextSanitizeStats:
     fixed_image_blocks: int = 0
     fixed_audio_blocks: int = 0
+    fixed_video_blocks: int = 0
     fixed_tool_messages: int = 0
     removed_tool_calls: int = 0
 
@@ -21,6 +22,7 @@ def changed(self) -> bool:
         return bool(
             self.fixed_image_blocks
             or self.fixed_audio_blocks
+            or self.fixed_video_blocks
             or self.fixed_tool_messages
             or self.removed_tool_calls
         )
@@ -34,6 +36,33 @@ def _message_to_dict(message: dict[str, Any] | Message) -> dict[str, Any] | None
     return None
 
 
+def has_video_blocks(contexts: Sequence[dict[str, Any] | Message]) -> bool:
+    """Whether any message in `contexts` carries a `video_url` block.
+
+    Accepts both `Message` objects (whose content holds `ContentPart` instances)
+    and plain dicts, since callers pass either form.
+
+    Args:
+        contexts: Messages to inspect.
+
+    Returns:
+        True as soon as a `video_url` block is found.
+    """
+    for msg in contexts:
+        content = msg.content if isinstance(msg, Message) else msg.get("content")
+        if not isinstance(content, list):
+            continue
+        for part in content:
+            part_type = (
+                part.get("type")
+                if isinstance(part, dict)
+                else getattr(part, "type", None)
+            )
+            if part_type == "video_url":
+                return True
+    return False
+
+
 def sanitize_contexts_by_modalities(
     contexts: Sequence[dict[str, Any] | Message],
     modalities: list[str] | None,
@@ -41,17 +70,43 @@ def sanitize_contexts_by_modalities(
     if not contexts:
         return [], ContextSanitizeStats()
     if not modalities or not isinstance(modalities, list):
+        # An unconfigured provider (None or an empty list) keeps the historical
+        # "supports everything" default for image/audio/tool_use. Video postdates
+        # that convention and stays strictly opt-in, so it is degraded even here:
+        # this function also sanitizes a context that was assembled for another
+        # provider, which may have produced a `video_url` block because *it*
+        # declared the video modality. Everything else is copied through
+        # untouched, exactly as before.
         copied_contexts = []
+        stats = ContextSanitizeStats()
         for msg in contexts:
             copied_msg = _message_to_dict(msg)
-            if copied_msg:
-                copied_contexts.append(copied_msg)
-        return copied_contexts, ContextSanitizeStats()
+            if not copied_msg:
+                continue
+            content = copied_msg.get("content")
+            if isinstance(content, list):
+                filtered_parts: list[Any] = []
+                removed_video = False
+                for part in content:
+                    if (
+                        isinstance(part, dict)
+                        and str(part.get("type", "")).lower() == "video_url"
+                    ):
+                        removed_video = True
+                        stats.fixed_video_blocks += 1
+                        filtered_parts.append({"type": "text", "text": "[Video]"})
+                        continue
+                    filtered_parts.append(part)
+                if removed_video:
+                    copied_msg["content"] = filtered_parts
+            copied_contexts.append(copied_msg)
+        return copied_contexts, stats
 
     supports_image = "image" in modalities
     supports_audio = "audio" in modalities
+    supports_video = "video" in modalities
     supports_tool_use = "tool_use" in modalities
-    if supports_image and supports_audio and supports_tool_use:
+    if supports_image and supports_audio and supports_video and supports_tool_use:
         copied_contexts = []
         for msg in contexts:
             copied_msg = _message_to_dict(msg)
@@ -83,7 +138,7 @@ def sanitize_contexts_by_modalities(
                 msg.pop("tool_calls", None)
                 msg.pop("tool_call_id", None)
 
-        if not supports_image or not supports_audio:
+        if not supports_image or not supports_audio or not supports_video:
             content = msg.get("content")
             if isinstance(content, list):
                 filtered_parts: list[Any] = []
@@ -104,6 +159,11 @@ def sanitize_contexts_by_modalities(
                             stats.fixed_audio_blocks += 1
                             filtered_parts.append({"type": "text", "text": "[Audio]"})
                             continue
+                        if not supports_video and part_type == "video_url":
+                            removed_any_multimodal = True
+                            stats.fixed_video_blocks += 1
+                            filtered_parts.append({"type": "text", "text": "[Video]"})
+                            continue
                     filtered_parts.append(part)
                 if removed_any_multimodal:
                     msg["content"] = filtered_parts
@@ -136,6 +196,8 @@ def _tool_result_placeholder(content: Any) -> str:
                     text_parts.append("[Image]")
                 elif part_type in {"audio_url", "input_audio"}:
                     text_parts.append("[Audio]")
+                elif part_type == "video_url":
+                    text_parts.append("[Video]")
         content_text = "\n".join(part for part in text_parts if part).strip()
     else:
         content_text = ""
@@ -149,10 +211,11 @@ def log_context_sanitize_stats(stats: ContextSanitizeStats) -> None:
         return
     logger.debug(
         "context modality fix applied: "
-        "fixed_image_blocks=%s, fixed_audio_blocks=%s, "
+        "fixed_image_blocks=%s, fixed_audio_blocks=%s, fixed_video_blocks=%s, "
         "fixed_tool_messages=%s, removed_tool_calls=%s",
         stats.fixed_image_blocks,
         stats.fixed_audio_blocks,
+        stats.fixed_video_blocks,
         stats.fixed_tool_messages,
         stats.removed_tool_calls,
     )
diff --git a/astrbot/core/provider/sources/anthropic_source.py b/astrbot/core/provider/sources/anthropic_source.py
index 20a3a813d5d..a07ce575133 100644
--- a/astrbot/core/provider/sources/anthropic_source.py
+++ b/astrbot/core/provider/sources/anthropic_source.py
@@ -13,7 +13,13 @@
 
 from astrbot import logger
 from astrbot.api.provider import Provider
-from astrbot.core.agent.message import AudioURLPart, ContentPart, ImageURLPart, TextPart
+from astrbot.core.agent.message import (
+    AudioURLPart,
+    ContentPart,
+    ImageURLPart,
+    TextPart,
+    VideoURLPart,
+)
 from astrbot.core.exceptions import EmptyModelOutputError
 from astrbot.core.provider.entities import LLMResponse, TokenUsage
 from astrbot.core.provider.func_tool_manager import ToolSet
@@ -1023,6 +1029,10 @@ async def resolve_image_url(image_url: str) -> dict | None:
                         content.append(image_dict)
                 elif isinstance(block, AudioURLPart):
                     content.append({"type": "text", "text": "[Audio]"})
+                elif isinstance(block, VideoURLPart):
+                    # No native video input here and history is replayed, so
+                    # degrade to a text placeholder instead of resolving base64.
+                    content.append({"type": "text", "text": "[Video]"})
                 else:
                     raise ValueError(f"不支持的额外内容块类型: {type(block)}")
 
diff --git a/astrbot/core/provider/sources/gemini_source.py b/astrbot/core/provider/sources/gemini_source.py
index 0d8df6af99b..63151aff74a 100644
--- a/astrbot/core/provider/sources/gemini_source.py
+++ b/astrbot/core/provider/sources/gemini_source.py
@@ -15,7 +15,13 @@
 import astrbot.core.message.components as Comp
 from astrbot import logger
 from astrbot.api.provider import Provider
-from astrbot.core.agent.message import AudioURLPart, ContentPart, ImageURLPart, TextPart
+from astrbot.core.agent.message import (
+    AudioURLPart,
+    ContentPart,
+    ImageURLPart,
+    TextPart,
+    VideoURLPart,
+)
 from astrbot.core.exceptions import EmptyModelOutputError
 from astrbot.core.message.message_event_result import MessageChain
 from astrbot.core.provider.entities import LLMResponse, TokenUsage
@@ -407,6 +413,22 @@ def process_audio_url(audio_url_dict: dict) -> types.Part:
             audio_bytes = base64.b64decode(url.split(",", 1)[1])
             return types.Part.from_bytes(data=audio_bytes, mime_type=mime_type)
 
+        async def process_video_url(video_url_dict: dict) -> types.Part:
+            url = video_url_dict["url"]
+            video_data = await resolve_media_ref_to_base64_data(
+                url,
+                media_type="video",
+                strict=True,
+            )
+            if video_data is None:
+                raise ValueError(
+                    f"Failed to resolve Gemini history video: {describe_media_ref(url)}"
+                )
+            return types.Part.from_bytes(
+                data=base64.b64decode(video_data.base64_data),
+                mime_type=video_data.mime_type,
+            )
+
         def append_or_extend(
             contents: list[types.Content],
             part: list[types.Part],
@@ -430,8 +452,16 @@ def append_or_extend(
                             parts.append(types.Part.from_text(text=item["text"] or " "))
                         elif item["type"] == "image_url":
                             parts.append(await process_image_url(item["image_url"]))
-                        else:
+                        elif item["type"] == "audio_url":
                             parts.append(process_audio_url(item["audio_url"]))
+                        elif item["type"] == "video_url":
+                            parts.append(await process_video_url(item["video_url"]))
+                        else:
+                            # Never raise for an unknown block type here. This
+                            # message is replayed from the session history on
+                            # every later request, so an exception would keep
+                            # failing long after the original turn.
+                            parts.append(create_text_part(f"[{item['type']}]"))
                 else:
                     parts = [create_text_part(content)]
                 append_or_extend(gemini_contents, parts, types.UserContent)
@@ -944,6 +974,7 @@ async def text_chat(
         session_id=None,
         image_urls=None,
         audio_urls=None,
+        video_urls=None,
         func_tool=None,
         contexts=None,
         system_prompt=None,
@@ -963,6 +994,7 @@ async def text_chat(
                 prompt or "",
                 image_urls,
                 audio_urls,
+                video_urls,
                 extra_user_content_parts,
             )
         context_query = self._ensure_message_to_dicts(contexts)
@@ -1016,6 +1048,7 @@ async def text_chat_stream(
         session_id=None,
         image_urls=None,
         audio_urls=None,
+        video_urls=None,
         func_tool=None,
         contexts=None,
         system_prompt=None,
@@ -1035,6 +1068,7 @@ async def text_chat_stream(
                 prompt or "",
                 image_urls,
                 audio_urls,
+                video_urls,
                 extra_user_content_parts,
             )
         context_query = self._ensure_message_to_dicts(contexts)
@@ -1113,6 +1147,7 @@ async def assemble_context(
         text: str,
         image_urls: list[str] | None = None,
         audio_urls: list[str] | None = None,
+        video_urls: list[str] | None = None,
         extra_user_content_parts: list[ContentPart] | None = None,
     ):
         """组装上下文。"""
@@ -1151,6 +1186,27 @@ async def resolve_audio_part(audio_path: str) -> dict | None:
                 "audio_url": {"url": audio_data.to_data_url()},
             }
 
+        async def resolve_video_part(video_ref: str) -> dict | None:
+            try:
+                video_data = await resolve_media_ref_to_base64_data(
+                    video_ref,
+                    media_type="video",
+                    strict=True,
+                )
+            except Exception as exc:
+                logger.warning(
+                    "Video preprocessing failed; ignoring it. Error: %s", exc
+                )
+                return None
+
+            if not video_data:
+                logger.warning("Video preprocessing returned no data; ignoring it.")
+                return None
+            return {
+                "type": "video_url",
+                "video_url": {"url": video_data.to_data_url()},
+            }
+
         # 构建内容块列表
         content_blocks = []
 
@@ -1162,6 +1218,9 @@ async def resolve_audio_part(audio_path: str) -> dict | None:
             content_blocks.append({"type": "text", "text": "[Image]"})
         elif audio_urls:
             content_blocks.append({"type": "text", "text": "[Audio]"})
+        elif video_urls:
+            # If there is no text but there is video, add a placeholder text.
+            content_blocks.append({"type": "text", "text": "[Video]"})
         elif extra_user_content_parts:
             # 如果只有额外内容块，也需要添加占位文本
             content_blocks.append({"type": "text", "text": " "})
@@ -1179,6 +1238,10 @@ async def resolve_audio_part(audio_path: str) -> dict | None:
                     audio_part = await resolve_audio_part(part.audio_url.url)
                     if audio_part:
                         content_blocks.append(audio_part)
+                elif isinstance(part, VideoURLPart):
+                    video_part = await resolve_video_part(part.video_url.url)
+                    if video_part:
+                        content_blocks.append(video_part)
                 else:
                     raise ValueError(
                         f"Unsupported extra content part type: {type(part)}"
@@ -1197,12 +1260,19 @@ async def resolve_audio_part(audio_path: str) -> dict | None:
                 if audio_part:
                     content_blocks.append(audio_part)
 
+        if video_urls:
+            for video_ref in video_urls:
+                video_part = await resolve_video_part(video_ref)
+                if video_part:
+                    content_blocks.append(video_part)
+
         # 如果只有主文本且没有额外内容块和图片，返回简单格式以保持向后兼容
         if (
             text
             and not extra_user_content_parts
             and not image_urls
             and not audio_urls
+            and not video_urls
             and len(content_blocks) == 1
             and content_blocks[0]["type"] == "text"
         ):
diff --git a/astrbot/core/provider/sources/openai_source.py b/astrbot/core/provider/sources/openai_source.py
index 9b0c89d5d28..f4f7534d008 100644
--- a/astrbot/core/provider/sources/openai_source.py
+++ b/astrbot/core/provider/sources/openai_source.py
@@ -24,6 +24,7 @@
     ImageURLPart,
     Message,
     TextPart,
+    VideoURLPart,
 )
 from astrbot.core.agent.tool import ToolSet
 from astrbot.core.exceptions import EmptyModelOutputError
@@ -1420,6 +1421,10 @@ async def assemble_context(
                     audio_part = await self._resolve_audio_part(part.audio_url.url)
                     if audio_part:
                         content_blocks.append(audio_part)
+                elif isinstance(part, VideoURLPart):
+                    # No native video input here and history is replayed, so
+                    # degrade to a text placeholder instead of resolving base64.
+                    content_blocks.append({"type": "text", "text": "[Video]"})
                 else:
                     raise ValueError(f"不支持的额外内容块类型: {type(part)}")
 
diff --git a/dashboard/src/components/provider/ProviderChatCompletionPanel.vue b/dashboard/src/components/provider/ProviderChatCompletionPanel.vue
index 2500782e611..a6c334db80f 100644
--- a/dashboard/src/components/provider/ProviderChatCompletionPanel.vue
+++ b/dashboard/src/components/provider/ProviderChatCompletionPanel.vue
@@ -98,6 +98,7 @@
                 :is-source-modified="isSourceModified"
                 :supports-image-input="supportsImageInput"
                 :supports-audio-input="supportsAudioInput"
+                :supports-video-input="supportsVideoInput"
                 :supports-tool-call="supportsToolCall"
                 :supports-reasoning="supportsReasoning"
                 :format-context-limit="formatContextLimit"
@@ -248,6 +249,7 @@ const {
   getSourceDisplayName,
   supportsImageInput,
   supportsAudioInput,
+  supportsVideoInput,
   supportsToolCall,
   supportsReasoning,
   formatContextLimit,
diff --git a/dashboard/src/components/provider/ProviderModelsPanel.vue b/dashboard/src/components/provider/ProviderModelsPanel.vue
index 9826e1cb7ea..70ebead4451 100644
--- a/dashboard/src/components/provider/ProviderModelsPanel.vue
+++ b/dashboard/src/components/provider/ProviderModelsPanel.vue
@@ -296,6 +296,10 @@ const props = defineProps({
     type: Function,
     required: true
   },
+  supportsVideoInput: {
+    type: Function,
+    required: true
+  },
   supportsToolCall: {
     type: Function,
     required: true
@@ -367,6 +371,15 @@ const capabilityBadges = (entry) => {
       enabled: !isConfigured || modalities.includes('audio'),
       label: props.tm('models.metadata.audio')
     },
+    {
+      key: 'video',
+      icon: 'mdi-video-outline',
+      supported: props.supportsVideoInput(metadata),
+      // Video is opt-in on the backend, so an unconfigured provider must not
+      // report it as enabled the way image/audio do.
+      enabled: modalities.includes('video'),
+      label: props.tm('models.metadata.video')
+    },
     {
       key: 'tool_use',
       icon: 'mdi-wrench-outline',
diff --git a/dashboard/src/composables/useProviderSources.ts b/dashboard/src/composables/useProviderSources.ts
index aea46f37083..b53e9d5d9ef 100644
--- a/dashboard/src/composables/useProviderSources.ts
+++ b/dashboard/src/composables/useProviderSources.ts
@@ -365,6 +365,11 @@ export function useProviderSources(options: UseProviderSourcesOptions) {
     return inputs.includes('audio')
   }
 
+  function supportsVideoInput(meta: any) {
+    const inputs = meta?.modalities?.input || []
+    return inputs.includes('video')
+  }
+
   function supportsToolCall(meta: any) {
     return Boolean(meta?.tool_call)
   }
@@ -648,6 +653,11 @@ export function useProviderSources(options: UseProviderSourcesOptions) {
       if (supportsAudioInput(metadata)) {
         modalities.push('audio')
       }
+      // Video is opt-in: only enable it when the model metadata declares video
+      // input, never as part of the "assume everything" fallback above.
+      if (supportsVideoInput(metadata)) {
+        modalities.push('video')
+      }
       if (supportsToolCall(metadata)) {
         modalities.push('tool_use')
       }
@@ -831,6 +841,7 @@ export function useProviderSources(options: UseProviderSourcesOptions) {
     getModelMetadata,
     supportsImageInput,
     supportsAudioInput,
+    supportsVideoInput,
     supportsToolCall,
     supportsReasoning,
     formatContextLimit,
diff --git a/dashboard/src/i18n/locales/en-US/features/provider.json b/dashboard/src/i18n/locales/en-US/features/provider.json
index aad194fcd6f..5bd1cfc5f9b 100644
--- a/dashboard/src/i18n/locales/en-US/features/provider.json
+++ b/dashboard/src/i18n/locales/en-US/features/provider.json
@@ -165,6 +165,7 @@
     "metadata": {
       "image": "Image input",
       "audio": "Audio input",
+      "video": "Video input",
       "toolUse": "Tool use",
       "reasoning": "Reasoning",
       "context": "{tokens} context window",
diff --git a/dashboard/src/i18n/locales/ja-JP/features/provider.json b/dashboard/src/i18n/locales/ja-JP/features/provider.json
index ef95b6b1e1a..0086bf16b55 100644
--- a/dashboard/src/i18n/locales/ja-JP/features/provider.json
+++ b/dashboard/src/i18n/locales/ja-JP/features/provider.json
@@ -165,6 +165,7 @@
     "metadata": {
       "image": "画像入力",
       "audio": "音声入力",
+      "video": "動画入力",
       "toolUse": "ツール使用",
       "reasoning": "推論",
       "context": "{tokens} トークンのコンテキストウィンドウ",
diff --git a/dashboard/src/i18n/locales/ru-RU/features/provider.json b/dashboard/src/i18n/locales/ru-RU/features/provider.json
index 3a8dbc7bbe8..8d9282dfab4 100644
--- a/dashboard/src/i18n/locales/ru-RU/features/provider.json
+++ b/dashboard/src/i18n/locales/ru-RU/features/provider.json
@@ -165,6 +165,7 @@
         "metadata": {
             "image": "Ввод изображений",
             "audio": "Ввод аудио",
+            "video": "Ввод видео",
             "toolUse": "Использование инструментов",
             "reasoning": "Рассуждение",
             "context": "Контекстное окно {tokens}",
diff --git a/dashboard/src/i18n/locales/zh-CN/features/provider.json b/dashboard/src/i18n/locales/zh-CN/features/provider.json
index 0818e442d9c..84d69cb6648 100644
--- a/dashboard/src/i18n/locales/zh-CN/features/provider.json
+++ b/dashboard/src/i18n/locales/zh-CN/features/provider.json
@@ -165,6 +165,7 @@
     "metadata": {
       "image": "图像输入",
       "audio": "音频输入",
+      "video": "视频输入",
       "toolUse": "工具使用",
       "reasoning": "推理",
       "context": "{tokens} 上下文窗口",
diff --git a/dashboard/src/utils/providerMetadata.ts b/dashboard/src/utils/providerMetadata.ts
index e5fe7e2e51b..9e37a135354 100644
--- a/dashboard/src/utils/providerMetadata.ts
+++ b/dashboard/src/utils/providerMetadata.ts
@@ -65,6 +65,13 @@ export function providerCapabilityBadges(
       enabled: modalities.includes('audio'),
       label: tm('models.metadata.audio')
     },
+    {
+      key: 'video',
+      icon: 'mdi-video-outline',
+      supported: inputs.includes('video'),
+      enabled: modalities.includes('video'),
+      label: tm('models.metadata.video')
+    },
     {
       key: 'tool_use',
       icon: 'mdi-wrench-outline',
diff --git a/dashboard/src/views/ProviderPage.vue b/dashboard/src/views/ProviderPage.vue
index 5065bc8afdf..c2f92796600 100644
--- a/dashboard/src/views/ProviderPage.vue
+++ b/dashboard/src/views/ProviderPage.vue
@@ -138,6 +138,7 @@
                     :is-source-modified="isSourceModified"
                     :supports-image-input="supportsImageInput"
                     :supports-audio-input="supportsAudioInput"
+                    :supports-video-input="supportsVideoInput"
                     :supports-tool-call="supportsToolCall"
                     :supports-reasoning="supportsReasoning"
                     :format-context-limit="formatContextLimit"
@@ -364,6 +365,7 @@ const {
   getSourceDisplayName,
   supportsImageInput,
   supportsAudioInput,
+  supportsVideoInput,
   supportsToolCall,
   supportsReasoning,
   formatContextLimit,
diff --git a/tests/test_video_input_modality.py b/tests/test_video_input_modality.py
new file mode 100644
index 00000000000..21cffa93136
--- /dev/null
+++ b/tests/test_video_input_modality.py
@@ -0,0 +1,401 @@
+from types import SimpleNamespace
+
+import pytest
+
+from astrbot.core.agent.message import (
+    Message,
+    TextPart,
+    VideoURLPart,
+    dump_messages_with_checkpoints,
+)
+from astrbot.core.agent.runners.tool_loop_agent_runner import ToolLoopAgentRunner
+from astrbot.core.provider.entities import ProviderRequest
+from astrbot.core.provider.modalities import sanitize_contexts_by_modalities
+from astrbot.core.provider.sources.gemini_source import ProviderGoogleGenAI
+
+VIDEO_DATA_URL = "data:video/mp4;base64,AAAAGGZ0eXBtcDQyAAAAAG1wNDJpc29t"
+
+
+def _gemini_provider() -> ProviderGoogleGenAI:
+    provider = ProviderGoogleGenAI.__new__(ProviderGoogleGenAI)
+    provider.provider_config = {}
+    provider.provider_settings = {}
+    return provider
+
+
+def _runner(modalities) -> ToolLoopAgentRunner:
+    runner = ToolLoopAgentRunner.__new__(ToolLoopAgentRunner)
+    runner.provider = SimpleNamespace(provider_config={"modalities": modalities})
+    return runner
+
+
+def test_video_url_part_is_registered_and_round_trips():
+    """A `video_url` dict must resolve through the ContentPart registry.
+
+    The registry lookup in `ContentPart.__get_pydantic_core_schema__` is
+    unguarded, so an unregistered type raises `KeyError` the moment the block is
+    validated from persisted history.
+    """
+    message = Message.model_validate(
+        {
+            "role": "user",
+            "content": [{"type": "video_url", "video_url": {"url": VIDEO_DATA_URL}}],
+        }
+    )
+
+    assert isinstance(message.content[0], VideoURLPart)
+    assert message.content[0].video_url.url == VIDEO_DATA_URL
+
+
+@pytest.mark.asyncio
+async def test_assemble_context_marks_video_block_as_transient():
+    request = ProviderRequest(prompt="", video_urls=[VIDEO_DATA_URL])
+
+    context = await request.assemble_context()
+
+    video_blocks = [
+        block for block in context["content"] if block.get("type") == "video_url"
+    ]
+    assert len(video_blocks) == 1
+    assert video_blocks[0]["_no_save"] is True
+
+
+@pytest.mark.asyncio
+async def test_video_block_is_not_written_to_history():
+    """The video payload must not be persisted and replayed on later requests."""
+    request = ProviderRequest(prompt="look", video_urls=[VIDEO_DATA_URL])
+    context = await request.assemble_context()
+
+    messages = [Message.model_validate(context)]
+    dumped = dump_messages_with_checkpoints(messages)
+
+    persisted_types = [
+        part["type"] for part in dumped[0]["content"] if isinstance(part, dict)
+    ]
+    assert "video_url" not in persisted_types
+    assert "text" in persisted_types
+
+
+@pytest.mark.asyncio
+async def test_transient_video_reaches_provider_but_not_history():
+    """Prove both halves of the design in one place.
+
+    Providers serialize messages with `model_dump()`, which does not emit
+    `_no_save`, so the video still reaches the provider payload. Persistence goes
+    through `dump_messages_with_checkpoints`, which drops it.
+    """
+    request = ProviderRequest(prompt="look", video_urls=[VIDEO_DATA_URL])
+    context = await request.assemble_context()
+    message = Message.model_validate(context)
+
+    assert message.content[1]._no_save is True
+
+    provider_payload = message.model_dump()
+    assert any(part["type"] == "video_url" for part in provider_payload["content"])
+
+    persisted = dump_messages_with_checkpoints([message])
+    assert not any(part["type"] == "video_url" for part in persisted[0]["content"])
+
+
+@pytest.mark.asyncio
+async def test_gemini_accepts_video_url_block():
+    provider = _gemini_provider()
+    payloads = {
+        "messages": [
+            {
+                "role": "user",
+                "content": [
+                    {"type": "text", "text": "describe this"},
+                    {"type": "video_url", "video_url": {"url": VIDEO_DATA_URL}},
+                ],
+            }
+        ]
+    }
+
+    contents = await provider._prepare_conversation(payloads)
+
+    parts = contents[0].parts
+    assert len(parts) == 2
+    assert parts[1].inline_data is not None
+    assert parts[1].inline_data.mime_type == "video/mp4"
+
+
+@pytest.mark.asyncio
+async def test_gemini_degrades_unknown_block_type_instead_of_raising():
+    """An unknown block must not raise.
+
+    The message is replayed from history on every later request, so raising here
+    would keep failing long after the turn that produced it.
+    """
+    provider = _gemini_provider()
+    payloads = {
+        "messages": [
+            {
+                "role": "user",
+                "content": [{"type": "some_future_block", "payload": "x"}],
+            }
+        ]
+    }
+
+    contents = await provider._prepare_conversation(payloads)
+
+    assert contents[0].parts[0].text == "[some_future_block]"
+
+
+@pytest.mark.asyncio
+async def test_video_is_opt_in_for_providers_without_modalities():
+    """An unconfigured provider must not receive video.
+
+    An empty or missing `modalities` list means "supports everything" for
+    backward compatibility, but video postdates that convention and is not
+    universally supported.
+    """
+    runner = _runner([])
+    request = ProviderRequest(prompt="look", video_urls=[VIDEO_DATA_URL])
+
+    context = await runner._assemble_request_context_for_provider(request)
+
+    # With the video removed this collapses to the plain-text form, which is the
+    # pre-existing behaviour for a text-only request.
+    content = context["content"]
+    if isinstance(content, str):
+        assert "video_url" not in content
+    else:
+        assert not any(block.get("type") == "video_url" for block in content)
+
+
+@pytest.mark.asyncio
+async def test_video_is_sent_when_declared():
+    runner = _runner(["text", "image", "audio", "video", "tool_use"])
+    request = ProviderRequest(prompt="look", video_urls=[VIDEO_DATA_URL])
+
+    context = await runner._assemble_request_context_for_provider(request)
+
+    assert any(block.get("type") == "video_url" for block in context["content"])
+
+
+@pytest.mark.asyncio
+async def test_video_is_replaced_with_placeholder_when_not_declared():
+    runner = _runner(["text", "image", "audio", "tool_use"])
+    request = ProviderRequest(prompt="look", video_urls=[VIDEO_DATA_URL])
+
+    context = await runner._assemble_request_context_for_provider(request)
+
+    content = context["content"]
+    assert not any(block.get("type") == "video_url" for block in content)
+    assert {"type": "text", "text": "[Video]"} in content
+
+
+def test_modality_sanitizer_degrades_video_to_text():
+    contexts = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "look"},
+                {"type": "video_url", "video_url": {"url": VIDEO_DATA_URL}},
+            ],
+        }
+    ]
+
+    sanitized, stats = sanitize_contexts_by_modalities(contexts, ["text", "image"])
+
+    assert stats.fixed_video_blocks == 1
+    assert {"type": "text", "text": "[Video]"} in sanitized[0]["content"]
+
+
+def test_modality_sanitizer_keeps_video_when_declared():
+    contexts = [
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "look"},
+                {"type": "video_url", "video_url": {"url": VIDEO_DATA_URL}},
+            ],
+        }
+    ]
+
+    sanitized, stats = sanitize_contexts_by_modalities(
+        contexts, ["text", "image", "audio", "video"]
+    )
+
+    assert stats.fixed_video_blocks == 0
+    assert any(block.get("type") == "video_url" for block in sanitized[0]["content"])
+
+
+@pytest.mark.asyncio
+async def test_video_placeholder_is_added_for_video_only_requests():
+    request = ProviderRequest(prompt="", video_urls=[VIDEO_DATA_URL])
+
+    context = await request.assemble_context()
+
+    assert context["content"][0] == {"type": "text", "text": "[Video]"}
+
+
+def test_text_only_request_is_unchanged():
+    """The simple string form must be preserved for text-only requests."""
+    request = ProviderRequest(prompt="hello")
+
+    assert request.video_urls == []
+    assert TextPart(text="hello").text == "hello"
+
+
+def _video_context() -> list[dict]:
+    return [
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": "look"},
+                {"type": "video_url", "video_url": {"url": VIDEO_DATA_URL}},
+            ],
+        }
+    ]
+
+
+@pytest.mark.parametrize("modalities", [[], None])
+def test_unconfigured_sanitizer_degrades_video(modalities):
+    """Video must stay opt-in on the sanitize path too.
+
+    `sanitize_contexts_by_modalities` also runs on an already-assembled context
+    when a request falls back to another provider. That context may carry a
+    `video_url` block produced by the primary provider, so an unconfigured
+    fallback must not inherit it just because empty modalities means "supports
+    everything" for the other modalities.
+    """
+    sanitized, stats = sanitize_contexts_by_modalities(_video_context(), modalities)
+
+    assert stats.fixed_video_blocks == 1
+    assert stats.changed is True
+    assert not any(part.get("type") == "video_url" for part in sanitized[0]["content"])
+    assert {"type": "text", "text": "[Video]"} in sanitized[0]["content"]
+
+
+def test_unconfigured_sanitizer_leaves_the_rest_of_the_history_alone():
+    """Only video is degraded for an unconfigured provider.
+
+    Everything else must be copied through untouched, including messages that
+    the configured-provider path would filter out.
+    """
+    contexts = [
+        {"role": "user", "content": "hello"},
+        {"role": "assistant", "content": ""},
+        {"content": "message without a role"},
+        {
+            "role": "user",
+            "content": [{"type": "image_url", "image_url": {"url": "https://x/i.png"}}],
+        },
+    ]
+
+    sanitized, stats = sanitize_contexts_by_modalities(contexts, [])
+
+    assert sanitized == contexts
+    assert stats.changed is False
+
+
+def test_fallback_sanitizer_strips_video_for_unconfigured_provider():
+    """The runner's fallback path must apply the same opt-in rule.
+
+    The primary provider may have declared video while the fallback provider is
+    unconfigured, and the context is sanitized only once the fallback is tried.
+    """
+    runner = _runner([])
+
+    sanitized = runner._sanitize_contexts_for_provider(_video_context())
+
+    content = sanitized[0]["content"]
+    assert not any(part.get("type") == "video_url" for part in content)
+    assert {"type": "text", "text": "[Video]"} in content
+
+
+def test_unconfigured_provider_keeps_its_context_objects_without_video():
+    """The sanitizer is only entered when there is a video block to degrade.
+
+    An unconfigured provider must keep receiving the very same context objects
+    it always did; converting a plain-text history to dicts is an unrelated
+    behaviour change.
+    """
+    runner = _runner([])
+    contexts = [{"role": "user", "content": "hello"}]
+
+    assert runner._sanitize_contexts_for_provider(contexts) is contexts
+
+
+def test_fallback_sanitizer_strips_video_from_message_objects():
+    """The production path hands the sanitizer `Message` objects, not dicts."""
+    runner = _runner([])
+    message = Message.model_validate(_video_context()[0])
+
+    sanitized = runner._sanitize_contexts_for_provider([message])
+
+    content = sanitized[0]["content"]
+    assert not any(part.get("type") == "video_url" for part in content)
+    assert {"type": "text", "text": "[Video]"} in content
+
+
+@pytest.mark.asyncio
+async def test_extra_video_part_is_resolved_and_marked_transient():
+    """A `VideoURLPart` in `extra_user_content_parts` must behave like `video_urls`.
+
+    It has to be resolved to a portable data URI and flagged transient, or the
+    payload is persisted and a local path is replayed after the file is gone.
+    """
+    request = ProviderRequest(
+        prompt="look",
+        extra_user_content_parts=[
+            VideoURLPart(video_url=VideoURLPart.VideoURL(url=VIDEO_DATA_URL))
+        ],
+    )
+
+    context = await request.assemble_context()
+
+    video_blocks = [b for b in context["content"] if b.get("type") == "video_url"]
+    assert len(video_blocks) == 1
+    assert video_blocks[0]["_no_save"] is True
+    assert video_blocks[0]["video_url"]["url"].startswith("data:video/mp4;base64,")
+
+
+@pytest.mark.asyncio
+async def test_extra_video_part_reaches_provider_but_not_history():
+    request = ProviderRequest(
+        prompt="look",
+        extra_user_content_parts=[
+            VideoURLPart(video_url=VideoURLPart.VideoURL(url=VIDEO_DATA_URL))
+        ],
+    )
+    message = Message.model_validate(await request.assemble_context())
+
+    assert any(part["type"] == "video_url" for part in message.model_dump()["content"])
+
+    persisted = dump_messages_with_checkpoints([message])
+    assert not any(part["type"] == "video_url" for part in persisted[0]["content"])
+
+
+@pytest.mark.asyncio
+async def test_extra_video_only_message_keeps_a_text_placeholder():
+    """A message of only transient blocks must not persist as empty content."""
+    request = ProviderRequest(
+        prompt="",
+        extra_user_content_parts=[
+            VideoURLPart(video_url=VideoURLPart.VideoURL(url=VIDEO_DATA_URL))
+        ],
+    )
+    message = Message.model_validate(await request.assemble_context())
+
+    persisted = dump_messages_with_checkpoints([message])
+    assert not any(part.get("type") == "video_url" for part in persisted[0]["content"])
+    assert {"type": "text", "text": "[Video]"} in persisted[0]["content"]
+
+
+@pytest.mark.asyncio
+async def test_extra_video_part_is_dropped_when_it_cannot_be_resolved(tmp_path):
+    request = ProviderRequest(
+        prompt="look",
+        extra_user_content_parts=[
+            VideoURLPart(
+                video_url=VideoURLPart.VideoURL(url=str(tmp_path / "missing.mp4"))
+            )
+        ],
+    )
+
+    context = await request.assemble_context()
+
+    assert not any(b.get("type") == "video_url" for b in context["content"])