feat(api): Add image multimodal support for LLMNode (#17372)

Enhance `LLMNode` with multimodal capability, introducing support for
image outputs.

This implementation extracts base64-encoded images from LLM responses,
saves them to the storage service, and records the file metadata in the
`ToolFile` table. In conversations, these images are rendered as
markdown-based inline images.
Additionally, the images are included in the LLMNode's output as
file variables, enabling subsequent nodes in the workflow to utilize them.

To integrate file outputs into workflows, adjustments to the frontend code
are necessary.

For multimodal output functionality, updates to related model configurations
are required. Currently, this capability has been applied exclusively to
Google's Gemini models.

Close #15814.

Signed-off-by: -LAN- <laipz8200@outlook.com>
Co-authored-by: -LAN- <laipz8200@outlook.com>

api/constants/mimetypes.py

@@ -0,0 +1,7 @@
+# The two constants below should keep in sync.
+# Default content type for files which have no explicit content type.
+
+DEFAULT_MIME_TYPE = "application/octet-stream"
+# Default file extension for files which have no explicit content type, should
+# correspond to the `DEFAULT_MIME_TYPE` above.
+DEFAULT_EXTENSION = ".bin"

api/controllers/files/tool_files.py

@@ -4,7 +4,9 @@ from werkzeug.exceptions import Forbidden, NotFound
 
 from controllers.files import api
 from controllers.files.error import UnsupportedFileTypeError
+from core.tools.signature import verify_tool_file_signature
 from core.tools.tool_file_manager import ToolFileManager
+from models import db as global_db
 
 
 class ToolFilePreviewApi(Resource):
@@ -19,17 +21,14 @@ class ToolFilePreviewApi(Resource):
         parser.add_argument("as_attachment", type=bool, required=False, default=False, location="args")
 
         args = parser.parse_args()
-
+        if not verify_tool_file_signature(
-        if not ToolFileManager.verify_file(
+            file_id=file_id, timestamp=args["timestamp"], nonce=args["nonce"], sign=args["sign"]
-            file_id=file_id,
-            timestamp=args["timestamp"],
-            nonce=args["nonce"],
-            sign=args["sign"],
         ):
             raise Forbidden("Invalid request.")
 
         try:
-            stream, tool_file = ToolFileManager.get_file_generator_by_tool_file_id(
+            tool_file_manager = ToolFileManager(engine=global_db.engine)
+            stream, tool_file = tool_file_manager.get_file_generator_by_tool_file_id(
                 file_id,
             )
 

api/controllers/files/upload.py

@@ -53,7 +53,7 @@ class PluginUploadFileApi(Resource):
             raise Forbidden("Invalid request.")
 
         try:
-            tool_file = ToolFileManager.create_file_by_raw(
+            tool_file = ToolFileManager().create_file_by_raw(
                 user_id=user.id,
                 tenant_id=tenant_id,
                 file_binary=file.read(),

api/core/app/task_pipeline/message_cycle_manage.py

@@ -24,7 +24,7 @@ from core.app.entities.task_entities import (
     WorkflowTaskState,
 )
 from core.llm_generator.llm_generator import LLMGenerator
-from core.tools.tool_file_manager import ToolFileManager
+from core.tools.signature import sign_tool_file
 from extensions.ext_database import db
 from models.model import AppMode, Conversation, MessageAnnotation, MessageFile
 from services.annotation_service import AppAnnotationService
@@ -154,7 +154,7 @@ class MessageCycleManage:
             if message_file.url.startswith("http"):
                 url = message_file.url
             else:
-                url = ToolFileManager.sign_file(tool_file_id=tool_file_id, extension=extension)
+                url = sign_tool_file(tool_file_id=tool_file_id, extension=extension)
 
             return MessageFileStreamResponse(
                 task_id=self._application_generate_entity.task_id,

api/core/file/file_manager.py

@@ -10,12 +10,12 @@ from core.model_runtime.entities import (
     VideoPromptMessageContent,
 )
 from core.model_runtime.entities.message_entities import PromptMessageContentUnionTypes
+from core.tools.signature import sign_tool_file
 from extensions.ext_storage import storage
 
 from . import helpers
 from .enums import FileAttribute
 from .models import File, FileTransferMethod, FileType
-from .tool_file_parser import ToolFileParser
 
 
 def get_attr(*, file: File, attr: FileAttribute):
@@ -130,6 +130,6 @@ def _to_url(f: File, /):
         # add sign url
         if f.related_id is None or f.extension is None:
             raise ValueError("Missing file related_id or extension")
-        return ToolFileParser.get_tool_file_manager().sign_file(tool_file_id=f.related_id, extension=f.extension)
+        return sign_tool_file(tool_file_id=f.related_id, extension=f.extension)
     else:
         raise ValueError(f"Unsupported transfer method: {f.transfer_method}")

api/core/file/models.py

@@ -4,11 +4,11 @@ from typing import Any, Optional
 from pydantic import BaseModel, Field, model_validator
 
 from core.model_runtime.entities.message_entities import ImagePromptMessageContent
+from core.tools.signature import sign_tool_file
 
 from . import helpers
 from .constants import FILE_MODEL_IDENTITY
 from .enums import FileTransferMethod, FileType
-from .tool_file_parser import ToolFileParser
 
 
 class ImageConfig(BaseModel):
@@ -34,13 +34,21 @@ class FileUploadConfig(BaseModel):
 
 
 class File(BaseModel):
+    # NOTE: dify_model_identity is a special identifier used to distinguish between
+    # new and old data formats during serialization and deserialization.
     dify_model_identity: str = FILE_MODEL_IDENTITY
 
     id: Optional[str] = None  # message file id
     tenant_id: str
     type: FileType
     transfer_method: FileTransferMethod
+    # If `transfer_method` is `FileTransferMethod.remote_url`, the
+    # `remote_url` attribute must not be `None`.
     remote_url: Optional[str] = None  # remote url
+    # If `transfer_method` is `FileTransferMethod.local_file` or
+    # `FileTransferMethod.tool_file`, the `related_id` attribute must not be `None`.
+    #
+    # It should be set to `ToolFile.id` when `transfer_method` is `tool_file`.
     related_id: Optional[str] = None
     filename: Optional[str] = None
     extension: Optional[str] = Field(default=None, description="File extension, should contains dot")
@@ -110,9 +118,7 @@ class File(BaseModel):
         elif self.transfer_method == FileTransferMethod.TOOL_FILE:
             assert self.related_id is not None
             assert self.extension is not None
-            return ToolFileParser.get_tool_file_manager().sign_file(
+            return sign_tool_file(tool_file_id=self.related_id, extension=self.extension)
-                tool_file_id=self.related_id, extension=self.extension
-            )
 
     def to_plugin_parameter(self) -> dict[str, Any]:
         return {

api/core/file/tool_file_parser.py

@@ -1,12 +1,19 @@
-from typing import TYPE_CHECKING, Any, cast
+from collections.abc import Callable
+from typing import TYPE_CHECKING
 
 if TYPE_CHECKING:
     from core.tools.tool_file_manager import ToolFileManager
 
-tool_file_manager: dict[str, Any] = {"manager": None}
+_tool_file_manager_factory: Callable[[], "ToolFileManager"] | None = None
 
 
 class ToolFileParser:
     @staticmethod
     def get_tool_file_manager() -> "ToolFileManager":
-        return cast("ToolFileManager", tool_file_manager["manager"])
+        assert _tool_file_manager_factory is not None
+        return _tool_file_manager_factory()
+
+
+def set_tool_file_manager_factory(factory: Callable[[], "ToolFileManager"]) -> None:
+    global _tool_file_manager_factory
+    _tool_file_manager_factory = factory

api/core/model_manager.py

@@ -101,7 +101,7 @@ class ModelInstance:
     @overload
     def invoke_llm(
         self,
-        prompt_messages: list[PromptMessage],
+        prompt_messages: Sequence[PromptMessage],
         model_parameters: Optional[dict] = None,
         tools: Sequence[PromptMessageTool] | None = None,
         stop: Optional[list[str]] = None,

api/core/model_runtime/entities/message_entities.py

@@ -1,4 +1,5 @@
-from collections.abc import Sequence
+from abc import ABC
+from collections.abc import Mapping, Sequence
 from enum import Enum, StrEnum
 from typing import Annotated, Any, Literal, Optional, Union
 
@@ -60,8 +61,12 @@ class PromptMessageContentType(StrEnum):
     DOCUMENT = "document"
 
 
-class PromptMessageContent(BaseModel):
+class PromptMessageContent(ABC, BaseModel):
-    pass
+    """
+    Model class for prompt message content.
+    """
+
+    type: PromptMessageContentType
 
 
 class TextPromptMessageContent(PromptMessageContent):
@@ -125,7 +130,16 @@ PromptMessageContentUnionTypes = Annotated[
 ]
 
 
-class PromptMessage(BaseModel):
+CONTENT_TYPE_MAPPING: Mapping[PromptMessageContentType, type[PromptMessageContent]] = {
+    PromptMessageContentType.TEXT: TextPromptMessageContent,
+    PromptMessageContentType.IMAGE: ImagePromptMessageContent,
+    PromptMessageContentType.AUDIO: AudioPromptMessageContent,
+    PromptMessageContentType.VIDEO: VideoPromptMessageContent,
+    PromptMessageContentType.DOCUMENT: DocumentPromptMessageContent,
+}
+
+
+class PromptMessage(ABC, BaseModel):
     """
     Model class for prompt message.
     """
@@ -142,6 +156,23 @@ class PromptMessage(BaseModel):
         """
         return not self.content
 
+    @field_validator("content", mode="before")
+    @classmethod
+    def validate_content(cls, v):
+        if isinstance(v, list):
+            prompts = []
+            for prompt in v:
+                if isinstance(prompt, PromptMessageContent):
+                    if not isinstance(prompt, TextPromptMessageContent | MultiModalPromptMessageContent):
+                        prompt = CONTENT_TYPE_MAPPING[prompt.type].model_validate(prompt.model_dump())
+                elif isinstance(prompt, dict):
+                    prompt = CONTENT_TYPE_MAPPING[prompt["type"]].model_validate(prompt)
+                else:
+                    raise ValueError(f"invalid prompt message {prompt}")
+                prompts.append(prompt)
+            return prompts
+        return v
+
     @field_serializer("content")
     def serialize_content(
         self, content: Optional[Union[str, Sequence[PromptMessageContent]]]

api/core/model_runtime/model_providers/__base/large_language_model.py

@@ -2,7 +2,7 @@ import logging
 import time
 import uuid
 from collections.abc import Generator, Sequence
-from typing import Optional, Union, cast
+from typing import Optional, Union
 
 from pydantic import ConfigDict
 
@@ -13,14 +13,15 @@ from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk,
 from core.model_runtime.entities.message_entities import (
     AssistantPromptMessage,
     PromptMessage,
+    PromptMessageContentUnionTypes,
     PromptMessageTool,
+    TextPromptMessageContent,
 )
 from core.model_runtime.entities.model_entities import (
     ModelType,
     PriceType,
 )
 from core.model_runtime.model_providers.__base.ai_model import AIModel
-from core.model_runtime.utils.helper import convert_llm_result_chunk_to_str
 from core.plugin.impl.model import PluginModelClient
 
 logger = logging.getLogger(__name__)
@@ -238,7 +239,7 @@ class LargeLanguageModel(AIModel):
     def _invoke_result_generator(
         self,
         model: str,
-        result: Generator,
+        result: Generator[LLMResultChunk, None, None],
         credentials: dict,
         prompt_messages: Sequence[PromptMessage],
         model_parameters: dict,
@@ -255,11 +256,21 @@ class LargeLanguageModel(AIModel):
         :return: result generator
         """
         callbacks = callbacks or []
-        assistant_message = AssistantPromptMessage(content="")
+        message_content: list[PromptMessageContentUnionTypes] = []
         usage = None
         system_fingerprint = None
         real_model = model
 
+        def _update_message_content(content: str | list[PromptMessageContentUnionTypes] | None):
+            if not content:
+                return
+            if isinstance(content, list):
+                message_content.extend(content)
+                return
+            if isinstance(content, str):
+                message_content.append(TextPromptMessageContent(data=content))
+                return
+
         try:
             for chunk in result:
                 # Following https://github.com/langgenius/dify/issues/17799,
@@ -281,9 +292,8 @@ class LargeLanguageModel(AIModel):
                     callbacks=callbacks,
                 )
 
-                text = convert_llm_result_chunk_to_str(chunk.delta.message.content)
+                _update_message_content(chunk.delta.message.content)
-                current_content = cast(str, assistant_message.content)
+
-                assistant_message.content = current_content + text
                 real_model = chunk.model
                 if chunk.delta.usage:
                     usage = chunk.delta.usage
@@ -293,6 +303,7 @@ class LargeLanguageModel(AIModel):
         except Exception as e:
             raise self._transform_invoke_error(e)
 
+        assistant_message = AssistantPromptMessage(content=message_content)
         self._trigger_after_invoke_callbacks(
             model=model,
             result=LLMResult(

api/core/model_runtime/utils/helper.py

@@ -1,8 +1,6 @@
 import pydantic
 from pydantic import BaseModel
 
-from core.model_runtime.entities.message_entities import PromptMessageContentUnionTypes
-
 
 def dump_model(model: BaseModel) -> dict:
     if hasattr(pydantic, "model_dump"):
@@ -10,18 +8,3 @@ def dump_model(model: BaseModel) -> dict:
         return pydantic.model_dump(model)  # type: ignore
     else:
         return model.model_dump()
-
-
-def convert_llm_result_chunk_to_str(content: None | str | list[PromptMessageContentUnionTypes]) -> str:
-    if content is None:
-        message_text = ""
-    elif isinstance(content, str):
-        message_text = content
-    elif isinstance(content, list):
-        # Assuming the list contains PromptMessageContent objects with a "data" attribute
-        message_text = "".join(
-            item.data if hasattr(item, "data") and isinstance(item.data, str) else str(item) for item in content
-        )
-    else:
-        message_text = str(content)
-    return message_text

api/core/tools/signature.py

@@ -0,0 +1,41 @@
+import base64
+import hashlib
+import hmac
+import os
+import time
+
+from configs import dify_config
+
+
+def sign_tool_file(tool_file_id: str, extension: str) -> str:
+    """
+    sign file to get a temporary url
+    """
+    base_url = dify_config.FILES_URL
+    file_preview_url = f"{base_url}/files/tools/{tool_file_id}{extension}"
+
+    timestamp = str(int(time.time()))
+    nonce = os.urandom(16).hex()
+    data_to_sign = f"file-preview|{tool_file_id}|{timestamp}|{nonce}"
+    secret_key = dify_config.SECRET_KEY.encode() if dify_config.SECRET_KEY else b""
+    sign = hmac.new(secret_key, data_to_sign.encode(), hashlib.sha256).digest()
+    encoded_sign = base64.urlsafe_b64encode(sign).decode()
+
+    return f"{file_preview_url}?timestamp={timestamp}&nonce={nonce}&sign={encoded_sign}"
+
+
+def verify_tool_file_signature(file_id: str, timestamp: str, nonce: str, sign: str) -> bool:
+    """
+    verify signature
+    """
+    data_to_sign = f"file-preview|{file_id}|{timestamp}|{nonce}"
+    secret_key = dify_config.SECRET_KEY.encode() if dify_config.SECRET_KEY else b""
+    recalculated_sign = hmac.new(secret_key, data_to_sign.encode(), hashlib.sha256).digest()
+    recalculated_encoded_sign = base64.urlsafe_b64encode(recalculated_sign).decode()
+
+    # verify signature
+    if sign != recalculated_encoded_sign:
+        return False
+
+    current_time = int(time.time())
+    return current_time - int(timestamp) <= dify_config.FILES_ACCESS_TIMEOUT

api/core/tools/tool_file_manager.py

@@ -9,18 +9,28 @@ from typing import Optional, Union
 from uuid import uuid4
 
 import httpx
+from sqlalchemy.orm import Session
 
 from configs import dify_config
 from core.helper import ssrf_proxy
-from extensions.ext_database import db
+from extensions.ext_database import db as global_db
 from extensions.ext_storage import storage
 from models.model import MessageFile
 from models.tools import ToolFile
 
 logger = logging.getLogger(__name__)
 
+from sqlalchemy.engine import Engine
+
 
 class ToolFileManager:
+    _engine: Engine
+
+    def __init__(self, engine: Engine | None = None):
+        if engine is None:
+            engine = global_db.engine
+        self._engine = engine
+
     @staticmethod
     def sign_file(tool_file_id: str, extension: str) -> str:
         """
@@ -55,8 +65,8 @@ class ToolFileManager:
         current_time = int(time.time())
         return current_time - int(timestamp) <= dify_config.FILES_ACCESS_TIMEOUT
 
-    @staticmethod
     def create_file_by_raw(
+        self,
         *,
         user_id: str,
         tenant_id: str,
@@ -77,6 +87,7 @@ class ToolFileManager:
         filepath = f"tools/{tenant_id}/{unique_filename}"
         storage.save(filepath, file_binary)
 
+        with Session(self._engine, expire_on_commit=False) as session:
             tool_file = ToolFile(
                 user_id=user_id,
                 tenant_id=tenant_id,
@@ -87,14 +98,14 @@ class ToolFileManager:
                 size=len(file_binary),
             )
 
-        db.session.add(tool_file)
+            session.add(tool_file)
-        db.session.commit()
+            session.commit()
-        db.session.refresh(tool_file)
+            session.refresh(tool_file)
 
         return tool_file
 
-    @staticmethod
     def create_file_by_url(
+        self,
         user_id: str,
         tenant_id: str,
         file_url: str,
@@ -119,6 +130,7 @@ class ToolFileManager:
         filepath = f"tools/{tenant_id}/{filename}"
         storage.save(filepath, blob)
 
+        with Session(self._engine, expire_on_commit=False) as session:
             tool_file = ToolFile(
                 user_id=user_id,
                 tenant_id=tenant_id,
@@ -130,13 +142,12 @@ class ToolFileManager:
                 size=len(blob),
             )
 
-        db.session.add(tool_file)
+            session.add(tool_file)
-        db.session.commit()
+            session.commit()
 
         return tool_file
 
-    @staticmethod
+    def get_file_binary(self, id: str) -> Union[tuple[bytes, str], None]:
-    def get_file_binary(id: str) -> Union[tuple[bytes, str], None]:
         """
         get file binary
 
@@ -144,8 +155,9 @@ class ToolFileManager:
 
         :return: the binary of the file, mime type
         """
+        with Session(self._engine, expire_on_commit=False) as session:
             tool_file: ToolFile | None = (
-            db.session.query(ToolFile)
+                session.query(ToolFile)
                 .filter(
                     ToolFile.id == id,
                 )
@@ -159,8 +171,7 @@ class ToolFileManager:
 
         return blob, tool_file.mimetype
 
-    @staticmethod
+    def get_file_binary_by_message_file_id(self, id: str) -> Union[tuple[bytes, str], None]:
-    def get_file_binary_by_message_file_id(id: str) -> Union[tuple[bytes, str], None]:
         """
         get file binary
 
@@ -168,8 +179,9 @@ class ToolFileManager:
 
         :return: the binary of the file, mime type
         """
+        with Session(self._engine, expire_on_commit=False) as session:
             message_file: MessageFile | None = (
-            db.session.query(MessageFile)
+                session.query(MessageFile)
                 .filter(
                     MessageFile.id == id,
                 )
@@ -189,7 +201,7 @@ class ToolFileManager:
                 tool_file_id = None
 
             tool_file: ToolFile | None = (
-            db.session.query(ToolFile)
+                session.query(ToolFile)
                 .filter(
                     ToolFile.id == tool_file_id,
                 )
@@ -203,8 +215,7 @@ class ToolFileManager:
 
         return blob, tool_file.mimetype
 
-    @staticmethod
+    def get_file_generator_by_tool_file_id(self, tool_file_id: str):
-    def get_file_generator_by_tool_file_id(tool_file_id: str):
         """
         get file binary
 
@@ -212,8 +223,9 @@ class ToolFileManager:
 
         :return: the binary of the file, mime type
         """
+        with Session(self._engine, expire_on_commit=False) as session:
             tool_file: ToolFile | None = (
-            db.session.query(ToolFile)
+                session.query(ToolFile)
                 .filter(
                     ToolFile.id == tool_file_id,
                 )
@@ -229,6 +241,11 @@ class ToolFileManager:
 
 
 # init tool_file_parser
-from core.file.tool_file_parser import tool_file_manager
+from core.file.tool_file_parser import set_tool_file_manager_factory
 
-tool_file_manager["manager"] = ToolFileManager
+
+def _factory() -> ToolFileManager:
+    return ToolFileManager()
+
+
+set_tool_file_manager_factory(_factory)

api/core/tools/utils/message_transformer.py

@@ -31,8 +31,8 @@ class ToolFileMessageTransformer:
                 # try to download image
                 try:
                     assert isinstance(message.message, ToolInvokeMessage.TextMessage)
-
+                    tool_file_manager = ToolFileManager()
-                    file = ToolFileManager.create_file_by_url(
+                    file = tool_file_manager.create_file_by_url(
                         user_id=user_id,
                         tenant_id=tenant_id,
                         file_url=message.message.text,
@@ -68,7 +68,8 @@ class ToolFileMessageTransformer:
 
                 # FIXME: should do a type check here.
                 assert isinstance(message.message.blob, bytes)
-                file = ToolFileManager.create_file_by_raw(
+                tool_file_manager = ToolFileManager()
+                file = tool_file_manager.create_file_by_raw(
                     user_id=user_id,
                     tenant_id=tenant_id,
                     conversation_id=conversation_id,

api/core/workflow/nodes/http_request/node.py

@@ -191,8 +191,9 @@ class HttpRequestNode(BaseNode[HttpRequestNodeData]):
         mime_type = (
             content_disposition_type or content_type or mimetypes.guess_type(filename)[0] or "application/octet-stream"
         )
+        tool_file_manager = ToolFileManager()
 
-        tool_file = ToolFileManager.create_file_by_raw(
+        tool_file = tool_file_manager.create_file_by_raw(
             user_id=self.user_id,
             tenant_id=self.tenant_id,
             conversation_id=None,

api/core/workflow/nodes/llm/exc.py

@@ -38,3 +38,8 @@ class MemoryRolePrefixRequiredError(LLMNodeError):
 class FileTypeNotSupportError(LLMNodeError):
     def __init__(self, *, type_name: str):
         super().__init__(f"{type_name} type is not supported by this model")
+
+
+class UnsupportedPromptContentTypeError(LLMNodeError):
+    def __init__(self, *, type_name: str) -> None:
+        super().__init__(f"Prompt content type {type_name} is not supported.")

api/core/workflow/nodes/llm/file_saver.py

@@ -0,0 +1,160 @@
+import mimetypes
+import typing as tp
+
+from sqlalchemy import Engine
+
+from constants.mimetypes import DEFAULT_EXTENSION, DEFAULT_MIME_TYPE
+from core.file import File, FileTransferMethod, FileType
+from core.helper import ssrf_proxy
+from core.tools.signature import sign_tool_file
+from core.tools.tool_file_manager import ToolFileManager
+from models import db as global_db
+
+
+class LLMFileSaver(tp.Protocol):
+    """LLMFileSaver is responsible for save multimodal output returned by
+    LLM.
+    """
+
+    def save_binary_string(
+        self,
+        data: bytes,
+        mime_type: str,
+        file_type: FileType,
+        extension_override: str | None = None,
+    ) -> File:
+        """save_binary_string saves the inline file data returned by LLM.
+
+        Currently (2025-04-30), only some of Google Gemini models will return
+        multimodal output as inline data.
+
+        :param data: the contents of the file
+        :param mime_type: the media type of the file, specified by rfc6838
+            (https://datatracker.ietf.org/doc/html/rfc6838)
+        :param file_type: The file type of the inline file.
+        :param extension_override: Override the auto-detected file extension while saving this file.
+
+            The default value is `None`, which means do not override the file extension and guessing it
+            from the `mime_type` attribute while saving the file.
+
+            Setting it to values other than `None` means override the file's extension, and
+            will bypass the extension guessing saving the file.
+
+            Specially, setting it to empty string (`""`) will leave the file extension empty.
+
+            When it is not `None` or empty string (`""`), it should be a string beginning with a
+            dot (`.`). For example, `.py` and `.tar.gz` are both valid values, while `py`
+            and `tar.gz` are not.
+        """
+        pass
+
+    def save_remote_url(self, url: str, file_type: FileType) -> File:
+        """save_remote_url saves the file from a remote url returned by LLM.
+
+        Currently (2025-04-30), no model returns multimodel output as a url.
+
+        :param url: the url of the file.
+        :param file_type: the file type of the file, check `FileType` enum for reference.
+        """
+        pass
+
+
+EngineFactory: tp.TypeAlias = tp.Callable[[], Engine]
+
+
+class FileSaverImpl(LLMFileSaver):
+    _engine_factory: EngineFactory
+    _tenant_id: str
+    _user_id: str
+
+    def __init__(self, user_id: str, tenant_id: str, engine_factory: EngineFactory | None = None):
+        if engine_factory is None:
+
+            def _factory():
+                return global_db.engine
+
+            engine_factory = _factory
+        self._engine_factory = engine_factory
+        self._user_id = user_id
+        self._tenant_id = tenant_id
+
+    def _get_tool_file_manager(self):
+        return ToolFileManager(engine=self._engine_factory())
+
+    def save_remote_url(self, url: str, file_type: FileType) -> File:
+        http_response = ssrf_proxy.get(url)
+        http_response.raise_for_status()
+        data = http_response.content
+        mime_type_from_header = http_response.headers.get("Content-Type")
+        mime_type, extension = _extract_content_type_and_extension(url, mime_type_from_header)
+        return self.save_binary_string(data, mime_type, file_type, extension_override=extension)
+
+    def save_binary_string(
+        self,
+        data: bytes,
+        mime_type: str,
+        file_type: FileType,
+        extension_override: str | None = None,
+    ) -> File:
+        tool_file_manager = self._get_tool_file_manager()
+        tool_file = tool_file_manager.create_file_by_raw(
+            user_id=self._user_id,
+            tenant_id=self._tenant_id,
+            # TODO(QuantumGhost): what is conversation id?
+            conversation_id=None,
+            file_binary=data,
+            mimetype=mime_type,
+        )
+        extension_override = _validate_extension_override(extension_override)
+        extension = _get_extension(mime_type, extension_override)
+        url = sign_tool_file(tool_file.id, extension)
+
+        return File(
+            tenant_id=self._tenant_id,
+            type=file_type,
+            transfer_method=FileTransferMethod.TOOL_FILE,
+            filename=tool_file.name,
+            extension=extension,
+            mime_type=mime_type,
+            size=len(data),
+            related_id=tool_file.id,
+            url=url,
+            # TODO(QuantumGhost): how should I set the following key?
+            # What's the difference between `remote_url` and `url`?
+            # What's the purpose of `storage_key` and `dify_model_identity`?
+            storage_key=tool_file.file_key,
+        )
+
+
+def _get_extension(mime_type: str, extension_override: str | None = None) -> str:
+    """get_extension return the extension of file.
+
+    If the `extension_override` parameter is set, this function should honor it and
+    return its value.
+    """
+    if extension_override is not None:
+        return extension_override
+    return mimetypes.guess_extension(mime_type) or DEFAULT_EXTENSION
+
+
+def _extract_content_type_and_extension(url: str, content_type_header: str | None) -> tuple[str, str]:
+    """_extract_content_type_and_extension tries to
+    guess content type of file from url and `Content-Type` header in response.
+    """
+    if content_type_header:
+        extension = mimetypes.guess_extension(content_type_header) or DEFAULT_EXTENSION
+        return content_type_header, extension
+    content_type = mimetypes.guess_type(url)[0] or DEFAULT_MIME_TYPE
+    extension = mimetypes.guess_extension(content_type) or DEFAULT_EXTENSION
+    return content_type, extension
+
+
+def _validate_extension_override(extension_override: str | None) -> str | None:
+    # `extension_override` is allow to be `None or `""`.
+    if extension_override is None:
+        return None
+    if extension_override == "":
+        return ""
+    if not extension_override.startswith("."):
+        raise ValueError("extension_override should start with '.' if not None or empty.", extension_override)
+    return extension_override

api/core/workflow/nodes/llm/node.py

@@ -1,3 +1,5 @@
+import base64
+import io
 import json
 import logging
 from collections.abc import Generator, Mapping, Sequence
@@ -21,7 +23,7 @@ from core.model_runtime.entities import (
     PromptMessageContentType,
     TextPromptMessageContent,
 )
-from core.model_runtime.entities.llm_entities import LLMResult, LLMUsage
+from core.model_runtime.entities.llm_entities import LLMResult, LLMResultChunk, LLMUsage
 from core.model_runtime.entities.message_entities import (
     AssistantPromptMessage,
     PromptMessageContentUnionTypes,
@@ -38,7 +40,6 @@ from core.model_runtime.entities.model_entities import (
 )
 from core.model_runtime.model_providers.__base.large_language_model import LargeLanguageModel
 from core.model_runtime.utils.encoders import jsonable_encoder
-from core.model_runtime.utils.helper import convert_llm_result_chunk_to_str
 from core.plugin.entities.plugin import ModelProviderID
 from core.prompt.entities.advanced_prompt_entities import CompletionModelPromptTemplate, MemoryConfig
 from core.prompt.utils.prompt_message_util import PromptMessageUtil
@@ -95,9 +96,13 @@ from .exc import (
     TemplateTypeNotSupportError,
     VariableNotFoundError,
 )
+from .file_saver import FileSaverImpl, LLMFileSaver
 
 if TYPE_CHECKING:
     from core.file.models import File
+    from core.workflow.graph_engine.entities.graph import Graph
+    from core.workflow.graph_engine.entities.graph_init_params import GraphInitParams
+    from core.workflow.graph_engine.entities.graph_runtime_state import GraphRuntimeState
 
 logger = logging.getLogger(__name__)
 
@@ -106,6 +111,43 @@ class LLMNode(BaseNode[LLMNodeData]):
     _node_data_cls = LLMNodeData
     _node_type = NodeType.LLM
 
+    # Instance attributes specific to LLMNode.
+    # Output variable for file
+    _file_outputs: list["File"]
+
+    _llm_file_saver: LLMFileSaver
+
+    def __init__(
+        self,
+        id: str,
+        config: Mapping[str, Any],
+        graph_init_params: "GraphInitParams",
+        graph: "Graph",
+        graph_runtime_state: "GraphRuntimeState",
+        previous_node_id: Optional[str] = None,
+        thread_pool_id: Optional[str] = None,
+        *,
+        llm_file_saver: LLMFileSaver | None = None,
+    ) -> None:
+        super().__init__(
+            id=id,
+            config=config,
+            graph_init_params=graph_init_params,
+            graph=graph,
+            graph_runtime_state=graph_runtime_state,
+            previous_node_id=previous_node_id,
+            thread_pool_id=thread_pool_id,
+        )
+        # LLM file outputs, used for MultiModal outputs.
+        self._file_outputs: list[File] = []
+
+        if llm_file_saver is None:
+            llm_file_saver = FileSaverImpl(
+                user_id=graph_init_params.user_id,
+                tenant_id=graph_init_params.tenant_id,
+            )
+        self._llm_file_saver = llm_file_saver
+
     def _run(self) -> Generator[NodeEvent | InNodeEvent, None, None]:
         def process_structured_output(text: str) -> Optional[dict[str, Any] | list[Any]]:
             """Process structured output if enabled"""
@@ -215,6 +257,9 @@ class LLMNode(BaseNode[LLMNodeData]):
             structured_output = process_structured_output(result_text)
             if structured_output:
                 outputs["structured_output"] = structured_output
+            if self._file_outputs is not None:
+                outputs["files"] = self._file_outputs
+
             yield RunCompletedEvent(
                 run_result=NodeRunResult(
                     status=WorkflowNodeExecutionStatus.SUCCEEDED,
@@ -240,6 +285,7 @@ class LLMNode(BaseNode[LLMNodeData]):
                 )
             )
         except Exception as e:
+            logger.exception("error while executing llm node")
             yield RunCompletedEvent(
                 run_result=NodeRunResult(
                     status=WorkflowNodeExecutionStatus.FAILED,
@@ -268,44 +314,45 @@ class LLMNode(BaseNode[LLMNodeData]):
 
         return self._handle_invoke_result(invoke_result=invoke_result)
 
-    def _handle_invoke_result(self, invoke_result: LLMResult | Generator) -> Generator[NodeEvent, None, None]:
+    def _handle_invoke_result(
+        self, invoke_result: LLMResult | Generator[LLMResultChunk, None, None]
+    ) -> Generator[NodeEvent, None, None]:
+        # For blocking mode
         if isinstance(invoke_result, LLMResult):
-            message_text = convert_llm_result_chunk_to_str(invoke_result.message.content)
+            event = self._handle_blocking_result(invoke_result=invoke_result)
-
+            yield event
-            yield ModelInvokeCompletedEvent(
-                text=message_text,
-                usage=invoke_result.usage,
-                finish_reason=None,
-            )
             return
 
-        model = None
+        # For streaming mode
+        model = ""
         prompt_messages: list[PromptMessage] = []
-        full_text = ""
+
-        usage = None
+        usage = LLMUsage.empty_usage()
         finish_reason = None
+        full_text_buffer = io.StringIO()
         for result in invoke_result:
-            text = convert_llm_result_chunk_to_str(result.delta.message.content)
+            contents = result.delta.message.content
-            full_text += text
+            for text_part in self._save_multimodal_output_and_convert_result_to_markdown(contents):
+                full_text_buffer.write(text_part)
+                yield RunStreamChunkEvent(chunk_content=text_part, from_variable_selector=[self.node_id, "text"])
 
-            yield RunStreamChunkEvent(chunk_content=text, from_variable_selector=[self.node_id, "text"])
+            # Update the whole metadata
-
+            if not model and result.model:
-            if not model:
                 model = result.model
-
+            if len(prompt_messages) == 0:
-            if not prompt_messages:
+                # TODO(QuantumGhost): it seems that this update has no visable effect.
-                prompt_messages = result.prompt_messages
+                # What's the purpose of the line below?
-
+                prompt_messages = list(result.prompt_messages)
-            if not usage and result.delta.usage:
+            if usage.prompt_tokens == 0 and result.delta.usage:
                 usage = result.delta.usage
-
+            if finish_reason is None and result.delta.finish_reason:
-            if not finish_reason and result.delta.finish_reason:
                 finish_reason = result.delta.finish_reason
 
-        if not usage:
+        yield ModelInvokeCompletedEvent(text=full_text_buffer.getvalue(), usage=usage, finish_reason=finish_reason)
-            usage = LLMUsage.empty_usage()
 
-        yield ModelInvokeCompletedEvent(text=full_text, usage=usage, finish_reason=finish_reason)
+    def _image_file_to_markdown(self, file: "File", /):
+        text_chunk = f"![]({file.generate_url()})"
+        return text_chunk
 
     def _transform_chat_messages(
         self, messages: Sequence[LLMNodeChatModelMessage] | LLMNodeCompletionModelPromptTemplate, /
@@ -963,6 +1010,42 @@ class LLMNode(BaseNode[LLMNodeData]):
 
         return prompt_messages
 
+    def _handle_blocking_result(self, *, invoke_result: LLMResult) -> ModelInvokeCompletedEvent:
+        buffer = io.StringIO()
+        for text_part in self._save_multimodal_output_and_convert_result_to_markdown(invoke_result.message.content):
+            buffer.write(text_part)
+
+        return ModelInvokeCompletedEvent(
+            text=buffer.getvalue(),
+            usage=invoke_result.usage,
+            finish_reason=None,
+        )
+
+    def _save_multimodal_image_output(self, content: ImagePromptMessageContent) -> "File":
+        """_save_multimodal_output saves multi-modal contents generated by LLM plugins.
+
+        There are two kinds of multimodal outputs:
+
+          - Inlined data encoded in base64, which would be saved to storage directly.
+          - Remote files referenced by an url, which would be downloaded and then saved to storage.
+
+        Currently, only image files are supported.
+        """
+        # Inject the saver somehow...
+        _saver = self._llm_file_saver
+
+        # If this
+        if content.url != "":
+            saved_file = _saver.save_remote_url(content.url, FileType.IMAGE)
+        else:
+            saved_file = _saver.save_binary_string(
+                data=base64.b64decode(content.base64_data),
+                mime_type=content.mime_type,
+                file_type=FileType.IMAGE,
+            )
+        self._file_outputs.append(saved_file)
+        return saved_file
+
     def _handle_native_json_schema(self, model_parameters: dict, rules: list[ParameterRule]) -> dict:
         """
         Handle structured output for models with native JSON schema support.
@@ -1123,6 +1206,41 @@ class LLMNode(BaseNode[LLMNodeData]):
             else SupportStructuredOutputStatus.UNSUPPORTED
         )
 
+    def _save_multimodal_output_and_convert_result_to_markdown(
+        self,
+        contents: str | list[PromptMessageContentUnionTypes] | None,
+    ) -> Generator[str, None, None]:
+        """Convert intermediate prompt messages into strings and yield them to the caller.
+
+        If the messages contain non-textual content (e.g., multimedia like images or videos),
+        it will be saved separately, and the corresponding Markdown representation will
+        be yielded to the caller.
+        """
+
+        # NOTE(QuantumGhost): This function should yield results to the caller immediately
+        # whenever new content or partial content is available. Avoid any intermediate buffering
+        # of results. Additionally, do not yield empty strings; instead, yield from an empty list
+        # if necessary.
+        if contents is None:
+            yield from []
+            return
+        if isinstance(contents, str):
+            yield contents
+        elif isinstance(contents, list):
+            for item in contents:
+                if isinstance(item, TextPromptMessageContent):
+                    yield item.data
+                elif isinstance(item, ImagePromptMessageContent):
+                    file = self._save_multimodal_image_output(item)
+                    self._file_outputs.append(file)
+                    yield self._image_file_to_markdown(file)
+                else:
+                    logger.warning("unknown item type encountered, type=%s", type(item))
+                    yield str(item)
+        else:
+            logger.warning("unknown contents type encountered, type=%s", type(contents))
+            yield str(contents)
+
 
 def _combine_message_content_with_role(
     *, contents: Optional[str | list[PromptMessageContentUnionTypes]] = None, role: PromptMessageRole

api/models/engine.py

@@ -10,4 +10,16 @@ POSTGRES_INDEXES_NAMING_CONVENTION = {
 }
 
 metadata = MetaData(naming_convention=POSTGRES_INDEXES_NAMING_CONVENTION)
+
+# ****** IMPORTANT NOTICE ******
+#
+# NOTE(QuantumGhost): Avoid directly importing and using `db` in modules outside of the
+# `controllers` package.
+#
+# Instead, import `db` within the `controllers` package and pass it as an argument to
+# functions or class constructors.
+#
+# Directly importing `db` in other modules can make the code more difficult to read, test, and maintain.
+#
+# Whenever possible, avoid this pattern in new code.
 db = SQLAlchemy(metadata=metadata)

api/models/model.py

@@ -8,6 +8,7 @@ from typing import TYPE_CHECKING, Any, Literal, Optional, cast
 
 from core.plugin.entities.plugin import GenericProviderID
 from core.tools.entities.tool_entities import ToolProviderType
+from core.tools.signature import sign_tool_file
 from services.plugin.plugin_service import PluginService
 
 if TYPE_CHECKING:
@@ -23,7 +24,6 @@ from configs import dify_config
 from constants import DEFAULT_FILE_NUMBER_LIMITS
 from core.file import FILE_MODEL_IDENTITY, File, FileTransferMethod, FileType
 from core.file import helpers as file_helpers
-from core.file.tool_file_parser import ToolFileParser
 from libs.helper import generate_string
 from models.base import Base
 from models.enums import CreatedByRole
@@ -986,9 +986,7 @@ class Message(db.Model):  # type: ignore[name-defined]
                 if not tool_file_id:
                     continue
 
-                sign_url = ToolFileParser.get_tool_file_manager().sign_file(
+                sign_url = sign_tool_file(tool_file_id=tool_file_id, extension=extension)
-                    tool_file_id=tool_file_id, extension=extension
-                )
             elif "file-preview" in url:
                 # get upload file id
                 upload_file_id_pattern = r"\/files\/([\w-]+)\/file-preview?\?timestamp="

api/models/tools.py

@@ -263,8 +263,8 @@ class ToolConversationVariables(Base):
 
 
 class ToolFile(Base):
-    """
+    """This table stores file metadata generated in workflows,
-    store the file created by agent
+    not only files created by agent.
     """
 
     __tablename__ = "tool_files"

api/tests/unit_tests/core/workflow/nodes/llm/test_file_saver.py

@@ -0,0 +1,192 @@
+import uuid
+from typing import NamedTuple
+from unittest import mock
+
+import httpx
+import pytest
+from sqlalchemy import Engine
+
+from core.file import FileTransferMethod, FileType, models
+from core.helper import ssrf_proxy
+from core.tools import signature
+from core.tools.tool_file_manager import ToolFileManager
+from core.workflow.nodes.llm.file_saver import (
+    FileSaverImpl,
+    _extract_content_type_and_extension,
+    _get_extension,
+    _validate_extension_override,
+)
+from models import ToolFile
+
+_PNG_DATA = b"\x89PNG\r\n\x1a\n"
+
+
+def _gen_id():
+    return str(uuid.uuid4())
+
+
+class TestFileSaverImpl:
+    def test_save_binary_string(self, monkeypatch):
+        user_id = _gen_id()
+        tenant_id = _gen_id()
+        file_type = FileType.IMAGE
+        mime_type = "image/png"
+        mock_signed_url = "https://example.com/image.png"
+        mock_tool_file = ToolFile(
+            id=_gen_id(),
+            user_id=user_id,
+            tenant_id=tenant_id,
+            conversation_id=None,
+            file_key="test-file-key",
+            mimetype=mime_type,
+            original_url=None,
+            name=f"{_gen_id()}.png",
+            size=len(_PNG_DATA),
+        )
+        mocked_tool_file_manager = mock.MagicMock(spec=ToolFileManager)
+        mocked_engine = mock.MagicMock(spec=Engine)
+
+        mocked_tool_file_manager.create_file_by_raw.return_value = mock_tool_file
+        monkeypatch.setattr(FileSaverImpl, "_get_tool_file_manager", lambda _: mocked_tool_file_manager)
+        # Since `File.generate_url` used `ToolFileManager.sign_file` directly, we also need to patch it here.
+        mocked_sign_file = mock.MagicMock(spec=signature.sign_tool_file)
+        # Since `File.generate_url` used `signature.sign_tool_file` directly, we also need to patch it here.
+        monkeypatch.setattr(models, "sign_tool_file", mocked_sign_file)
+        mocked_sign_file.return_value = mock_signed_url
+
+        storage_file_manager = FileSaverImpl(
+            user_id=user_id,
+            tenant_id=tenant_id,
+            engine_factory=mocked_engine,
+        )
+
+        file = storage_file_manager.save_binary_string(_PNG_DATA, mime_type, file_type)
+        assert file.tenant_id == tenant_id
+        assert file.type == file_type
+        assert file.transfer_method == FileTransferMethod.TOOL_FILE
+        assert file.extension == ".png"
+        assert file.mime_type == mime_type
+        assert file.size == len(_PNG_DATA)
+        assert file.related_id == mock_tool_file.id
+
+        assert file.generate_url() == mock_signed_url
+
+        mocked_tool_file_manager.create_file_by_raw.assert_called_once_with(
+            user_id=user_id,
+            tenant_id=tenant_id,
+            conversation_id=None,
+            file_binary=_PNG_DATA,
+            mimetype=mime_type,
+        )
+        mocked_sign_file.assert_called_once_with(mock_tool_file.id, ".png")
+
+    def test_save_remote_url_request_failed(self, monkeypatch):
+        _TEST_URL = "https://example.com/image.png"
+        mock_request = httpx.Request("GET", _TEST_URL)
+        mock_response = httpx.Response(
+            status_code=401,
+            request=mock_request,
+        )
+        file_saver = FileSaverImpl(
+            user_id=_gen_id(),
+            tenant_id=_gen_id(),
+        )
+        mock_get = mock.MagicMock(spec=ssrf_proxy.get, return_value=mock_response)
+        monkeypatch.setattr(ssrf_proxy, "get", mock_get)
+
+        with pytest.raises(httpx.HTTPStatusError) as exc:
+            file_saver.save_remote_url(_TEST_URL, FileType.IMAGE)
+        mock_get.assert_called_once_with(_TEST_URL)
+        assert exc.value.response.status_code == 401
+
+    def test_save_remote_url_success(self, monkeypatch):
+        _TEST_URL = "https://example.com/image.png"
+        mime_type = "image/png"
+        user_id = _gen_id()
+        tenant_id = _gen_id()
+
+        mock_request = httpx.Request("GET", _TEST_URL)
+        mock_response = httpx.Response(
+            status_code=200,
+            content=b"test-data",
+            headers={"Content-Type": mime_type},
+            request=mock_request,
+        )
+
+        file_saver = FileSaverImpl(user_id=user_id, tenant_id=tenant_id)
+        mock_tool_file = ToolFile(
+            id=_gen_id(),
+            user_id=user_id,
+            tenant_id=tenant_id,
+            conversation_id=None,
+            file_key="test-file-key",
+            mimetype=mime_type,
+            original_url=None,
+            name=f"{_gen_id()}.png",
+            size=len(_PNG_DATA),
+        )
+        mock_get = mock.MagicMock(spec=ssrf_proxy.get, return_value=mock_response)
+        monkeypatch.setattr(ssrf_proxy, "get", mock_get)
+        mock_save_binary_string = mock.MagicMock(spec=file_saver.save_binary_string, return_value=mock_tool_file)
+        monkeypatch.setattr(file_saver, "save_binary_string", mock_save_binary_string)
+
+        file = file_saver.save_remote_url(_TEST_URL, FileType.IMAGE)
+        mock_save_binary_string.assert_called_once_with(
+            mock_response.content,
+            mime_type,
+            FileType.IMAGE,
+            extension_override=".png",
+        )
+        assert file == mock_tool_file
+
+
+def test_validate_extension_override():
+    class TestCase(NamedTuple):
+        extension_override: str | None
+        expected: str | None
+
+    cases = [TestCase(None, None), TestCase("", ""), ".png", ".png", ".tar.gz", ".tar.gz"]
+
+    for valid_ext_override in [None, "", ".png", ".tar.gz"]:
+        assert valid_ext_override == _validate_extension_override(valid_ext_override)
+
+    for invalid_ext_override in ["png", "tar.gz"]:
+        with pytest.raises(ValueError) as exc:
+            _validate_extension_override(invalid_ext_override)
+
+
+class TestExtractContentTypeAndExtension:
+    def test_with_both_content_type_and_extension(self):
+        content_type, extension = _extract_content_type_and_extension("https://example.com/image.jpg", "image/png")
+        assert content_type == "image/png"
+        assert extension == ".png"
+
+    def test_url_with_file_extension(self):
+        for content_type in [None, ""]:
+            content_type, extension = _extract_content_type_and_extension("https://example.com/image.png", content_type)
+            assert content_type == "image/png"
+            assert extension == ".png"
+
+    def test_response_with_content_type(self):
+        content_type, extension = _extract_content_type_and_extension("https://example.com/image", "image/png")
+        assert content_type == "image/png"
+        assert extension == ".png"
+
+    def test_no_content_type_and_no_extension(self):
+        for content_type in [None, ""]:
+            content_type, extension = _extract_content_type_and_extension("https://example.com/image", content_type)
+            assert content_type == "application/octet-stream"
+            assert extension == ".bin"
+
+
+class TestGetExtension:
+    def test_with_extension_override(self):
+        mime_type = "image/png"
+        for override in [".jpg", ""]:
+            extension = _get_extension(mime_type, override)
+            assert extension == override
+
+    def test_without_extension_override(self):
+        mime_type = "image/png"
+        extension = _get_extension(mime_type)
+        assert extension == ".png"

api/tests/unit_tests/core/workflow/nodes/llm/test_node.py

@@ -1,5 +1,8 @@
+import base64
+import uuid
 from collections.abc import Sequence
 from typing import Optional
+from unittest import mock
 
 import pytest
 
@@ -30,6 +33,7 @@ from core.workflow.nodes.llm.entities import (
     VisionConfig,
     VisionConfigOptions,
 )
+from core.workflow.nodes.llm.file_saver import LLMFileSaver
 from core.workflow.nodes.llm.node import LLMNode
 from models.enums import UserFrom
 from models.provider import ProviderType
@@ -49,8 +53,8 @@ class MockTokenBufferMemory:
 
 
 @pytest.fixture
-def llm_node():
+def llm_node_data() -> LLMNodeData:
-    data = LLMNodeData(
+    return LLMNodeData(
         title="Test LLM",
         model=ModelConfig(provider="openai", name="gpt-3.5-turbo", mode="chat", completion_params={}),
         prompt_template=[],
@@ -64,17 +68,11 @@ def llm_node():
             ),
         ),
     )
-    variable_pool = VariablePool(
+
-        system_variables={},
+
-        user_inputs={},
+@pytest.fixture
-    )
+def graph_init_params() -> GraphInitParams:
-    node = LLMNode(
+    return GraphInitParams(
-        id="1",
-        config={
-            "id": "1",
-            "data": data.model_dump(),
-        },
-        graph_init_params=GraphInitParams(
         tenant_id="1",
         app_id="1",
         workflow_type=WorkflowType.WORKFLOW,
@@ -84,8 +82,12 @@ def llm_node():
         user_from=UserFrom.ACCOUNT,
         invoke_from=InvokeFrom.SERVICE_API,
         call_depth=0,
-        ),
+    )
-        graph=Graph(
+
+
+@pytest.fixture
+def graph() -> Graph:
+    return Graph(
         root_node_id="1",
         answer_stream_generate_routes=AnswerStreamGenerateRoute(
             answer_dependencies={},
@@ -95,11 +97,36 @@ def llm_node():
             end_dependencies={},
             end_stream_variable_selector_mapping={},
         ),
-        ),
+    )
-        graph_runtime_state=GraphRuntimeState(
+
+
+@pytest.fixture
+def graph_runtime_state() -> GraphRuntimeState:
+    variable_pool = VariablePool(
+        system_variables={},
+        user_inputs={},
+    )
+    return GraphRuntimeState(
         variable_pool=variable_pool,
         start_at=0,
-        ),
+    )
+
+
+@pytest.fixture
+def llm_node(
+    llm_node_data: LLMNodeData, graph_init_params: GraphInitParams, graph: Graph, graph_runtime_state: GraphRuntimeState
+) -> LLMNode:
+    mock_file_saver = mock.MagicMock(spec=LLMFileSaver)
+    node = LLMNode(
+        id="1",
+        config={
+            "id": "1",
+            "data": llm_node_data.model_dump(),
+        },
+        graph_init_params=graph_init_params,
+        graph=graph,
+        graph_runtime_state=graph_runtime_state,
+        llm_file_saver=mock_file_saver,
     )
     return node
 
@@ -465,3 +492,167 @@ def test_handle_list_messages_basic(llm_node):
     assert len(result) == 1
     assert isinstance(result[0], UserPromptMessage)
     assert result[0].content == [TextPromptMessageContent(data="Hello, world")]
+
+
+@pytest.fixture
+def llm_node_for_multimodal(
+    llm_node_data, graph_init_params, graph, graph_runtime_state
+) -> tuple[LLMNode, LLMFileSaver]:
+    mock_file_saver: LLMFileSaver = mock.MagicMock(spec=LLMFileSaver)
+    node = LLMNode(
+        id="1",
+        config={
+            "id": "1",
+            "data": llm_node_data.model_dump(),
+        },
+        graph_init_params=graph_init_params,
+        graph=graph,
+        graph_runtime_state=graph_runtime_state,
+        llm_file_saver=mock_file_saver,
+    )
+    return node, mock_file_saver
+
+
+class TestLLMNodeSaveMultiModalImageOutput:
+    def test_llm_node_save_inline_output(self, llm_node_for_multimodal: tuple[LLMNode, LLMFileSaver]):
+        llm_node, mock_file_saver = llm_node_for_multimodal
+        content = ImagePromptMessageContent(
+            format="png",
+            base64_data=base64.b64encode(b"test-data").decode(),
+            mime_type="image/png",
+        )
+        mock_file = File(
+            id=str(uuid.uuid4()),
+            tenant_id="1",
+            type=FileType.IMAGE,
+            transfer_method=FileTransferMethod.TOOL_FILE,
+            related_id=str(uuid.uuid4()),
+            filename="test-file.png",
+            extension=".png",
+            mime_type="image/png",
+            size=9,
+        )
+        mock_file_saver.save_binary_string.return_value = mock_file
+        file = llm_node._save_multimodal_image_output(content=content)
+        assert llm_node._file_outputs == [mock_file]
+        assert file == mock_file
+        mock_file_saver.save_binary_string.assert_called_once_with(
+            data=b"test-data", mime_type="image/png", file_type=FileType.IMAGE
+        )
+
+    def test_llm_node_save_url_output(self, llm_node_for_multimodal: tuple[LLMNode, LLMFileSaver]):
+        llm_node, mock_file_saver = llm_node_for_multimodal
+        content = ImagePromptMessageContent(
+            format="png",
+            url="https://example.com/image.png",
+            mime_type="image/jpg",
+        )
+        mock_file = File(
+            id=str(uuid.uuid4()),
+            tenant_id="1",
+            type=FileType.IMAGE,
+            transfer_method=FileTransferMethod.TOOL_FILE,
+            related_id=str(uuid.uuid4()),
+            filename="test-file.png",
+            extension=".png",
+            mime_type="image/png",
+            size=9,
+        )
+        mock_file_saver.save_remote_url.return_value = mock_file
+        file = llm_node._save_multimodal_image_output(content=content)
+        assert llm_node._file_outputs == [mock_file]
+        assert file == mock_file
+        mock_file_saver.save_remote_url.assert_called_once_with(content.url, FileType.IMAGE)
+
+
+def test_llm_node_image_file_to_markdown(llm_node: LLMNode):
+    mock_file = mock.MagicMock(spec=File)
+    mock_file.generate_url.return_value = "https://example.com/image.png"
+    markdown = llm_node._image_file_to_markdown(mock_file)
+    assert markdown == "![](https://example.com/image.png)"
+
+
+class TestSaveMultimodalOutputAndConvertResultToMarkdown:
+    def test_str_content(self, llm_node_for_multimodal):
+        llm_node, mock_file_saver = llm_node_for_multimodal
+        gen = llm_node._save_multimodal_output_and_convert_result_to_markdown("hello world")
+        assert list(gen) == ["hello world"]
+        mock_file_saver.save_binary_string.assert_not_called()
+        mock_file_saver.save_remote_url.assert_not_called()
+
+    def test_text_prompt_message_content(self, llm_node_for_multimodal):
+        llm_node, mock_file_saver = llm_node_for_multimodal
+        gen = llm_node._save_multimodal_output_and_convert_result_to_markdown(
+            [TextPromptMessageContent(data="hello world")]
+        )
+        assert list(gen) == ["hello world"]
+        mock_file_saver.save_binary_string.assert_not_called()
+        mock_file_saver.save_remote_url.assert_not_called()
+
+    def test_image_content_with_inline_data(self, llm_node_for_multimodal, monkeypatch):
+        llm_node, mock_file_saver = llm_node_for_multimodal
+
+        image_raw_data = b"PNG_DATA"
+        image_b64_data = base64.b64encode(image_raw_data).decode()
+
+        mock_saved_file = File(
+            id=str(uuid.uuid4()),
+            tenant_id="1",
+            type=FileType.IMAGE,
+            transfer_method=FileTransferMethod.TOOL_FILE,
+            filename="test.png",
+            extension=".png",
+            size=len(image_raw_data),
+            related_id=str(uuid.uuid4()),
+            url="https://example.com/test.png",
+            storage_key="test_storage_key",
+        )
+        mock_file_saver.save_binary_string.return_value = mock_saved_file
+        gen = llm_node._save_multimodal_output_and_convert_result_to_markdown(
+            [
+                ImagePromptMessageContent(
+                    format="png",
+                    base64_data=image_b64_data,
+                    mime_type="image/png",
+                )
+            ]
+        )
+        yielded_strs = list(gen)
+        assert len(yielded_strs) == 1
+
+        # This assertion requires careful handling.
+        # `FILES_URL` settings can vary across environments, which might lead to fragile tests.
+        #
+        # Rather than asserting the complete URL returned by _save_multimodal_output_and_convert_result_to_markdown,
+        # we verify that the result includes the markdown image syntax and the expected file URL path.
+        expected_file_url_path = f"/files/tools/{mock_saved_file.related_id}.png"
+        assert yielded_strs[0].startswith("![](")
+        assert expected_file_url_path in yielded_strs[0]
+        assert yielded_strs[0].endswith(")")
+        mock_file_saver.save_binary_string.assert_called_once_with(
+            data=image_raw_data,
+            mime_type="image/png",
+            file_type=FileType.IMAGE,
+        )
+        assert mock_saved_file in llm_node._file_outputs
+
+    def test_unknown_content_type(self, llm_node_for_multimodal):
+        llm_node, mock_file_saver = llm_node_for_multimodal
+        gen = llm_node._save_multimodal_output_and_convert_result_to_markdown(frozenset(["hello world"]))
+        assert list(gen) == ["frozenset({'hello world'})"]
+        mock_file_saver.save_binary_string.assert_not_called()
+        mock_file_saver.save_remote_url.assert_not_called()
+
+    def test_unknown_item_type(self, llm_node_for_multimodal):
+        llm_node, mock_file_saver = llm_node_for_multimodal
+        gen = llm_node._save_multimodal_output_and_convert_result_to_markdown([frozenset(["hello world"])])
+        assert list(gen) == ["frozenset({'hello world'})"]
+        mock_file_saver.save_binary_string.assert_not_called()
+        mock_file_saver.save_remote_url.assert_not_called()
+
+    def test_none_content(self, llm_node_for_multimodal):
+        llm_node, mock_file_saver = llm_node_for_multimodal
+        gen = llm_node._save_multimodal_output_and_convert_result_to_markdown(None)
+        assert list(gen) == []
+        mock_file_saver.save_binary_string.assert_not_called()
+        mock_file_saver.save_remote_url.assert_not_called()

web/app/components/base/mermaid/index.tsx

@@ -476,15 +476,15 @@ const Flowchart = React.forwardRef((props: {
       'bg-white': currentTheme === Theme.light,
       'bg-slate-900': currentTheme === Theme.dark,
     }),
-    mermaidDiv: cn('mermaid cursor-pointer h-auto w-full relative', {
+    mermaidDiv: cn('mermaid relative h-auto w-full cursor-pointer', {
       'bg-white': currentTheme === Theme.light,
       'bg-slate-900': currentTheme === Theme.dark,
     }),
-    errorMessage: cn('py-4 px-[26px]', {
+    errorMessage: cn('px-[26px] py-4', {
       'text-red-500': currentTheme === Theme.light,
       'text-red-400': currentTheme === Theme.dark,
     }),
-    errorIcon: cn('w-6 h-6', {
+    errorIcon: cn('h-6 w-6', {
       'text-red-500': currentTheme === Theme.light,
       'text-red-400': currentTheme === Theme.dark,
     }),
@@ -492,7 +492,7 @@ const Flowchart = React.forwardRef((props: {
       'text-gray-700': currentTheme === Theme.light,
       'text-gray-300': currentTheme === Theme.dark,
     }),
-    themeToggle: cn('flex items-center justify-center w-10 h-10 rounded-full transition-all duration-300 shadow-md backdrop-blur-sm', {
+    themeToggle: cn('flex h-10 w-10 items-center justify-center rounded-full shadow-md backdrop-blur-sm transition-all duration-300', {
       'bg-white/80 hover:bg-white hover:shadow-lg text-gray-700 border border-gray-200': currentTheme === Theme.light,
       'bg-slate-800/80 hover:bg-slate-700 hover:shadow-lg text-yellow-300 border border-slate-600': currentTheme === Theme.dark,
     }),
@@ -501,7 +501,7 @@ const Flowchart = React.forwardRef((props: {
   // Style classes for look options
   const getLookButtonClass = (lookType: 'classic' | 'handDrawn') => {
     return cn(
-      'flex items-center justify-center mb-4 w-[calc((100%-8px)/2)] h-8 rounded-lg border border-components-option-card-option-border bg-components-option-card-option-bg cursor-pointer system-sm-medium text-text-secondary',
+      'system-sm-medium mb-4 flex h-8 w-[calc((100%-8px)/2)] cursor-pointer items-center justify-center rounded-lg border border-components-option-card-option-border bg-components-option-card-option-bg text-text-secondary',
       look === lookType && 'border-[1.5px] border-components-option-card-option-selected-border bg-components-option-card-option-selected-bg text-text-primary',
       currentTheme === Theme.dark && 'border-slate-600 bg-slate-800 text-slate-300',
       look === lookType && currentTheme === Theme.dark && 'border-blue-500 bg-slate-700 text-white',
@@ -512,7 +512,7 @@ const Flowchart = React.forwardRef((props: {
     <div ref={ref as React.RefObject<HTMLDivElement>} className={themeClasses.container}>
       <div className={themeClasses.segmented}>
         <div className="msh-segmented-group">
-          <label className="msh-segmented-item flex items-center space-x-1 m-2 w-[200px]">
+          <label className="msh-segmented-item m-2 flex w-[200px] items-center space-x-1">
             <div
               key='classic'
               className={getLookButtonClass('classic')}
@@ -534,7 +534,7 @@ const Flowchart = React.forwardRef((props: {
       <div ref={containerRef} style={{ position: 'absolute', visibility: 'hidden', height: 0, overflow: 'hidden' }} />
 
       {isLoading && !svgCode && (
-        <div className='py-4 px-[26px]'>
+        <div className='px-[26px] py-4'>
           <LoadingAnim type='text'/>
           {!isCodeComplete && (
             <div className="mt-2 text-sm text-gray-500">
@@ -546,7 +546,7 @@ const Flowchart = React.forwardRef((props: {
 
       {svgCode && (
         <div className={themeClasses.mermaidDiv} style={{ objectFit: 'cover' }} onClick={() => setImagePreviewUrl(svgCode)}>
-          <div className="absolute left-2 bottom-2 z-[100]">
+          <div className="absolute bottom-2 left-2 z-[100]">
             <button
               onClick={(e) => {
                 e.stopPropagation()