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>
This commit is contained in:
QuantumGhost
2025-04-30 17:28:02 +08:00
committed by GitHub
parent 6c9a9d344a
commit 349c3cf7b8
24 changed files with 971 additions and 191 deletions

View File

@@ -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

View File

@@ -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,
@@ -77,24 +87,25 @@ class ToolFileManager:
filepath = f"tools/{tenant_id}/{unique_filename}"
storage.save(filepath, file_binary)
tool_file = ToolFile(
user_id=user_id,
tenant_id=tenant_id,
conversation_id=conversation_id,
file_key=filepath,
mimetype=mimetype,
name=present_filename,
size=len(file_binary),
)
with Session(self._engine, expire_on_commit=False) as session:
tool_file = ToolFile(
user_id=user_id,
tenant_id=tenant_id,
conversation_id=conversation_id,
file_key=filepath,
mimetype=mimetype,
name=present_filename,
size=len(file_binary),
)
db.session.add(tool_file)
db.session.commit()
db.session.refresh(tool_file)
session.add(tool_file)
session.commit()
session.refresh(tool_file)
return tool_file
@staticmethod
def create_file_by_url(
self,
user_id: str,
tenant_id: str,
file_url: str,
@@ -119,24 +130,24 @@ class ToolFileManager:
filepath = f"tools/{tenant_id}/{filename}"
storage.save(filepath, blob)
tool_file = ToolFile(
user_id=user_id,
tenant_id=tenant_id,
conversation_id=conversation_id,
file_key=filepath,
mimetype=mimetype,
original_url=file_url,
name=filename,
size=len(blob),
)
with Session(self._engine, expire_on_commit=False) as session:
tool_file = ToolFile(
user_id=user_id,
tenant_id=tenant_id,
conversation_id=conversation_id,
file_key=filepath,
mimetype=mimetype,
original_url=file_url,
name=filename,
size=len(blob),
)
db.session.add(tool_file)
db.session.commit()
session.add(tool_file)
session.commit()
return tool_file
@staticmethod
def get_file_binary(id: str) -> Union[tuple[bytes, str], None]:
def get_file_binary(self, id: str) -> Union[tuple[bytes, str], None]:
"""
get file binary
@@ -144,13 +155,14 @@ class ToolFileManager:
:return: the binary of the file, mime type
"""
tool_file: ToolFile | None = (
db.session.query(ToolFile)
.filter(
ToolFile.id == id,
with Session(self._engine, expire_on_commit=False) as session:
tool_file: ToolFile | None = (
session.query(ToolFile)
.filter(
ToolFile.id == id,
)
.first()
)
.first()
)
if not tool_file:
return None
@@ -159,8 +171,7 @@ class ToolFileManager:
return blob, tool_file.mimetype
@staticmethod
def get_file_binary_by_message_file_id(id: str) -> Union[tuple[bytes, str], None]:
def get_file_binary_by_message_file_id(self, id: str) -> Union[tuple[bytes, str], None]:
"""
get file binary
@@ -168,33 +179,34 @@ class ToolFileManager:
:return: the binary of the file, mime type
"""
message_file: MessageFile | None = (
db.session.query(MessageFile)
.filter(
MessageFile.id == id,
with Session(self._engine, expire_on_commit=False) as session:
message_file: MessageFile | None = (
session.query(MessageFile)
.filter(
MessageFile.id == id,
)
.first()
)
.first()
)
# Check if message_file is not None
if message_file is not None:
# get tool file id
if message_file.url is not None:
tool_file_id = message_file.url.split("/")[-1]
# trim extension
tool_file_id = tool_file_id.split(".")[0]
# Check if message_file is not None
if message_file is not None:
# get tool file id
if message_file.url is not None:
tool_file_id = message_file.url.split("/")[-1]
# trim extension
tool_file_id = tool_file_id.split(".")[0]
else:
tool_file_id = None
else:
tool_file_id = None
else:
tool_file_id = None
tool_file: ToolFile | None = (
db.session.query(ToolFile)
.filter(
ToolFile.id == tool_file_id,
tool_file: ToolFile | None = (
session.query(ToolFile)
.filter(
ToolFile.id == tool_file_id,
)
.first()
)
.first()
)
if not tool_file:
return None
@@ -203,8 +215,7 @@ class ToolFileManager:
return blob, tool_file.mimetype
@staticmethod
def get_file_generator_by_tool_file_id(tool_file_id: str):
def get_file_generator_by_tool_file_id(self, tool_file_id: str):
"""
get file binary
@@ -212,13 +223,14 @@ class ToolFileManager:
:return: the binary of the file, mime type
"""
tool_file: ToolFile | None = (
db.session.query(ToolFile)
.filter(
ToolFile.id == tool_file_id,
with Session(self._engine, expire_on_commit=False) as session:
tool_file: ToolFile | None = (
session.query(ToolFile)
.filter(
ToolFile.id == tool_file_id,
)
.first()
)
.first()
)
if not tool_file:
return None, None
@@ -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)

View File

@@ -31,8 +31,8 @@ class ToolFileMessageTransformer:
# try to download image
try:
assert isinstance(message.message, ToolInvokeMessage.TextMessage)
file = ToolFileManager.create_file_by_url(
tool_file_manager = ToolFileManager()
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,