模型端点配置化

.env.example

@@ -1,6 +1,16 @@
 # 阿里云 DashScope API Key (用于通义千问等模型)
 DASHSCOPE_API_KEY=your_dashscope_api_key_here
 
+# DashScope 模型请求端点
+# 视频生成端点
+DASHSCOPE_VIDEO_CREATE_URL=https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis
+# 文生图端点
+DASHSCOPE_IMAGE_T2I_URL=https://dashscope.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation
+# 图生图端点
+DASHSCOPE_IMAGE_I2I_URL=https://dashscope.aliyuncs.com/api/v1/services/aigc/image-generation/generation
+# 任务查询端点（必须包含 {task_id}）
+DASHSCOPE_TASK_QUERY_URL_TEMPLATE=https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}
+
 # 阿里云 Access Key (用于 OSS、视频超分等服务)
 ALIBABA_CLOUD_ACCESS_KEY_ID=your_aliyun_access_key_id_here
 ALIBABA_CLOUD_ACCESS_KEY_SECRET=your_aliyun_access_key_secret_here
@@ -13,3 +23,37 @@ OSS_BASE_PATH=comic_gen/
 # API 服务配置
 API_HOST=0.0.0.0
 API_PORT=8000
+
+# ===============================
+# 模型名称配置
+# ===============================
+# Wanx 默认模型
+WANX_T2V_MODEL_NAME_DEFAULT=wan2.5-t2v-preview
+WANX_I2V_MODEL_NAME_DEFAULT=wan2.6-i2v
+WANX_R2V_MODEL_NAME_DEFAULT=wan2.6-r2v
+
+# Wanx 走 HTTP 接口的模型列表（逗号分隔）
+WANX_HTTP_I2V_MODEL_NAMES=wan2.6-i2v,wan2.5-i2v
+WANX_HTTP_R2V_MODEL_NAMES=wan2.6-r2v
+
+# Wanx 图像模型
+WANX_IMAGE_T2I_MODEL_NAME_DEFAULT=wan2.6-t2i
+WANX_IMAGE_I2I_MODEL_NAME_DEFAULT=wan2.6-image
+WANX_IMAGE_HTTP_T2I_MODEL_NAMES=wan2.6-t2i
+WANX_IMAGE_HTTP_I2I_MODEL_NAMES=wan2.6-image
+WANX_IMAGE_FOUR_REF_MODELS=wan2.6-image
+
+# LLM 各环节模型
+LLM_PARSE_NOVEL_MODEL_NAME=qwen-max
+LLM_STORYBOARD_ANALYSIS_MODEL_NAME=qwen-max
+LLM_STYLE_RECOMMEND_MODEL_NAME=qwen-plus
+LLM_STORYBOARD_POLISH_MODEL_NAME=qwen-plus
+LLM_VIDEO_POLISH_MODEL_NAME=qwen-plus
+LLM_R2V_POLISH_MODEL_NAME=qwen-plus
+
+# 其他模型默认名称 / URL
+QWEN_VL_MODEL_NAME_DEFAULT=qwen-vl-plus
+DOUBAO_BASE_URL=https://ark.cn-beijing.volces.com/api/v3
+DOUBAO_MODEL_NAME_DEFAULT=doubao-seedance-1-0-pro-fast-251015
+KLING_BASE_URL=https://api.klingai.com/v1
+KLING_MODEL_NAME_DEFAULT=kling-v2-5-turbo

src/apps/comic_gen/api.py

@@ -14,6 +14,7 @@
 from .pipeline import ComicGenPipeline
 from .models import Script, VideoTask
 from .llm import ScriptProcessor
+from ...model_request_settings import MODEL_REQUEST_SETTINGS
 from ...utils.oss_utils import OSSImageUploader, sign_oss_urls_in_data
 from ...utils import setup_logging
 from fastapi.responses import JSONResponse
@@ -666,7 +667,7 @@ class CreateVideoTaskRequest(BaseModel):
     prompt_extend: bool = True
     negative_prompt: Optional[str] = None
     batch_size: int = 1
-    model: str = "wan2.6-i2v"
+    model: str = MODEL_REQUEST_SETTINGS.wanx_i2v_model_name_default
     shot_type: str = "single"  # 'single' or 'multi' (only for wan2.6-i2v)
     generation_mode: str = "i2v"  # 'i2v' (image-to-video) or 'r2v' (reference-to-video)
     reference_video_urls: List[str] = []  # Reference video URLs for R2V (max 3)
@@ -1733,4 +1734,3 @@ async def reorder_frames(script_id: str, request: ReorderFramesRequest):
     pipeline._save_data()
 
     return {"status": "success", "message": "Frames reordered", "frame_count": len(script.frames)}
-

src/apps/comic_gen/assets.py

@@ -5,6 +5,7 @@
 from urllib.parse import quote
 from .models import Character, Scene, Prop, GenerationStatus, ImageAsset, ImageVariant, MAX_VARIANTS_PER_ASSET
 from ...models.image import WanxImageModel
+from ...model_request_settings import MODEL_REQUEST_SETTINGS
 from ...utils import get_logger
 from ...utils.oss_utils import is_object_key
 
@@ -138,7 +139,7 @@ def generate_character(self, character: Character, generation_type: str = "all",
                         effective_generation_prompt = generation_prompt
                         if ref_image_path:
                             # Override to I2I model when using reference image
-                            effective_model_name = i2i_model_name or "wan2.6-image"
+                            effective_model_name = i2i_model_name or MODEL_REQUEST_SETTINGS.wanx_image_i2i_model_name_default
                             logger.debug(f"Reverse generation: Using I2I model {effective_model_name} with reference image")
 
                             # Enhance prompt for reverse generation to emphasize reference consistency (only if not already present)

src/apps/comic_gen/llm.py

@@ -9,12 +9,19 @@
 
 from .models import Script, Character, Scene, Prop, StoryboardFrame, GenerationStatus
 from ...utils import get_logger
+from ...model_request_settings import MODEL_REQUEST_SETTINGS
 
 logger = get_logger(__name__)
 
 class ScriptProcessor:
     def __init__(self, api_key: str = None):
         self._api_key = api_key
+        self.parse_novel_model_name = MODEL_REQUEST_SETTINGS.llm_parse_novel_model_name
+        self.storyboard_analysis_model_name = MODEL_REQUEST_SETTINGS.llm_storyboard_analysis_model_name
+        self.style_recommend_model_name = MODEL_REQUEST_SETTINGS.llm_style_recommend_model_name
+        self.storyboard_polish_model_name = MODEL_REQUEST_SETTINGS.llm_storyboard_polish_model_name
+        self.video_polish_model_name = MODEL_REQUEST_SETTINGS.llm_video_polish_model_name
+        self.r2v_polish_model_name = MODEL_REQUEST_SETTINGS.llm_r2v_polish_model_name
 
     @property
     def api_key(self):
@@ -40,8 +47,7 @@ def parse_novel(self, title: str, text: str) -> Script:
             dashscope.api_key = self.api_key
 
             response = dashscope.Generation.call(
-                # model='deepseek-v3.2',
-                model='qwen-max',
+                model=self.parse_novel_model_name,
                 prompt=prompt,
                 result_format='message',
             )
@@ -416,7 +422,7 @@ def analyze_script_for_styles(self, script_text: str) -> List[Dict[str, Any]]:
             dashscope.api_key = self.api_key
 
             response = dashscope.Generation.call(
-                model='qwen-plus',
+                model=self.style_recommend_model_name,
                 messages=[
                     {"role": "system", "content": system_prompt},
                     {"role": "user", "content": user_prompt}
@@ -671,7 +677,7 @@ def analyze_to_storyboard(self, text: str, entities_json: Dict[str, Any]) -> Lis
             dashscope.api_key = self.api_key
 
             response = dashscope.Generation.call(
-                model='qwen-max',
+                model=self.storyboard_analysis_model_name,
                 messages=[
                     {"role": "system", "content": system_prompt},
                     {"role": "user", "content": "请开始生成分镜帧列表，确保覆盖剧本中的所有内容。"}
@@ -792,7 +798,7 @@ def polish_storyboard_prompt(self, draft_prompt: str, assets: List[Dict[str, Any
             dashscope.api_key = self.api_key
 
             response = dashscope.Generation.call(
-                model='qwen-plus',
+                model=self.storyboard_polish_model_name,
                 prompt=system_prompt,
                 result_format='message',
                 response_format={'type': 'json_object'}
@@ -866,7 +872,7 @@ def polish_video_prompt(self, draft_prompt: str) -> Dict[str, str]:
             dashscope.api_key = self.api_key
 
             response = dashscope.Generation.call(
-                model='qwen-plus',
+                model=self.video_polish_model_name,
                 messages=[
                     {'role': 'system', 'content': system_prompt},
                     {'role': 'user', 'content': draft_prompt}
@@ -964,7 +970,7 @@ def polish_r2v_prompt(self, draft_prompt: str, slots: List[Dict[str, str]]) -> D
             dashscope.api_key = self.api_key
 
             response = dashscope.Generation.call(
-                model='qwen-plus',
+                model=self.r2v_polish_model_name,
                 messages=[
                     {'role': 'system', 'content': system_prompt},
                     {'role': 'user', 'content': draft_prompt}

src/apps/comic_gen/models.py

@@ -2,6 +2,7 @@
 from enum import Enum
 import time
 from pydantic import BaseModel, Field
+from ...model_request_settings import MODEL_REQUEST_SETTINGS
 
 class AspectRatio(str, Enum):
     SQUARE = "1:1"
@@ -76,7 +77,7 @@ class VideoTask(BaseModel):
     audio_url: Optional[str] = Field(None, description="URL of generated/uploaded audio")
     prompt_extend: bool = Field(True, description="Whether to use prompt extension")
     negative_prompt: Optional[str] = Field(None, description="Negative prompt")
-    model: str = Field("wan2.6-i2v", description="Model used for generation")
+    model: str = Field(MODEL_REQUEST_SETTINGS.wanx_i2v_model_name_default, description="Model used for generation")
     shot_type: str = Field("single", description="Shot type: 'single' or 'multi' (only for wan2.6-i2v)")
     generation_mode: str = Field("i2v", description="Generation mode: 'i2v' (image-to-video) or 'r2v' (reference-to-video)")
     reference_video_urls: List[str] = Field(default_factory=list, description="Reference video URLs for R2V generation (max 3)")
@@ -221,9 +222,9 @@ class StoryboardFrame(BaseModel):
 
 class ModelSettings(BaseModel):
     """Model selection settings for different generation stages"""
-    t2i_model: str = Field("wan2.6-t2i", description="Text-to-Image model for Assets")
-    i2i_model: str = Field("wan2.6-image", description="Image-to-Image model for Storyboard")
-    i2v_model: str = Field("wan2.6-i2v", description="Image-to-Video model for Motion")
+    t2i_model: str = Field(MODEL_REQUEST_SETTINGS.wanx_image_t2i_model_name_default, description="Text-to-Image model for Assets")
+    i2i_model: str = Field(MODEL_REQUEST_SETTINGS.wanx_image_i2i_model_name_default, description="Image-to-Image model for Storyboard")
+    i2v_model: str = Field(MODEL_REQUEST_SETTINGS.wanx_i2v_model_name_default, description="Image-to-Video model for Motion")
     character_aspect_ratio: str = Field("9:16", description="Aspect ratio for Characters (9:16, 16:9, 1:1)")
     scene_aspect_ratio: str = Field("16:9", description="Aspect ratio for Scenes (9:16, 16:9, 1:1)")
     prop_aspect_ratio: str = Field("1:1", description="Aspect ratio for Props (9:16, 16:9, 1:1)")

src/apps/comic_gen/pipeline.py

@@ -14,6 +14,7 @@
 from .video import VideoGenerator
 from .audio import AudioGenerator
 from .export import ExportManager
+from ...model_request_settings import MODEL_REQUEST_SETTINGS
 from ...utils import get_logger
 from ...utils.oss_utils import is_object_key
 from ...utils.system_check import get_ffmpeg_path, get_ffmpeg_install_instructions
@@ -1123,7 +1124,7 @@ def generate_motion_ref(
                             duration=duration,
                             created_at=time.time(),
                             generate_audio=bool(audio_url),
-                            model="wan2.6-i2v",
+                            model=MODEL_REQUEST_SETTINGS.wanx_i2v_model_name_default,
                             generation_mode="i2v"  # Image to video (motion reference)
                         )
 
@@ -1273,7 +1274,7 @@ def generate_video(self, script_id: str) -> Script:
         self._save_data()
         return script
 
-    def create_video_task(self, script_id: str, image_url: str, prompt: str, duration: int = 5, seed: int = None, resolution: str = "720p", generate_audio: bool = False, audio_url: str = None, prompt_extend: bool = True, negative_prompt: str = None, model: str = "wan2.6-i2v", frame_id: str = None, shot_type: str = "single", generation_mode: str = "i2v", reference_video_urls: list = None) -> Tuple[Script, str]:
+    def create_video_task(self, script_id: str, image_url: str, prompt: str, duration: int = 5, seed: int = None, resolution: str = "720p", generate_audio: bool = False, audio_url: str = None, prompt_extend: bool = True, negative_prompt: str = None, model: str = MODEL_REQUEST_SETTINGS.wanx_i2v_model_name_default, frame_id: str = None, shot_type: str = "single", generation_mode: str = "i2v", reference_video_urls: list = None) -> Tuple[Script, str]:
         """Creates a new video generation task."""
         script = self.get_script(script_id)
         if not script:
@@ -1283,7 +1284,7 @@ def create_video_task(self, script_id: str, image_url: str, prompt: str, duratio
 
         # If R2V mode is selected, use the R2V model
         if generation_mode == "r2v":
-            model = "wan2.6-r2v"
+            model = MODEL_REQUEST_SETTINGS.wanx_r2v_model_name_default
 
         # Snapshot the input image to ensure consistency
         snapshot_url = image_url
@@ -1687,7 +1688,7 @@ def create_asset_video_task(self, script_id: str, asset_id: str, asset_type: str
             prompt=prompt or f"Cinematic shot of {target_asset.name}",
             status="pending",
             duration=duration,
-            model="wan2.6-r2v", # Force R2V model
+            model=MODEL_REQUEST_SETTINGS.wanx_r2v_model_name_default, # Force configured R2V model
             created_at=time.time()
         )
 
@@ -1888,7 +1889,7 @@ def create_asset_video_task(self, script_id: str, asset_id: str, asset_type: str
             status="pending",
             duration=duration,
             resolution=resolution,
-            model="wan2.6-i2v", # Asset video uses I2V
+            model=MODEL_REQUEST_SETTINGS.wanx_i2v_model_name_default, # Asset video uses configured I2V model
             created_at=time.time()
         )
 

src/model_request_settings.py

@@ -0,0 +1,103 @@
+import os
+from pathlib import Path
+from typing import List
+
+from dotenv import load_dotenv
+
+
+_PROJECT_ROOT = Path(__file__).resolve().parent.parent
+_ENV_PATH = _PROJECT_ROOT / ".env"
+if _ENV_PATH.exists():
+    load_dotenv(_ENV_PATH, override=False)
+
+
+def _split_csv(value: str, default: List[str]) -> List[str]:
+    if not value:
+        return default
+    items = [item.strip() for item in value.split(",") if item.strip()]
+    return items or default
+
+
+class ModelRequestSettings:
+    # DashScope request URLs
+    dashscope_video_create_url: str = os.getenv(
+        "DASHSCOPE_VIDEO_CREATE_URL",
+        "https://dashscope.aliyuncs.com/api/v1/services/aigc/video-generation/video-synthesis",
+    )
+    dashscope_image_t2i_url: str = os.getenv(
+        "DASHSCOPE_IMAGE_T2I_URL",
+        "https://dashscope.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation",
+    )
+    dashscope_image_i2i_url: str = os.getenv(
+        "DASHSCOPE_IMAGE_I2I_URL",
+        "https://dashscope.aliyuncs.com/api/v1/services/aigc/image-generation/generation",
+    )
+    dashscope_task_query_url_template: str = os.getenv(
+        "DASHSCOPE_TASK_QUERY_URL_TEMPLATE",
+        "https://dashscope.aliyuncs.com/api/v1/tasks/{task_id}",
+    )
+
+    # Wanx model names
+    wanx_t2v_model_name_default: str = os.getenv(
+        "WANX_T2V_MODEL_NAME_DEFAULT",
+        "wan2.5-t2v-preview",
+    )
+    wanx_i2v_model_name_default: str = os.getenv(
+        "WANX_I2V_MODEL_NAME_DEFAULT",
+        "wan2.6-i2v",
+    )
+    wanx_r2v_model_name_default: str = os.getenv(
+        "WANX_R2V_MODEL_NAME_DEFAULT",
+        "wan2.6-r2v",
+    )
+    wanx_http_i2v_model_names: List[str] = _split_csv(
+        os.getenv("WANX_HTTP_I2V_MODEL_NAMES", ""),
+        ["wan2.6-i2v", "wan2.5-i2v"],
+    )
+    wanx_http_r2v_model_names: List[str] = _split_csv(
+        os.getenv("WANX_HTTP_R2V_MODEL_NAMES", ""),
+        ["wan2.6-r2v"],
+    )
+
+    # Wanx image model names
+    wanx_image_t2i_model_name_default: str = os.getenv(
+        "WANX_IMAGE_T2I_MODEL_NAME_DEFAULT",
+        "wan2.6-t2i",
+    )
+    wanx_image_i2i_model_name_default: str = os.getenv(
+        "WANX_IMAGE_I2I_MODEL_NAME_DEFAULT",
+        "wan2.6-image",
+    )
+    wanx_image_four_ref_models: List[str] = _split_csv(
+        os.getenv("WANX_IMAGE_FOUR_REF_MODELS", ""),
+        ["wan2.6-image"],
+    )
+    wanx_image_http_t2i_model_names: List[str] = _split_csv(
+        os.getenv("WANX_IMAGE_HTTP_T2I_MODEL_NAMES", ""),
+        ["wan2.6-t2i"],
+    )
+    wanx_image_http_i2i_model_names: List[str] = _split_csv(
+        os.getenv("WANX_IMAGE_HTTP_I2I_MODEL_NAMES", ""),
+        ["wan2.6-image"],
+    )
+
+    # LLM model names
+    llm_parse_novel_model_name: str = os.getenv("LLM_PARSE_NOVEL_MODEL_NAME", "qwen-max")
+    llm_storyboard_analysis_model_name: str = os.getenv("LLM_STORYBOARD_ANALYSIS_MODEL_NAME", "qwen-max")
+    llm_style_recommend_model_name: str = os.getenv("LLM_STYLE_RECOMMEND_MODEL_NAME", "qwen-plus")
+    llm_storyboard_polish_model_name: str = os.getenv("LLM_STORYBOARD_POLISH_MODEL_NAME", "qwen-plus")
+    llm_video_polish_model_name: str = os.getenv("LLM_VIDEO_POLISH_MODEL_NAME", "qwen-plus")
+    llm_r2v_polish_model_name: str = os.getenv("LLM_R2V_POLISH_MODEL_NAME", "qwen-plus")
+
+    # Other model providers
+    qwen_vl_model_name_default: str = os.getenv("QWEN_VL_MODEL_NAME_DEFAULT", "qwen-vl-plus")
+    doubao_base_url: str = os.getenv("DOUBAO_BASE_URL", "https://ark.cn-beijing.volces.com/api/v3")
+    doubao_model_name_default: str = os.getenv(
+        "DOUBAO_MODEL_NAME_DEFAULT",
+        "doubao-seedance-1-0-pro-fast-251015",
+    )
+    kling_base_url: str = os.getenv("KLING_BASE_URL", "https://api.klingai.com/v1")
+    kling_model_name_default: str = os.getenv("KLING_MODEL_NAME_DEFAULT", "kling-v2-5-turbo")
+
+
+MODEL_REQUEST_SETTINGS = ModelRequestSettings()

src/models/doubao.py

@@ -4,6 +4,7 @@
 import base64
 from typing import Tuple, Optional
 from .base import VideoGenModel
+from ..model_request_settings import MODEL_REQUEST_SETTINGS
 
 # Try to import Ark, handle if not installed (though user said they installed it)
 try:
@@ -17,14 +18,17 @@ class DoubaoModel(VideoGenModel):
     def __init__(self, config: dict):
         super().__init__(config)
         self.api_key = os.getenv("ARK_API_KEY")
-        self.model_name = config.get('params', {}).get('model_name', 'doubao-seedance-1-0-pro-fast-251015')
+        self.model_name = config.get('params', {}).get(
+            'model_name', MODEL_REQUEST_SETTINGS.doubao_model_name_default
+        )
+        self.base_url = MODEL_REQUEST_SETTINGS.doubao_base_url
 
         if not self.api_key:
             logger.warning("ARK_API_KEY not found in environment variables.")
 
         if Ark:
             self.client = Ark(
-                base_url="https://ark.cn-beijing.volces.com/api/v3",
+                base_url=self.base_url,
                 api_key=self.api_key
             )
         else: