lian/banana-video
1
0
Fork 1
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

feat: 更新视频分析功能，支持提示词模板，提升复刻项目的灵活性和性能

Browse Source
This commit is contained in:
lk-eternal 2025-08-31 21:13:36 +08:00
parent c40c2b1039
commit c796ada84f
9 changed files with 27 additions and 44 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
.gitignore vendored
View File

@ -52,3 +52,4 @@ Thumbs.db
# Temporary files
*.tmp
*.temp
.qiniu_pythonsdk_hostscache.json

4
Dockerfile
View File

@ -43,5 +43,5 @@ EXPOSE 8000
HEALTHCHECK --interval=30s --timeout=30s --start-period=5s --retries=3 \
CMD curl -f http://localhost:8000/docs || exit 1
# 启动命令
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000", "--app-dir", "src"]
# 启动命令 - 使用多worker提升并发性能
CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "8000", "--app-dir", "src", "--workers", "4", "--loop", "uvloop"]

4
src/api/dependencies.py
View File

@ -90,8 +90,8 @@ class AIServiceImpl(AIService):
async def generate_video(self, frame_image_bytes: bytes, shot_prompt: str):
return await self.gemini_client.generate_video(frame_image_bytes, shot_prompt)
async def analyze_video(self, video_url: str):
return await self.gemini_client.analyze_video(video_url)
async def analyze_video(self, video_url: str, prompt_template: str):
return await self.gemini_client.analyze_video(video_url, prompt_template)
class StorageServiceImpl(StorageService):

8
src/api/routes/projects.py
View File

@ -434,8 +434,14 @@ async def replicate_from_video(
):
"""一键复刻：从视频URL生成项目、素材和分镜"""
try:
# 视频分析提示词模板不需要特定的占位符变量，因为视频内容直接传给AI模型
# 这里可以添加其他模板格式验证逻辑，如果需要的话
# 调用业务逻辑
result = await project_use_cases.replicate_from_video(request.video_url)
result = await project_use_cases.replicate_from_video(
request.video_url,
request.prompt_template
)
project = result["project"]
assets = result["assets"]

1
src/application/schemas/project.py
View File

@ -157,6 +157,7 @@ class ComposeVideoResponse(BaseModel):
class VideoReplicateRequest(BaseModel):
"""一键复刻请求模式"""
video_url: str = Field(..., description="要复刻的视频URL")
prompt_template: str = Field(..., description="视频分析提示词模板")
# 更新引用

5
src/application/use_cases/project_use_cases.py
View File

@ -590,12 +590,13 @@ class ProjectUseCases:
logger.error(f"生成视频失败: {e}")
raise ValueError(f"生成视频失败: {e}")
async def replicate_from_video(self, video_url: str) -> Dict[str, Any]:
async def replicate_from_video(self, video_url: str, prompt_template: str) -> Dict[str, Any]:
"""
一键复刻：从视频URL生成项目、素材和分镜
Args:
video_url: 要复刻的视频URL
prompt_template: 视频分析提示词模板
Returns:
包含project、assets、storyboards的字典
@ -605,7 +606,7 @@ class ProjectUseCases:
# 1. 使用Gemini分析视频内容
logger.info("正在分析视频内容...")
analysis_result = await self.ai_service.analyze_video(video_url)
analysis_result = await self.ai_service.analyze_video(video_url, prompt_template)
if not analysis_result:
raise ValueError("视频分析失败")

3
src/domain/services/ai_service.py
View File

@ -110,12 +110,13 @@ class AIService(ABC):
pass
@abstractmethod
async def analyze_video(self, video_url: str) -> Optional[Dict[str, Any]]:
async def analyze_video(self, video_url: str, prompt_template: str) -> Optional[Dict[str, Any]]:
"""
分析视频内容，提取关键素材帧和分镜关键帧
Args:
video_url: 视频URL
prompt_template: 视频分析提示词模板
Returns:
分析结果字典，包含key_assets_frames和key_storyboard_frames

39
src/infrastructure/external/gemini_client.py vendored
View File

@ -11,6 +11,7 @@ from google.genai import types
from ..config import settings
from ..utils import safe_json_loads
from .key_pool_manager import key_pool_manager
from ..services.template_service import TemplateService
from loguru import logger
import ssl
import urllib3
@ -217,7 +218,7 @@ class GeminiClient:
# 调用Veo-3.0 API生成视频
operation = self._current_client.models.generate_videos(
model="veo-3.0-fast-generate-preview", #veo-3.0-generate-preview
model="veo-3.0-generate-preview", #veo-3.0-fast-generate-preview
prompt=video_prompt,
image=image_input
)
@ -252,12 +253,13 @@ class GeminiClient:
logger.error(f"Veo-3.0视频生成失败: {e}")
raise e
async def analyze_video(self, video_url: str) -> Optional[Dict[str, Any]]:
async def analyze_video(self, video_url: str, prompt_template: str) -> Optional[Dict[str, Any]]:
"""
分析视频内容，提取关键素材帧和分镜关键帧
Args:
video_url: 视频URL
prompt_template: 视频分析提示词模板
Returns:
分析结果字典，包含key_assets_frames和key_storyboard_frames
@ -308,37 +310,8 @@ class GeminiClient:
return myfile
# 构建分析提示词
analysis_prompt = """
请仔细分析这个视频，并返回以下JSON格式的结果：
{
"title": "为这个视频生成一个简洁有吸引力的标题，不超过20个字符",
"script": "根据视频内容生成的完整剧本，包含对话、动作、场景描述等",
"key_assets_frames": [
{
"timestamp": "HH:MM:SS",
"name": "素材名称",
"description": "素材描述",
"tags": ["标签1", "标签2"]
}
],
"key_storyboard_frames": [
{
"timestamp": "HH:MM:SS",
"frame_prompt": "该帧画面描述",
"shot_prompt": "该关键帧到下一关键帧之间的剧情描述"
}
]
}
要求：
1. title: 根据视频主题和内容生成简洁有吸引力的标题，要能概括视频核心内容，不超过20个字符
2. script: 根据视频内容生成完整的剧本，包含场景描述、角色对话、动作指导等，要生动详细
3. key_assets_frames: 提取3-5个关键素材帧，包含视觉元素如角色、场景、道具、动物等
4. key_storyboard_frames: 提取分镜关键帧，约每8秒一帧
5. timestamp格式必须是HH:MM:SS
6. 确保返回的是有效的JSON格式，不要包含其他文字
"""
# 直接使用提示词模板作为分析提示词
analysis_prompt = prompt_template
myfile = self._execute_with_retry(_upload_and_analyze)

6
src/infrastructure/external/openrouter_client.py vendored
View File

@ -21,7 +21,7 @@ class OpenRouterClient:
async def generate_text(
self,
prompt: str,
model: str = "anthropic/claude-3.5-sonnet"
model: str = "google/gemini-2.5-pro"
) -> Optional[str]:
"""
生成文本内容
@ -66,7 +66,7 @@ class OpenRouterClient:
self,
prompt_template: str,
script_or_idea: str,
model: str = "anthropic/claude-3.5-sonnet"
model: str = "google/gemini-2.5-pro"
) -> Optional[Dict[str, Any]]:
"""
生成完整剧本和素材信息
@ -107,7 +107,7 @@ class OpenRouterClient:
text: str,
source_lang: str = "zh",
target_lang: str = "en",
model: str = "anthropic/claude-3.5-sonnet"
model: str = "google/gemini-2.5-pro"
) -> Optional[str]:
"""
翻译文本