# 本地大模型配置文件 models: # VLM模型(用于图片理解和修复) qwen2_vl: name: "Qwen2.5-VL-72B-Instruct-AWQ" api_base: "http://10.192.72.12:9991/v1" api_key: "${YUSYS_MULTIMODAL_API_KEY}" model_id: "Qwen2.5-VL-72B-Instruct-AWQ" type: "vlm" default_params: temperature: 0.1 max_tokens: 4096 timeout: 180 llava: name: "LLaVA-v1.6-34B" api_base: "http://localhost:11434/v1" api_key: "" model_id: "llava:34b" type: "vlm" default_params: temperature: 0.1 max_tokens: 4096 timeout: 180 cogvlm: name: "CogVLM2-19B" api_base: "http://localhost:11434/v1" api_key: "" model_id: "cogvlm2:19b" type: "vlm" default_params: temperature: 0.1 max_tokens: 4096 timeout: 180 # 图片生成模型 dashscope_wanx: name: "通义万相-风格重绘" api_base: "https://dashscope.aliyuncs.com/api/v1/services/aigc/image-generation/generation" api_key: "${DASHSCOPE_API_KEY}" model_id: "wanx-style-repaint-v1" type: "image_generation" generation_type: "style_repaint" # 风格重绘 default_params: style_index: 6 timeout: 300 poll_interval: 5 modelscope_qwen: name: "ModelScope-Qwen文生图" api_base: "https://api-inference.modelscope.cn/v1/images/generations" api_key: "${MODELSCOPE_API_KEY}" model_id: "Qwen/Qwen-Image" type: "image_generation" generation_type: "text_to_image" # 文生图 default_params: timeout: 300 poll_interval: 5 dashscope_flux: name: "通义万相-FLUX" api_base: "https://dashscope.aliyuncs.com/api/v1/services/aigc/text2image/image-synthesis" api_key: "${DASHSCOPE_API_KEY}" model_id: "flux-schnell" type: "image_generation" generation_type: "text_to_image" # 文生图 default_params: size: "1024*1024" timeout: 300 poll_interval: 5 # 新增背景生成模型 dashscope_background: name: "通义万相-背景生成" api_base: "https://dashscope.aliyuncs.com/api/v1/services/aigc/background-generation/generation/" api_key: "${DASHSCOPE_API_KEY}" model_id: "wanx-background-generation-v2" type: "image_generation" generation_type: "background_generation" # 背景生成 default_params: model_version: "v3" n: 1 timeout: 300 poll_interval: 5 # 提示词模板 prompts: # 原有的VLM提示词... photo_analysis: name: "照片分析" type: "vlm" template: | 请仔细观察这张照片,分析以下内容: 1. 照片的拍摄环境和背景 2. 人物的外貌特征和着装 3. 照片的拍摄时间推测(基于服装、环境等) 4. 照片的保存状况(是否有破损、褪色等) 5. 照片的历史价值和意义 请用中文详细描述,分条列出分析结果。 # 图片生成提示词 photo_style_repaint: name: "照片风格重绘" type: "image_generation" compatible_models: ["dashscope_wanx"] template: | 对输入的照片进行风格重绘,保持人物特征和基本构图,应用指定的艺术风格。 风格选项:0=复古漫画, 1=3D童话, 2=二次元, 3=小清新, 4=未来科技, 5=国画古风, 6=油画质感 text_to_image_simple: name: "简单文生图" type: "image_generation" compatible_models: ["modelscope_qwen", "dashscope_flux"] template: | 根据文本描述生成高质量图片。请提供详细的场景描述、人物特征、光线效果和艺术风格要求。 photo_restoration: name: "AI照片修复" type: "image_generation" compatible_models: ["dashscope_wanx"] template: | 修复老旧照片:去除折痕、划痕、污渍,提升清晰度,保持原有人物特征。 {prompt} # 新增背景生成提示词 background_classroom: name: "教室背景生成" type: "image_generation" compatible_models: ["dashscope_background"] template: | 1980年代中国大学教室场景:后方是深绿色黑板,略有粉笔痕迹,前方是深褐色木质课桌,教室光线明亮柔和,营造温馨的学习氛围 background_landscape: name: "风景背景生成" type: "image_generation" compatible_models: ["dashscope_background"] template: | {prompt} background_studio: name: "影棚背景生成" type: "image_generation" compatible_models: ["dashscope_background"] template: | 专业摄影影棚背景,柔和灯光,纯色或渐变背景,适合人像摄影 # 预设风格配置 style_presets: dashscope_styles: - {index: 0, name: "复古漫画", description: "复古漫画风格,线条清晰,色彩饱和"} - {index: 1, name: "3D童话", description: "3D童话风格,梦幻可爱,立体感强"} - {index: 2, name: "二次元", description: "二次元动漫风格,日式插画感"} - {index: 3, name: "小清新", description: "小清新风格,色彩柔和,清新自然"} - {index: 4, name: "未来科技", description: "未来科技风格,金属质感,科幻氛围"} - {index: 5, name: "国画古风", description: "中国风水墨画风格,古典雅致"} - {index: 6, name: "油画质感", description: "油画质感,色彩丰富,艺术感强"} # 默认配置 default: vlm_model: "qwen2_vl" image_gen_model: "dashscope_wanx" prompt: "photo_analysis" style_index: -1 # -1表示使用使用自定义风格, 0-6表示使用对应风格