使用Gemini Vision API分析皮肤照片，返回结构化分析结果。当需要实现AI皮肤分析功能时自动调用此技能。
调用Gemini 2.0 Flash Vision API分析用户面部照片，返回多维度皮肤评估结果。
image: UIImage - 用户面部照片userProfile: UserProfile? - 用户基础信息（可选）
struct GeminiConfig {
static let model = "gemini-2.0-flash"
static let baseURL = "https://generativelanguage.googleapis.com/v1beta"
static let endpoint = "/models/\(model):generateContent"
}
func buildRequest(image: UIImage, apiKey: String) -> URLRequest {
let url = URL(string: "\(GeminiConfig.baseURL)\(GeminiConfig.endpoint)?key=\(apiKey)")!
var request = URLRequest(url: url)
request.httpMethod = "POST"
request.setValue("application/json", forHTTPHeaderField: "Content-Type")
let imageData = image.jpegData(compressionQuality: 0.8)!
let base64Image = imageData.base64EncodedString()
let body: [String: Any] = [
"contents": [[
"parts": [
["text": promptTemplate],
["inline_data": [
"mime_type": "image/jpeg",
"data": base64Image
]]
]
]],
"generationConfig": [
"temperature": 0.1,
"topP": 0.8,
"maxOutputTokens": 2048
]
]
request.httpBody = try? JSONSerialization.data(withJSONObject: body)
return request
}
你是一位专业皮肤科医生，拥有20年临床经验。请仔细分析这张面部照片。
## 分析要求
请评估以下维度，以JSON格式返回结果：
1. skinType: 肤质类型
- "dry": 干性（紧绷、脱皮、细纹明显）
- "oily": 油性（T区泛油、毛孔明显）
- "combination": 混合性（T区油、两颊干）
- "sensitive": 敏感性（泛红、易过敏）
2. skinAge: 皮肤表观年龄（数字）
3. overallScore: 综合健康评分（0-100）
4. issues: 问题评分（每项0-10，0为无问题，10为严重）
- spots: 色斑/色素沉着
- acne: 痘痘/粉刺
- pores: 毛孔粗大
- wrinkles: 皱纹/细纹
- redness: 红血丝/泛红
- evenness: 肤色不均
- texture: 纹理粗糙
5. regions: 区域评分（0-100，100为最佳）
- tZone: T区（额头+鼻子）
- leftCheek: 左脸颊
- rightCheek: 右脸颊
- eyeArea: 眼周
- chin: 下巴
6. recommendations: 护肤建议数组（3-5条）
## 输出格式
仅返回JSON对象，不要包含其他文字或markdown代码块标记。
struct SkinAnalysis: Codable {
let skinType: SkinType
let skinAge: Int
let overallScore: Int
let issues: IssueScores
let regions: RegionScores
let recommendations: [String]
struct IssueScores: Codable {
let spots: Int
let acne: Int
let pores: Int
let wrinkles: Int
let redness: Int
let evenness: Int
let texture: Int
}
struct RegionScores: Codable {
let tZone: Int
let leftCheek: Int
let rightCheek: Int
let eyeArea: Int
let chin: Int
}
}
enum SkinType: String, Codable {
case dry, oily, combination, sensitive
}
enum AnalysisError: Error {
case invalidImage
case networkError(Error)
case apiError(String)
case parseError
case rateLimited
case unauthorized
}
// 重试策略
func analyzeWithRetry(image: UIImage, maxRetries: Int = 2) async throws -> SkinAnalysis {
var lastError: Error?
for attempt in 0..<maxRetries {
do {
return try await analyze(image: image)
} catch AnalysisError.rateLimited {
try await Task.sleep(nanoseconds: UInt64(pow(2.0, Double(attempt))) * 1_000_000_000)
lastError = AnalysisError.rateLimited
} catch {
lastError = error
break
}
}
throw lastError ?? AnalysisError.networkError(NSError())
}