mirror of
https://github.com/DrizzleTime/Foxel.git
synced 2026-05-12 02:20:28 +08:00
479 lines
18 KiB
C#
479 lines
18 KiB
C#
using System.Net.Http.Headers;
|
||
using System.Text.Json.Serialization;
|
||
using Foxel.Services.Configuration;
|
||
using Foxel.Utils;
|
||
|
||
namespace Foxel.Services.AI;
|
||
|
||
public class AiService(IHttpClientFactory httpClientFactory, IConfigService configService, ILogger<AiService> logger)
|
||
: IAiService
|
||
{
|
||
private HttpClient? _httpClient;
|
||
private string? _currentApiKey;
|
||
private string? _currentBaseUrl;
|
||
|
||
private HttpClient ConfigureHttpClient()
|
||
{
|
||
string apiKey = configService["AI:ApiKey"];
|
||
string baseUrl = configService["AI:ApiEndpoint"];
|
||
|
||
if (string.IsNullOrWhiteSpace(apiKey))
|
||
{
|
||
throw new InvalidOperationException("AI API Key 未配置或为空。请检查配置文件中的 AI:ApiKey 设置。");
|
||
}
|
||
|
||
if (string.IsNullOrWhiteSpace(baseUrl))
|
||
{
|
||
throw new InvalidOperationException("AI API Endpoint 未配置或为空。请检查配置文件中的 AI:ApiEndpoint 设置。");
|
||
}
|
||
|
||
if (!Uri.TryCreate(baseUrl, UriKind.Absolute, out Uri? validUri))
|
||
{
|
||
throw new InvalidOperationException($"AI API Endpoint 格式无效: {baseUrl}。请提供有效的 URL。");
|
||
}
|
||
|
||
if (_httpClient == null || _currentApiKey != apiKey || _currentBaseUrl != baseUrl)
|
||
{
|
||
_httpClient?.Dispose();
|
||
_httpClient = httpClientFactory.CreateClient();
|
||
_httpClient.BaseAddress = validUri;
|
||
_httpClient.DefaultRequestHeaders.Authorization = new AuthenticationHeaderValue("Bearer", apiKey);
|
||
_currentApiKey = apiKey;
|
||
_currentBaseUrl = baseUrl;
|
||
}
|
||
|
||
return _httpClient;
|
||
}
|
||
|
||
public async Task<(string title, string description)> AnalyzeImageAsync(string base64Image)
|
||
{
|
||
try
|
||
{
|
||
var client = ConfigureHttpClient();
|
||
string model = configService["AI:Model"];
|
||
var imageUrl = new ImageUrl
|
||
{
|
||
Url = $"data:image/jpeg;base64,{base64Image}"
|
||
};
|
||
|
||
var imageContent = new ImageUrlContent
|
||
{
|
||
Type = "image_url",
|
||
ImageUrl = imageUrl
|
||
};
|
||
|
||
var textContent = new TextContent
|
||
{
|
||
Type = "text",
|
||
Text = configService["AI:ImageAnalysisPrompt"] ??
|
||
"请详细分析这张图片，并提供全面的描述，以便用于向量嵌入和基于文本的图像搜索。描述需要包含：主体对象、场景环境、色彩特点、构图布局、风格特征、情绪氛围、细节特征等关键元素。请提供一个简短有力的标题，然后提供详细描述。\n\n请以JSON格式返回，格式如下：\n{\"title\": \"简短概括图片的核心内容\", \"description\": \"全面详细的描述，包含上述所有元素，使用丰富精确的词汇，避免笼统表达\"}\n\n请确保返回有效的JSON格式。"
|
||
};
|
||
|
||
var message = new ChatMessage
|
||
{
|
||
Role = "user",
|
||
Content = new MessageContent[] { imageContent, textContent }
|
||
};
|
||
|
||
var requestContent = new ChatCompletionRequest
|
||
{
|
||
Model = model,
|
||
Messages = [message],
|
||
Stream = false,
|
||
MaxTokens = 800,
|
||
Temperature = 0.5,
|
||
TopP = 0.8,
|
||
TopK = 50
|
||
};
|
||
|
||
var response = await client.PostAsJsonAsync("/v1/chat/completions", requestContent);
|
||
response.EnsureSuccessStatusCode();
|
||
|
||
var responseContent = await response.Content.ReadFromJsonAsync<AiResponse>();
|
||
if (responseContent?.Choices == null || responseContent.Choices.Length == 0)
|
||
{
|
||
return ("未能获取标题", "未能获取描述");
|
||
}
|
||
|
||
var aiMessage = responseContent.Choices[0].Message.Content;
|
||
return AiHelper.ExtractTitleAndDescription(aiMessage);
|
||
}
|
||
catch (Exception ex)
|
||
{
|
||
logger.LogError(ex, "AI分析图片时出错");
|
||
return ("处理失败", $"AI分析过程中发生错误: {ex.Message}");
|
||
}
|
||
}
|
||
|
||
public async Task<List<string>> MatchTagsAsync(string description, List<string> availableTags)
|
||
{
|
||
try
|
||
{
|
||
var client = ConfigureHttpClient();
|
||
|
||
if (availableTags.Count == 0)
|
||
return new List<string>();
|
||
|
||
string model = configService["AI:Model"];
|
||
var tagsText = string.Join(", ", availableTags);
|
||
|
||
string promptTemplate = configService["AI:TagMatchingPrompt"] ??
|
||
"以下是一组标签：[{tagsText}]。\n\n请从这些标签中严格选择与下面描述内容高度相关的标签（最多选择5个）。只选择确实匹配的标签，如果找不到完全匹配或高度相关的标签，宁可返回空数组也不要选择不太相关的标签。\n\n描述内容：{description}\n\n请以JSON格式返回，格式如下：\n{{\"tags\": [\"标签1\", \"标签2\", \"标签3\"]}}\n\n请确保返回有效的JSON格式前面不要加```，并且只包含确实匹配的标签名称。";
|
||
|
||
// 替换占位符
|
||
string promptText = promptTemplate
|
||
.Replace("{tagsText}", tagsText)
|
||
.Replace("{description}", description);
|
||
|
||
var textContent = new TextContent
|
||
{
|
||
Type = "text",
|
||
Text = promptText
|
||
};
|
||
|
||
var message = new ChatMessage
|
||
{
|
||
Role = "user",
|
||
Content = new MessageContent[] { textContent }
|
||
};
|
||
|
||
var requestContent = new ChatCompletionRequest
|
||
{
|
||
Model = model,
|
||
Messages = [message],
|
||
Stream = false,
|
||
MaxTokens = 200,
|
||
Temperature = 0.1, // 降低温度使结果更确定性
|
||
TopP = 0.95,
|
||
TopK = 50
|
||
};
|
||
|
||
var response = await client.PostAsJsonAsync("/v1/chat/completions", requestContent);
|
||
response.EnsureSuccessStatusCode();
|
||
|
||
var responseContent = await response.Content.ReadFromJsonAsync<AiResponse>();
|
||
if (responseContent?.Choices == null || responseContent.Choices.Length == 0)
|
||
{
|
||
return new List<string>();
|
||
}
|
||
|
||
var aiMessage = responseContent.Choices[0].Message.Content;
|
||
|
||
if (string.IsNullOrEmpty(aiMessage))
|
||
return new List<string>();
|
||
|
||
if (aiMessage.Contains("{") && aiMessage.Contains("}"))
|
||
{
|
||
try
|
||
{
|
||
int jsonStartIndex = aiMessage.IndexOf('{');
|
||
int jsonEndIndex = aiMessage.LastIndexOf('}') + 1;
|
||
|
||
if (jsonStartIndex >= 0 && jsonEndIndex > jsonStartIndex)
|
||
{
|
||
string jsonPart = aiMessage[jsonStartIndex..jsonEndIndex];
|
||
var options = new System.Text.Json.JsonSerializerOptions
|
||
{
|
||
PropertyNameCaseInsensitive = true
|
||
};
|
||
|
||
var result =
|
||
System.Text.Json.JsonSerializer.Deserialize<AiHelper.TagsResult>(jsonPart, options);
|
||
if (result is { Tags.Length: > 0 })
|
||
{
|
||
// 确保返回的标签真的在可用标签列表中
|
||
var matchedTags = new List<string>();
|
||
|
||
foreach (var tagName in result.Tags)
|
||
{
|
||
if (string.IsNullOrWhiteSpace(tagName))
|
||
continue;
|
||
|
||
// 找到大小写完全匹配的标签
|
||
var exactMatch = availableTags.FirstOrDefault(t =>
|
||
string.Equals(t, tagName, StringComparison.OrdinalIgnoreCase));
|
||
|
||
if (exactMatch != null)
|
||
{
|
||
matchedTags.Add(exactMatch);
|
||
}
|
||
}
|
||
|
||
return matchedTags.Distinct().ToList();
|
||
}
|
||
}
|
||
}
|
||
catch (System.Text.Json.JsonException)
|
||
{
|
||
// JSON解析失败，返回空列表
|
||
return new List<string>();
|
||
}
|
||
}
|
||
|
||
// 解析失败或没有找到匹配标签，返回空列表
|
||
return new List<string>();
|
||
}
|
||
catch (Exception ex)
|
||
{
|
||
logger.LogError(ex, "AI匹配标签时出错");
|
||
return new List<string>();
|
||
}
|
||
}
|
||
|
||
public async Task<List<string>> GenerateTagsFromImageAsync(string base64Image, List<string> availableTags,
|
||
bool allowNewTags = false)
|
||
{
|
||
try
|
||
{
|
||
// 获取配置好的 HttpClient
|
||
var client = ConfigureHttpClient();
|
||
|
||
string model = configService["AI:Model"];
|
||
|
||
var imageUrl = new ImageUrl
|
||
{
|
||
Url = $"data:image/jpeg;base64,{base64Image}"
|
||
};
|
||
|
||
var imageContent = new ImageUrlContent
|
||
{
|
||
Type = "image_url",
|
||
ImageUrl = imageUrl
|
||
};
|
||
|
||
string promptText;
|
||
|
||
if (allowNewTags)
|
||
{
|
||
// 获取配置的标签生成提示词，如果没有则使用默认提示词
|
||
string defaultPrompt = configService["AI:TagGenerationPrompt"] ??
|
||
"请为图片生成5个最相关的标签，每个标签应该是简短且描述性的词语或短语。\n\n请以JSON格式返回，格式如下：\n{\"tags\": [\"标签1\", \"标签2\", \"标签3\", \"标签4\", \"标签5\"]}\n\n请确保返回有效的JSON格式。";
|
||
|
||
// 如果允许新标签，则提供现有标签作为参考，但允许生成新标签
|
||
promptText = availableTags.Count > 0
|
||
? $"可以参考这些现有标签：[{string.Join(", ", availableTags)}]，但也可以生成其他与图片内容相关的新标签。\n\n{defaultPrompt}"
|
||
: defaultPrompt;
|
||
}
|
||
else
|
||
{
|
||
// 如果不允许新标签，则只能从已有标签中选择
|
||
if (availableTags.Count == 0)
|
||
return new List<string>();
|
||
|
||
var tagsText = string.Join(", ", availableTags);
|
||
string templatePrompt = configService["AI:TagMatchingPrompt"] ??
|
||
"以下是一组标签：[{tagsText}]。\n\n请从这些标签中严格选择与图片内容高度相关的标签（最多选择5个）。只选择确实匹配的标签，如果找不到完全匹配或高度相关的标签，宁可返回空数组也不要选择不太相关的标签。\n\n请以JSON格式返回，格式如下：\n{{\"tags\": [\"标签1\", \"标签2\", \"标签3\"]}}\n\n请确保返回有效的JSON格式，并且只包含上述列表中的标签名称。";
|
||
|
||
promptText = templatePrompt.Replace("{tagsText}", tagsText);
|
||
}
|
||
|
||
var textContent = new TextContent
|
||
{
|
||
Type = "text",
|
||
Text = promptText
|
||
};
|
||
|
||
var message = new ChatMessage
|
||
{
|
||
Role = "user",
|
||
Content = new MessageContent[] { imageContent, textContent }
|
||
};
|
||
|
||
var requestContent = new ChatCompletionRequest
|
||
{
|
||
Model = model,
|
||
Messages = [message],
|
||
Stream = false,
|
||
MaxTokens = 200,
|
||
Temperature = 0.1, // 降低温度使结果更确定性
|
||
TopP = 0.95,
|
||
TopK = 50
|
||
};
|
||
|
||
var response = await client.PostAsJsonAsync("/v1/chat/completions", requestContent);
|
||
response.EnsureSuccessStatusCode();
|
||
|
||
var responseContent = await response.Content.ReadFromJsonAsync<AiResponse>();
|
||
if (responseContent?.Choices == null || responseContent.Choices.Length == 0)
|
||
{
|
||
return new List<string>();
|
||
}
|
||
|
||
var aiMessage = responseContent.Choices[0].Message.Content;
|
||
|
||
if (string.IsNullOrEmpty(aiMessage))
|
||
return new List<string>();
|
||
|
||
if (aiMessage.Contains("{") && aiMessage.Contains("}"))
|
||
{
|
||
try
|
||
{
|
||
int jsonStartIndex = aiMessage.IndexOf('{');
|
||
int jsonEndIndex = aiMessage.LastIndexOf('}') + 1;
|
||
|
||
if (jsonStartIndex >= 0 && jsonEndIndex > jsonStartIndex)
|
||
{
|
||
string jsonPart = aiMessage[jsonStartIndex..jsonEndIndex];
|
||
var options = new System.Text.Json.JsonSerializerOptions
|
||
{
|
||
PropertyNameCaseInsensitive = true
|
||
};
|
||
|
||
var result =
|
||
System.Text.Json.JsonSerializer.Deserialize<AiHelper.TagsResult>(jsonPart, options);
|
||
if (result is { Tags.Length: > 0 })
|
||
{
|
||
var matchedTags = new List<string>();
|
||
|
||
foreach (var tagName in result.Tags)
|
||
{
|
||
if (string.IsNullOrWhiteSpace(tagName))
|
||
continue;
|
||
|
||
// 如果允许新标签，直接添加
|
||
if (allowNewTags)
|
||
{
|
||
matchedTags.Add(tagName.Trim());
|
||
}
|
||
else
|
||
{
|
||
// 否则只添加已有标签列表中的标签
|
||
var exactMatch = availableTags.FirstOrDefault(t =>
|
||
string.Equals(t, tagName, StringComparison.OrdinalIgnoreCase));
|
||
|
||
if (exactMatch != null)
|
||
{
|
||
matchedTags.Add(exactMatch);
|
||
}
|
||
}
|
||
}
|
||
|
||
return matchedTags.Distinct().ToList();
|
||
}
|
||
}
|
||
}
|
||
catch (System.Text.Json.JsonException)
|
||
{
|
||
// JSON解析失败，返回空列表
|
||
return new List<string>();
|
||
}
|
||
}
|
||
|
||
// 解析失败或没有找到匹配标签，返回空列表
|
||
return new List<string>();
|
||
}
|
||
catch (Exception ex)
|
||
{
|
||
logger.LogError(ex, "AI从图片生成标签时出错");
|
||
return new List<string>();
|
||
}
|
||
}
|
||
|
||
public async Task<float[]> GetEmbeddingAsync(string text)
|
||
{
|
||
try
|
||
{
|
||
// 获取配置好的 HttpClient
|
||
var client = ConfigureHttpClient();
|
||
|
||
string model = configService["AI:EmbeddingModel"];
|
||
|
||
var requestContent = new
|
||
{
|
||
model,
|
||
input = text,
|
||
encoding_format = "float"
|
||
};
|
||
|
||
var response = await client.PostAsJsonAsync("/v1/embeddings", requestContent);
|
||
response.EnsureSuccessStatusCode();
|
||
|
||
var embedResult = await response.Content.ReadFromJsonAsync<EmbeddingResponse>();
|
||
if (embedResult?.Data == null || embedResult.Data.Length == 0)
|
||
{
|
||
logger.LogWarning("嵌入向量API返回空结果");
|
||
return Array.Empty<float>();
|
||
}
|
||
|
||
return embedResult.Data[0].Embedding;
|
||
}
|
||
catch (Exception ex)
|
||
{
|
||
logger.LogError(ex, "获取嵌入向量时出错");
|
||
return Array.Empty<float>();
|
||
}
|
||
}
|
||
|
||
// 从EmbeddingService移植的私有记录类
|
||
private record EmbeddingResponse
|
||
{
|
||
[JsonPropertyName("data")] public EmbeddingData[] Data { get; set; } = Array.Empty<EmbeddingData>();
|
||
}
|
||
|
||
private record EmbeddingData
|
||
{
|
||
[JsonPropertyName("embedding")] public float[] Embedding { get; set; } = Array.Empty<float>();
|
||
}
|
||
|
||
private class AiResponse
|
||
{
|
||
[JsonPropertyName("choices")] public Choice[] Choices { get; set; } = Array.Empty<Choice>();
|
||
}
|
||
|
||
private class Choice
|
||
{
|
||
[JsonPropertyName("message")] public Message Message { get; set; } = new Message();
|
||
}
|
||
|
||
private class Message
|
||
{
|
||
[JsonPropertyName("content")] public string Content { get; set; } = string.Empty;
|
||
}
|
||
}
|
||
|
||
public class ChatCompletionRequest
|
||
{
|
||
[JsonPropertyName("model")] public string Model { get; set; } = string.Empty;
|
||
|
||
[JsonPropertyName("messages")] public ChatMessage[] Messages { get; set; } = Array.Empty<ChatMessage>();
|
||
|
||
[JsonPropertyName("stream")] public bool Stream { get; set; }
|
||
|
||
[JsonPropertyName("max_tokens")] public int MaxTokens { get; set; }
|
||
|
||
[JsonPropertyName("temperature")] public double Temperature { get; set; }
|
||
|
||
[JsonPropertyName("top_p")] public double TopP { get; set; }
|
||
|
||
[JsonPropertyName("top_k")] public int TopK { get; set; }
|
||
}
|
||
|
||
public class ChatMessage
|
||
{
|
||
[JsonPropertyName("role")] public string Role { get; set; } = string.Empty;
|
||
|
||
[JsonPropertyName("content")] public MessageContent[] Content { get; set; } = Array.Empty<MessageContent>();
|
||
}
|
||
|
||
[JsonPolymorphic(TypeDiscriminatorPropertyName = "type")]
|
||
[JsonDerivedType(typeof(TextContent), typeDiscriminator: "text")]
|
||
[JsonDerivedType(typeof(ImageUrlContent), typeDiscriminator: "image_url")]
|
||
public abstract class MessageContent
|
||
{
|
||
[JsonPropertyName("type")] public string Type { get; set; } = string.Empty;
|
||
}
|
||
|
||
public class TextContent : MessageContent
|
||
{
|
||
[JsonPropertyName("text")] public string Text { get; set; } = string.Empty;
|
||
}
|
||
|
||
public class ImageUrlContent : MessageContent
|
||
{
|
||
[JsonPropertyName("image_url")] public ImageUrl ImageUrl { get; set; } = new();
|
||
}
|
||
|
||
public class ImageUrl
|
||
{
|
||
[JsonPropertyName("url")] public string Url { get; set; } = string.Empty;
|
||
} |