You've already forked Nano-Banana-AI-Image-Editor
初始化提交
This commit is contained in:
327
v1/src/services/geminiService.ts
Normal file
327
v1/src/services/geminiService.ts
Normal file
@@ -0,0 +1,327 @@
|
||||
import { GoogleGenAI } from '@google/genai'
|
||||
|
||||
// 注意：在生产环境中，这应该通过后端代理处理
|
||||
const API_KEY = import.meta.env.VITE_GEMINI_API_KEY || 'demo-key'
|
||||
const genAI = new GoogleGenAI({ apiKey: API_KEY })
|
||||
|
||||
export interface GenerationRequest {
|
||||
prompt: string
|
||||
referenceImages?: Blob[] // Blob数组
|
||||
temperature?: number
|
||||
seed?: number
|
||||
}
|
||||
|
||||
export interface EditRequest {
|
||||
instruction: string
|
||||
originalImage: Blob // Blob
|
||||
referenceImages?: Blob[] // Blob数组
|
||||
maskImage?: Blob // Blob
|
||||
temperature?: number
|
||||
seed?: number
|
||||
}
|
||||
|
||||
export interface UsageMetadata {
|
||||
totalTokenCount?: number
|
||||
promptTokenCount?: number
|
||||
candidatesTokenCount?: number
|
||||
}
|
||||
|
||||
export interface SegmentationRequest {
|
||||
image: Blob // Blob
|
||||
query: string // "像素(x,y)处的对象" 或 "红色汽车"
|
||||
}
|
||||
|
||||
export class GeminiService {
|
||||
// 将Blob转换为base64的辅助函数
|
||||
private async blobToBase64(blob: Blob): Promise<string> {
|
||||
return new Promise((resolve, reject) => {
|
||||
const reader = new FileReader();
|
||||
reader.onload = () => {
|
||||
const result = reader.result as string;
|
||||
const base64 = result.split(',')[1]; // Remove data:image/png;base64, prefix
|
||||
resolve(base64);
|
||||
};
|
||||
reader.onerror = reject;
|
||||
reader.readAsDataURL(blob);
|
||||
});
|
||||
}
|
||||
|
||||
async generateImage(request: GenerationRequest): Promise<{ images: Blob[]; usageMetadata?: any }> {
|
||||
try {
|
||||
const contents: any[] = [{ text: request.prompt }]
|
||||
|
||||
// 如果提供了参考图像则添加
|
||||
if (request.referenceImages && request.referenceImages.length > 0) {
|
||||
// 将Blob转换为base64以发送到API
|
||||
const base64Images = await Promise.all(
|
||||
request.referenceImages.map(blob => this.blobToBase64(blob))
|
||||
);
|
||||
|
||||
base64Images.forEach(image => {
|
||||
contents.push({
|
||||
inlineData: {
|
||||
mimeType: 'image/png',
|
||||
data: image,
|
||||
},
|
||||
})
|
||||
})
|
||||
}
|
||||
|
||||
const response = await genAI.models.generateContent({
|
||||
model: 'gemini-2.5-flash-image-preview',
|
||||
contents,
|
||||
})
|
||||
|
||||
// 检查是否有被禁止的内容
|
||||
if (response.candidates && response.candidates.length > 0) {
|
||||
const candidate = response.candidates[0]
|
||||
if (candidate.finishReason === 'PROHIBITED_CONTENT') {
|
||||
throw new Error('内容被禁止：您的请求包含不允许的内容。请尝试其他提示。')
|
||||
}
|
||||
if (candidate.finishReason === 'IMAGE_SAFETY') {
|
||||
throw new Error('内容被禁止：您的请求包含不允许的内容。请尝试其他提示。')
|
||||
}
|
||||
// 检查finishReason为STOP但没有inlineData的情况
|
||||
if (candidate.finishReason === 'STOP') {
|
||||
// 检查是否有inlineData
|
||||
let hasInlineData = false;
|
||||
if (candidate.content && candidate.content.parts) {
|
||||
for (const part of candidate.content.parts) {
|
||||
if (part.inlineData) {
|
||||
hasInlineData = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 如果没有inlineData，则抛出错误
|
||||
if (!hasInlineData && candidate.content && candidate.content.parts && candidate.content.parts.length > 0) {
|
||||
throw new Error(candidate.content.parts[0].text || '生成失败：未返回图像数据');
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
const images: Blob[] = []
|
||||
|
||||
// 检查响应是否存在以及是否有内容
|
||||
if (response.candidates && response.candidates.length > 0 && response.candidates[0].content && response.candidates[0].content.parts) {
|
||||
for (const part of response.candidates[0].content.parts) {
|
||||
if (part.inlineData) {
|
||||
// 将返回的base64数据转换为Blob
|
||||
const byteString = atob(part.inlineData.data);
|
||||
const mimeString = part.inlineData.mimeType || 'image/png';
|
||||
const ab = new ArrayBuffer(byteString.length);
|
||||
const ia = new Uint8Array(ab);
|
||||
for (let i = 0; i < byteString.length; i++) {
|
||||
ia[i] = byteString.charCodeAt(i);
|
||||
}
|
||||
const blob = new Blob([ab], { type: mimeString });
|
||||
images.push(blob);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 获取usageMetadata（如果存在）
|
||||
const usageMetadata = response.usageMetadata
|
||||
|
||||
return { images, usageMetadata }
|
||||
} catch (error) {
|
||||
console.error('生成图像时出错:', error)
|
||||
if (error instanceof Error && error.message) {
|
||||
throw error
|
||||
}
|
||||
throw new Error(`生成图像失败: ${error instanceof Error ? error.message : '未知错误'}`)
|
||||
}
|
||||
}
|
||||
|
||||
async editImage(request: EditRequest): Promise<{ images: Blob[]; usageMetadata?: any }> {
|
||||
try {
|
||||
// 将Blob转换为base64以发送到API
|
||||
const originalImageBase64 = await this.blobToBase64(request.originalImage);
|
||||
|
||||
const contents = [
|
||||
{ text: this.buildEditPrompt(request) },
|
||||
{
|
||||
inlineData: {
|
||||
mimeType: 'image/png',
|
||||
data: originalImageBase64,
|
||||
},
|
||||
},
|
||||
]
|
||||
|
||||
// 如果提供了参考图像则添加
|
||||
if (request.referenceImages && request.referenceImages.length > 0) {
|
||||
// 将Blob转换为base64以发送到API
|
||||
const base64ReferenceImages = await Promise.all(
|
||||
request.referenceImages.map(blob => this.blobToBase64(blob))
|
||||
);
|
||||
|
||||
base64ReferenceImages.forEach(image => {
|
||||
contents.push({
|
||||
inlineData: {
|
||||
mimeType: 'image/png',
|
||||
data: image,
|
||||
},
|
||||
})
|
||||
})
|
||||
}
|
||||
|
||||
if (request.maskImage) {
|
||||
// 将Blob转换为base64以发送到API
|
||||
const maskImageBase64 = await this.blobToBase64(request.maskImage);
|
||||
contents.push({
|
||||
inlineData: {
|
||||
mimeType: 'image/png',
|
||||
data: maskImageBase64,
|
||||
},
|
||||
})
|
||||
}
|
||||
|
||||
const response = await genAI.models.generateContent({
|
||||
model: 'gemini-2.5-flash-image-preview',
|
||||
contents,
|
||||
})
|
||||
|
||||
// 检查是否有被禁止的内容
|
||||
if (response.candidates && response.candidates.length > 0) {
|
||||
const candidate = response.candidates[0]
|
||||
if (candidate.finishReason === 'PROHIBITED_CONTENT') {
|
||||
throw new Error('内容被禁止：您的请求包含不允许的内容。请尝试其他提示。')
|
||||
}
|
||||
// 检查finishReason为STOP但没有inlineData的情况
|
||||
if (candidate.finishReason === 'STOP') {
|
||||
// 检查是否有inlineData
|
||||
let hasInlineData = false;
|
||||
if (candidate.content && candidate.content.parts) {
|
||||
for (const part of candidate.content.parts) {
|
||||
if (part.inlineData) {
|
||||
hasInlineData = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 如果没有inlineData，则抛出错误
|
||||
if (!hasInlineData && candidate.content && candidate.content.parts && candidate.content.parts.length > 0) {
|
||||
throw new Error(candidate.content.parts[0].text || '编辑失败：未返回图像数据');
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
const images: Blob[] = []
|
||||
|
||||
// 检查响应是否存在以及是否有内容
|
||||
if (response.candidates && response.candidates.length > 0 && response.candidates[0].content && response.candidates[0].content.parts) {
|
||||
for (const part of response.candidates[0].content.parts) {
|
||||
if (part.inlineData) {
|
||||
// 将返回的base64数据转换为Blob
|
||||
const byteString = atob(part.inlineData.data);
|
||||
const mimeString = part.inlineData.mimeType || 'image/png';
|
||||
const ab = new ArrayBuffer(byteString.length);
|
||||
const ia = new Uint8Array(ab);
|
||||
for (let i = 0; i < byteString.length; i++) {
|
||||
ia[i] = byteString.charCodeAt(i);
|
||||
}
|
||||
const blob = new Blob([ab], { type: mimeString });
|
||||
images.push(blob);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 获取usageMetadata（如果存在）
|
||||
const usageMetadata = response.usageMetadata
|
||||
|
||||
return { images, usageMetadata }
|
||||
} catch (error) {
|
||||
console.error('编辑图像时出错:', error)
|
||||
if (error instanceof Error && error.message) {
|
||||
throw error
|
||||
}
|
||||
throw new Error(`编辑图像失败: ${error instanceof Error ? error.message : '未知错误'}`)
|
||||
}
|
||||
}
|
||||
|
||||
async segmentImage(request: SegmentationRequest): Promise<any> {
|
||||
try {
|
||||
// 将Blob转换为base64以发送到API
|
||||
const imageBase64 = await this.blobToBase64(request.image);
|
||||
|
||||
const prompt = [
|
||||
{
|
||||
text: `分析此图像并为以下对象创建分割遮罩: ${request.query}
|
||||
|
||||
返回具有此确切结构的JSON对象:
|
||||
{
|
||||
"masks": [
|
||||
{
|
||||
"label": "分割对象的描述",
|
||||
"box_2d": [x, y, width, height],
|
||||
"mask": "base64编码的二进制遮罩图像"
|
||||
}
|
||||
]
|
||||
}
|
||||
|
||||
仅分割请求的特定对象或区域。遮罩应该是二进制PNG，其中白色像素(255)表示选定区域，黑色像素(0)表示背景。`,
|
||||
},
|
||||
{
|
||||
inlineData: {
|
||||
mimeType: 'image/png',
|
||||
data: imageBase64,
|
||||
},
|
||||
},
|
||||
]
|
||||
|
||||
const response = await genAI.models.generateContent({
|
||||
model: 'gemini-2.5-flash-image-preview',
|
||||
contents: prompt,
|
||||
})
|
||||
|
||||
// 检查是否有被禁止的内容
|
||||
if (response.candidates && response.candidates.length > 0) {
|
||||
const candidate = response.candidates[0]
|
||||
if (candidate.finishReason === 'PROHIBITED_CONTENT') {
|
||||
throw new Error('内容被禁止：您的请求包含不允许的内容。请尝试其他提示。')
|
||||
}
|
||||
// 检查finishReason为STOP但没有inlineData的情况
|
||||
if (candidate.finishReason === 'STOP') {
|
||||
// 检查是否有inlineData
|
||||
let hasInlineData = false;
|
||||
if (candidate.content && candidate.content.parts) {
|
||||
for (const part of candidate.content.parts) {
|
||||
if (part.inlineData) {
|
||||
hasInlineData = true;
|
||||
break;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
// 如果没有inlineData，则抛出错误
|
||||
if (!hasInlineData && candidate.content && candidate.content.parts && candidate.content.parts.length > 0) {
|
||||
throw new Error(candidate.content.parts[0].text || '分割失败：未返回结果数据');
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
const responseText = response.candidates[0].content.parts[0].text
|
||||
return JSON.parse(responseText)
|
||||
} catch (error) {
|
||||
console.error('分割图像时出错:', error)
|
||||
if (error instanceof Error && error.message) {
|
||||
throw error
|
||||
}
|
||||
throw new Error(`分割图像失败: ${error instanceof Error ? error.message : '未知错误'}`)
|
||||
}
|
||||
}
|
||||
|
||||
private buildEditPrompt(request: EditRequest): string {
|
||||
const maskInstruction = request.maskImage ? '\n\n重要: 仅在遮罩图像显示白色像素(值255)的地方应用更改。完全不更改所有其他区域。精确遵守遮罩边界并在边缘保持无缝混合。' : ''
|
||||
|
||||
return `根据以下指令编辑此图像: ${request.instruction}
|
||||
|
||||
保持原始图像的光照、透视和整体构图。使更改看起来自然且无缝集成。${maskInstruction}
|
||||
|
||||
保持图像质量并确保编辑看起来专业且逼真。`
|
||||
}
|
||||
}
|
||||
|
||||
export const geminiService = new GeminiService()
|
||||
Reference in New Issue
Block a user