mirror of
https://github.com/basketikun/infinite-canvas.git
synced 2026-07-24 06:54:06 +08:00
feat(api): add support for Gemini API format and enhance channel configuration
This commit is contained in:
@@ -2,6 +2,8 @@
|
||||
|
||||
## Unreleased
|
||||
|
||||
+ [新增] 渠道兼容Gemini格式。
|
||||
|
||||
## v0.4.0 - 2026-06-16
|
||||
|
||||
+ [新增] 新增网页版Agent Loop模式。
|
||||
|
||||
@@ -5,3 +5,6 @@ description: 当前版本已实现但仍需人工验证的变更项
|
||||
|
||||
# 待测试
|
||||
|
||||
- 配置与用户偏好:每个模型渠道新增 OpenAI / Gemini 调用格式选择,默认 OpenAI,切换默认格式时同步更新默认 Base URL;渠道 Tab 增加紧凑醒目的提醒,说明模型是否可选需要到“模型”Tab 选择。
|
||||
- Gemini 调用格式:支持拉取 Gemini 模型列表,并用于文本问答、在线 Agent 工具调用、基础生图和图生图请求。
|
||||
- 音频和视频生成:选择 Gemini 调用格式时先给出不支持提示,仍需使用 OpenAI 兼容渠道。
|
||||
|
||||
@@ -1056,7 +1056,7 @@ function isResponseToolCall(value: unknown): value is ResponseToolCall {
|
||||
}
|
||||
|
||||
function toolCallToResponseInput(call: ResponseToolCall): ResponseInputMessage {
|
||||
return { type: "function_call", call_id: call.id, name: call.function.name, arguments: call.function.arguments };
|
||||
return { type: "function_call", call_id: call.id, name: call.function.name, arguments: call.function.arguments, ...(call.thoughtSignature ? { thoughtSignature: call.thoughtSignature } : {}) };
|
||||
}
|
||||
|
||||
function summarizeToolCalls(calls: ResponseToolCall[]) {
|
||||
|
||||
@@ -1,7 +1,7 @@
|
||||
"use client";
|
||||
|
||||
import { App, Button, Form, Input, Modal, Progress, Segmented, Select, Tabs } from "antd";
|
||||
import { Cloud, Plus, RefreshCw, Trash2, Wifi } from "lucide-react";
|
||||
import { CircleAlert, Cloud, Plus, RefreshCw, Trash2, Wifi } from "lucide-react";
|
||||
import { useState } from "react";
|
||||
|
||||
import { ModelPicker } from "@/components/model-picker";
|
||||
@@ -9,7 +9,7 @@ import { fetchChannelModels } from "@/services/api/image";
|
||||
import { syncAppDataToWebdav, type AppSyncDomainKey, type AppSyncProgressEvent } from "@/services/app-sync";
|
||||
import { testWebdavConnection, WEBDAV_MANIFEST_FILE_NAME } from "@/services/webdav-sync";
|
||||
import { audioFormatOptions, audioVoiceOptions, normalizeAudioSpeedValue } from "@/lib/audio-generation";
|
||||
import { createModelChannel, filterModelsByCapability, modelOptionLabel, modelOptionsFromChannels, normalizeModelOptionValue, useConfigStore, type AiConfig, type ModelCapability, type ModelChannel } from "@/stores/use-config-store";
|
||||
import { createModelChannel, defaultBaseUrlForApiFormat, filterModelsByCapability, modelOptionLabel, modelOptionsFromChannels, normalizeModelOptionValue, useConfigStore, type AiConfig, type ApiCallFormat, type ModelCapability, type ModelChannel } from "@/stores/use-config-store";
|
||||
|
||||
type ModelGroup = {
|
||||
capability: ModelCapability;
|
||||
@@ -34,6 +34,11 @@ const modelGroups: ModelGroup[] = [
|
||||
{ capability: "audio", modelKey: "audioModel", modelsKey: "audioModels", defaultLabel: "默认音频模型", optionsLabel: "音频模型可选项" },
|
||||
];
|
||||
|
||||
const apiFormatOptions: Array<{ label: string; value: ApiCallFormat }> = [
|
||||
{ label: "OpenAI", value: "openai" },
|
||||
{ label: "Gemini", value: "gemini" },
|
||||
];
|
||||
|
||||
const webdavDomainKeys: AppSyncDomainKey[] = ["canvas", "assets", "image-workbench", "video-workbench"];
|
||||
const webdavDomainLabels: Record<AppSyncDomainKey, string> = {
|
||||
canvas: "画布",
|
||||
@@ -92,6 +97,11 @@ export function AppConfigModal() {
|
||||
updateChannels(config.channels.map((channel) => (channel.id === id ? { ...channel, ...patch, models: patch.models ? uniqueModels(patch.models) : channel.models } : channel)));
|
||||
};
|
||||
|
||||
const updateChannelApiFormat = (channel: ModelChannel, apiFormat: ApiCallFormat) => {
|
||||
const baseUrl = !channel.baseUrl.trim() || channel.baseUrl.trim() === defaultBaseUrlForApiFormat(channel.apiFormat) ? defaultBaseUrlForApiFormat(apiFormat) : channel.baseUrl;
|
||||
updateChannel(channel.id, { apiFormat, baseUrl });
|
||||
};
|
||||
|
||||
const addChannel = () => {
|
||||
updateChannels([...config.channels, createModelChannel({ name: `渠道 ${config.channels.length + 1}` })]);
|
||||
};
|
||||
@@ -226,11 +236,17 @@ export function AppConfigModal() {
|
||||
children: (
|
||||
<Form layout="vertical" requiredMark={false}>
|
||||
<div className="mb-4 flex flex-wrap items-center justify-between gap-3 rounded-lg border border-stone-200 p-3 dark:border-stone-800">
|
||||
<div>
|
||||
<div className="text-sm font-semibold">多渠道聚合</div>
|
||||
<div className="mt-1 text-xs text-stone-500">每个渠道保存自己的 Base URL、API Key 和模型列表;模型选择时会显示模型名和渠道名。</div>
|
||||
<div className="min-w-0 flex-1">
|
||||
<div className="flex w-fit max-w-full flex-wrap items-center gap-1.5 rounded-md border border-amber-300 bg-amber-50 px-2.5 py-1.5 text-xs text-amber-900 dark:border-amber-700/60 dark:bg-amber-950/30 dark:text-amber-100">
|
||||
<CircleAlert className="size-3.5 shrink-0" />
|
||||
<span className="font-semibold">重要:</span>
|
||||
<span>新增或拉取模型后,需要到“模型”Tab 选择可选项才会显示。</span>
|
||||
<Button type="link" size="small" className="h-auto p-0 text-xs font-semibold text-amber-900 dark:text-amber-100" onClick={() => setActiveTab("models")}>
|
||||
去模型设置
|
||||
</Button>
|
||||
</div>
|
||||
</div>
|
||||
<div className="flex gap-2">
|
||||
<div className="flex shrink-0 gap-2">
|
||||
<Button icon={<RefreshCw className="size-4" />} loading={Boolean(loadingChannelId)} onClick={() => void refreshAllModels()}>
|
||||
拉取全部
|
||||
</Button>
|
||||
@@ -245,7 +261,9 @@ export function AppConfigModal() {
|
||||
<div className="mb-3 flex items-center justify-between gap-3">
|
||||
<div className="min-w-0">
|
||||
<div className="truncate text-sm font-semibold">{channel.name || "未命名渠道"}</div>
|
||||
<div className="mt-1 text-xs text-stone-500">已保存 {channel.models.length} 个模型</div>
|
||||
<div className="mt-1 text-xs text-stone-500">
|
||||
{apiFormatLabel(channel.apiFormat)} · 已保存 {channel.models.length} 个模型
|
||||
</div>
|
||||
</div>
|
||||
<div className="flex shrink-0 gap-2">
|
||||
<Button size="small" loading={loadingChannelId === channel.id} onClick={() => void refreshChannelModels(channel)}>
|
||||
@@ -258,13 +276,16 @@ export function AppConfigModal() {
|
||||
<Form.Item label="渠道名称" className="mb-0">
|
||||
<Input value={channel.name} onChange={(event) => updateChannel(channel.id, { name: event.target.value })} />
|
||||
</Form.Item>
|
||||
<Form.Item label="调用格式" className="mb-0">
|
||||
<Select value={channel.apiFormat} options={apiFormatOptions} onChange={(value: ApiCallFormat) => updateChannelApiFormat(channel, value)} />
|
||||
</Form.Item>
|
||||
<Form.Item label="Base URL" className="mb-0">
|
||||
<Input value={channel.baseUrl} onChange={(event) => updateChannel(channel.id, { baseUrl: event.target.value })} />
|
||||
</Form.Item>
|
||||
<Form.Item label="API Key" className="mb-0">
|
||||
<Input.Password value={channel.apiKey} onChange={(event) => updateChannel(channel.id, { apiKey: event.target.value })} />
|
||||
</Form.Item>
|
||||
<Form.Item label="模型列表" className="mb-0">
|
||||
<Form.Item label="模型列表" className="mb-0 md:col-span-2">
|
||||
<Select mode="tags" showSearch allowClear maxTagCount="responsive" placeholder="输入模型名,或点击拉取模型" value={channel.models} onChange={(models) => updateChannel(channel.id, { models })} />
|
||||
</Form.Item>
|
||||
</div>
|
||||
@@ -425,6 +446,7 @@ function withChannels(config: AiConfig, channels: ModelChannel[]): AiConfig {
|
||||
models,
|
||||
baseUrl: channels[0]?.baseUrl || config.baseUrl,
|
||||
apiKey: channels[0]?.apiKey || config.apiKey,
|
||||
apiFormat: channels[0]?.apiFormat || config.apiFormat,
|
||||
imageModels,
|
||||
videoModels,
|
||||
textModels,
|
||||
@@ -455,6 +477,10 @@ function uniqueModels(models: string[]) {
|
||||
return Array.from(new Set(models.map((model) => model.trim()).filter(Boolean)));
|
||||
}
|
||||
|
||||
function apiFormatLabel(apiFormat: ApiCallFormat) {
|
||||
return apiFormat === "gemini" ? "Gemini" : "OpenAI";
|
||||
}
|
||||
|
||||
function formatWebdavTime(value: string) {
|
||||
return new Date(value).toLocaleString("zh-CN", { month: "2-digit", day: "2-digit", hour: "2-digit", minute: "2-digit" });
|
||||
}
|
||||
|
||||
@@ -53,6 +53,7 @@ function assertAudioConfig(config: AiConfig, model: string) {
|
||||
if (!model) throw new Error("请先配置音频模型");
|
||||
if (!config.baseUrl.trim()) throw new Error("请先配置 Base URL");
|
||||
if (!config.apiKey.trim()) throw new Error("请先配置 API Key");
|
||||
if (config.apiFormat === "gemini") throw new Error("Gemini 调用格式暂不支持音频生成,请使用 OpenAI 格式渠道");
|
||||
}
|
||||
|
||||
async function assertAudioBlob(blob: Blob) {
|
||||
|
||||
@@ -16,11 +16,12 @@ export type ResponseToolCall = {
|
||||
id: string;
|
||||
type: "function";
|
||||
function: { name: string; arguments: string };
|
||||
thoughtSignature?: string;
|
||||
};
|
||||
|
||||
export type ResponseInputMessage =
|
||||
| AiTextMessage
|
||||
| { type: "function_call"; call_id: string; name: string; arguments: string }
|
||||
| { type: "function_call"; call_id: string; name: string; arguments: string; thoughtSignature?: string }
|
||||
| { role: "tool"; tool_call_id: string; content: string };
|
||||
|
||||
export type ResponseFunctionTool = {
|
||||
@@ -71,6 +72,24 @@ type ImageApiResponse = {
|
||||
code?: number;
|
||||
msg?: string;
|
||||
};
|
||||
type GeminiPart = {
|
||||
text?: string;
|
||||
inlineData?: { mimeType?: string; data?: string };
|
||||
inline_data?: { mime_type?: string; mimeType?: string; data?: string };
|
||||
fileData?: { mimeType?: string; fileUri?: string };
|
||||
functionCall?: { id?: string; name?: string; args?: Record<string, unknown> };
|
||||
functionResponse?: { id?: string; name?: string; response?: Record<string, unknown> };
|
||||
thoughtSignature?: string;
|
||||
thought_signature?: string;
|
||||
};
|
||||
type GeminiContent = { role?: "user" | "model"; parts: GeminiPart[] };
|
||||
type GeminiPayload = {
|
||||
candidates?: Array<{ content?: { parts?: GeminiPart[] }; finishReason?: string }>;
|
||||
models?: Array<{ name?: string }>;
|
||||
error?: { message?: string };
|
||||
promptFeedback?: { blockReason?: string };
|
||||
};
|
||||
type GeminiStreamState = { buffer: string; text: string; toolCalls: ResponseToolCall[]; error?: string };
|
||||
type RequestOptions = { signal?: AbortSignal };
|
||||
|
||||
const QUALITY_BASE: Record<string, number> = {
|
||||
@@ -220,6 +239,29 @@ function aiHeaders(config: AiConfig, contentType?: string) {
|
||||
};
|
||||
}
|
||||
|
||||
function geminiBaseUrl(config: Pick<AiConfig, "baseUrl">) {
|
||||
const normalizedBaseUrl = config.baseUrl.trim().replace(/\/+$/, "");
|
||||
const lowerBaseUrl = normalizedBaseUrl.toLowerCase();
|
||||
return lowerBaseUrl.endsWith("/v1") || lowerBaseUrl.endsWith("/v1beta") ? normalizedBaseUrl : `${normalizedBaseUrl}/v1beta`;
|
||||
}
|
||||
|
||||
function geminiModelName(model: string) {
|
||||
return model.trim().replace(/^models\//, "");
|
||||
}
|
||||
|
||||
function geminiApiUrl(config: Pick<AiConfig, "baseUrl" | "model">, action?: "generateContent" | "streamGenerateContent") {
|
||||
const baseUrl = geminiBaseUrl(config);
|
||||
if (!action) return `${baseUrl}/models`;
|
||||
return `${baseUrl}/models/${encodeURIComponent(geminiModelName(config.model))}:${action}`;
|
||||
}
|
||||
|
||||
function geminiHeaders(config: Pick<AiConfig, "apiKey">) {
|
||||
return {
|
||||
"x-goog-api-key": config.apiKey,
|
||||
"Content-Type": "application/json",
|
||||
};
|
||||
}
|
||||
|
||||
function withSystemMessage<T extends ResponseInputMessage>(config: AiConfig, messages: T[]): ResponseInputMessage[] {
|
||||
const systemPrompt = config.systemPrompt.trim();
|
||||
return systemPrompt ? [{ role: "system" as const, content: systemPrompt }, ...messages] : messages;
|
||||
@@ -288,6 +330,11 @@ function validateResponsePayload(payload: ResponseApiPayload) {
|
||||
if (payload.error?.message) throw new Error(payload.error.message);
|
||||
}
|
||||
|
||||
function validateGeminiPayload(payload: GeminiPayload) {
|
||||
if (payload.error?.message) throw new Error(payload.error.message);
|
||||
if (payload.promptFeedback?.blockReason) throw new Error(`Gemini 拒绝了本次请求:${payload.promptFeedback.blockReason}`);
|
||||
}
|
||||
|
||||
async function readFetchError(response: Response, fallback: string) {
|
||||
const text = await response.text();
|
||||
if (!text) return readStatusError(response.status, fallback);
|
||||
@@ -370,9 +417,206 @@ async function requestStreamingResponse(config: AiConfig, body: Record<string, u
|
||||
return { ...result, content: state.text || result.content };
|
||||
}
|
||||
|
||||
function toGeminiBody(config: AiConfig, messages: ResponseInputMessage[], extra?: Record<string, unknown>) {
|
||||
const systemText = [
|
||||
config.systemPrompt.trim(),
|
||||
...messages.flatMap((message) => (!("type" in message) && message.role === "system" ? [geminiTextContent(message.content)] : [])),
|
||||
]
|
||||
.filter(Boolean)
|
||||
.join("\n\n");
|
||||
const contents = toGeminiContents(messages.filter((message) => ("type" in message ? true : message.role !== "system")));
|
||||
return {
|
||||
contents,
|
||||
...(systemText ? { systemInstruction: { parts: [{ text: systemText }] } } : {}),
|
||||
...extra,
|
||||
};
|
||||
}
|
||||
|
||||
function toGeminiContents(messages: ResponseInputMessage[]): GeminiContent[] {
|
||||
const callNameById = new Map<string, string>();
|
||||
return messages.flatMap((message): GeminiContent[] => {
|
||||
if ("type" in message) {
|
||||
callNameById.set(message.call_id, message.name);
|
||||
return [{ role: "model", parts: [{ functionCall: { id: message.call_id, name: message.name, args: jsonObject(message.arguments) }, ...(message.thoughtSignature ? { thoughtSignature: message.thoughtSignature } : {}) }] }];
|
||||
}
|
||||
if (message.role === "tool") {
|
||||
const name = callNameById.get(message.tool_call_id) || "tool_result";
|
||||
return [{ role: "user", parts: [{ functionResponse: { id: message.tool_call_id, name, response: { result: jsonValue(message.content) } } }] }];
|
||||
}
|
||||
return [{ role: message.role === "assistant" ? "model" : "user", parts: toGeminiParts(message.content) }];
|
||||
});
|
||||
}
|
||||
|
||||
function toGeminiParts(content: ResponseMessageContent): GeminiPart[] {
|
||||
if (!Array.isArray(content)) return [{ text: String(content || "") }];
|
||||
return content.map((item) => (item.type === "text" ? { text: item.text } : toGeminiImagePart(item.image_url.url)));
|
||||
}
|
||||
|
||||
function toGeminiImagePart(url: string): GeminiPart {
|
||||
const match = url.match(/^data:([^;,]+);base64,(.+)$/);
|
||||
if (match) return { inlineData: { mimeType: match[1], data: match[2] } };
|
||||
return { fileData: { fileUri: url, mimeType: "image/png" } };
|
||||
}
|
||||
|
||||
function geminiTextContent(content: ResponseMessageContent) {
|
||||
if (!Array.isArray(content)) return String(content || "");
|
||||
return content.map((item) => (item.type === "text" ? item.text : item.image_url.url)).join("\n");
|
||||
}
|
||||
|
||||
function jsonObject(value: string): Record<string, unknown> {
|
||||
const parsed = jsonValue(value);
|
||||
return isRecord(parsed) ? parsed : {};
|
||||
}
|
||||
|
||||
function jsonValue(value: string): unknown {
|
||||
try {
|
||||
return JSON.parse(value);
|
||||
} catch {
|
||||
return value;
|
||||
}
|
||||
}
|
||||
|
||||
function toGeminiToolOptions(tools: ResponseFunctionTool[], toolChoice: ToolChoice) {
|
||||
if (!tools.length) return {};
|
||||
const functionDeclarations = tools.map((tool) => ({
|
||||
name: tool.function.name,
|
||||
description: tool.function.description,
|
||||
parameters: tool.function.parameters,
|
||||
}));
|
||||
const functionCallingConfig =
|
||||
typeof toolChoice === "object"
|
||||
? { mode: "ANY", allowedFunctionNames: [toolChoice.name] }
|
||||
: { mode: toolChoice === "required" ? "ANY" : "AUTO" };
|
||||
return {
|
||||
tools: [{ functionDeclarations }],
|
||||
toolConfig: { functionCallingConfig },
|
||||
};
|
||||
}
|
||||
|
||||
async function requestGeminiStreamingResponse(config: AiConfig, body: Record<string, unknown>, onDelta?: (text: string) => void, options?: RequestOptions): Promise<ToolResponseResult> {
|
||||
const response = await fetch(`${geminiApiUrl(config, "streamGenerateContent")}?alt=sse`, {
|
||||
method: "POST",
|
||||
headers: geminiHeaders(config),
|
||||
body: JSON.stringify(body),
|
||||
signal: options?.signal,
|
||||
});
|
||||
if (!response.ok) throw new Error(await readFetchError(response, "请求失败"));
|
||||
if (!response.body) {
|
||||
const payload = (await response.json()) as GeminiPayload;
|
||||
return parseGeminiToolResponse(payload);
|
||||
}
|
||||
|
||||
const reader = response.body.getReader();
|
||||
const decoder = new TextDecoder();
|
||||
const state: GeminiStreamState = { buffer: "", text: "", toolCalls: [] };
|
||||
for (;;) {
|
||||
const { done, value } = await reader.read();
|
||||
if (done) break;
|
||||
consumeGeminiStreamText(state, decoder.decode(value, { stream: true }), onDelta);
|
||||
if (state.error) throw new Error(state.error);
|
||||
}
|
||||
consumeGeminiStreamText(state, decoder.decode(), onDelta, true);
|
||||
if (state.error) throw new Error(state.error);
|
||||
return { content: state.text, toolCalls: state.toolCalls };
|
||||
}
|
||||
|
||||
function consumeGeminiStreamText(state: GeminiStreamState, text: string, onDelta?: (text: string) => void, flush = false) {
|
||||
state.buffer += text;
|
||||
for (;;) {
|
||||
const match = state.buffer.match(/\r?\n\r?\n/);
|
||||
if (!match) break;
|
||||
consumeGeminiStreamBlock(state.buffer.slice(0, match.index), state, onDelta);
|
||||
state.buffer = state.buffer.slice(match.index + match[0].length);
|
||||
}
|
||||
if (flush && state.buffer.trim()) {
|
||||
consumeGeminiStreamBlock(state.buffer, state, onDelta);
|
||||
state.buffer = "";
|
||||
}
|
||||
}
|
||||
|
||||
function consumeGeminiStreamBlock(block: string, state: GeminiStreamState, onDelta?: (text: string) => void) {
|
||||
const data = block
|
||||
.split(/\r?\n/)
|
||||
.filter((line) => line.startsWith("data:"))
|
||||
.map((line) => line.slice(5).replace(/^ /, ""))
|
||||
.join("\n")
|
||||
.trim();
|
||||
if (!data || data === "[DONE]") return;
|
||||
const result = parseGeminiToolResponse(JSON.parse(data) as GeminiPayload);
|
||||
if (result.content) {
|
||||
state.text += result.content;
|
||||
onDelta?.(state.text);
|
||||
}
|
||||
state.toolCalls.push(...result.toolCalls);
|
||||
}
|
||||
|
||||
function parseGeminiToolResponse(payload: GeminiPayload): ToolResponseResult {
|
||||
validateGeminiPayload(payload);
|
||||
const parts = payload.candidates?.flatMap((candidate) => candidate.content?.parts || []) || [];
|
||||
const content = parts.map((part) => part.text || "").join("");
|
||||
const toolCalls = parts
|
||||
.map((part) => part.functionCall)
|
||||
.filter((call): call is NonNullable<GeminiPart["functionCall"]> => Boolean(call?.name))
|
||||
.map((call) => {
|
||||
const part = parts.find((item) => item.functionCall === call);
|
||||
const thoughtSignature = part?.thoughtSignature || part?.thought_signature;
|
||||
return {
|
||||
id: call.id || nanoid(),
|
||||
type: "function" as const,
|
||||
function: { name: call.name || "", arguments: JSON.stringify(call.args || {}) },
|
||||
...(thoughtSignature ? { thoughtSignature } : {}),
|
||||
};
|
||||
});
|
||||
return { content, toolCalls };
|
||||
}
|
||||
|
||||
async function requestGeminiImages(config: AiConfig, prompt: string, references: ReferenceImage[], count: number, options?: RequestOptions) {
|
||||
const requests = Array.from({ length: count }, () => requestGeminiImagesOnce(config, prompt, references, options));
|
||||
return (await Promise.all(requests)).flat();
|
||||
}
|
||||
|
||||
async function requestGeminiImagesOnce(config: AiConfig, prompt: string, references: ReferenceImage[], options?: RequestOptions) {
|
||||
const parts: GeminiPart[] = [{ text: prompt }];
|
||||
for (const image of references) {
|
||||
parts.push(toGeminiImagePart(await imageToDataUrl(image)));
|
||||
}
|
||||
const response = await axios.post<GeminiPayload>(
|
||||
geminiApiUrl(config, "generateContent"),
|
||||
{
|
||||
...toGeminiBody(config, [{ role: "user", content: prompt }], { generationConfig: { responseModalities: ["TEXT", "IMAGE"] } }),
|
||||
contents: [{ role: "user", parts }],
|
||||
},
|
||||
{ headers: geminiHeaders(config), signal: options?.signal },
|
||||
);
|
||||
return parseGeminiImagePayload(response.data);
|
||||
}
|
||||
|
||||
function parseGeminiImagePayload(payload: GeminiPayload) {
|
||||
validateGeminiPayload(payload);
|
||||
const images =
|
||||
payload.candidates
|
||||
?.flatMap((candidate) => candidate.content?.parts || [])
|
||||
.map((part) => {
|
||||
const inlineData = part.inlineData || (part.inline_data ? { mimeType: part.inline_data.mimeType || part.inline_data.mime_type, data: part.inline_data.data } : undefined);
|
||||
if (inlineData?.data) return `data:${inlineData.mimeType || "image/png"};base64,${inlineData.data}`;
|
||||
return part.fileData?.fileUri || null;
|
||||
})
|
||||
.filter((value): value is string => Boolean(value))
|
||||
.map((dataUrl) => ({ id: nanoid(), dataUrl })) || [];
|
||||
if (!images.length) throw new Error("Gemini 接口没有返回图片");
|
||||
return images;
|
||||
}
|
||||
|
||||
export async function requestGeneration(config: AiConfig, prompt: string, options?: RequestOptions) {
|
||||
const requestConfig = resolveModelRequestConfig(config, config.model || config.imageModel);
|
||||
const n = Math.max(1, Math.min(15, Math.floor(Math.abs(Number(config.count)) || 1)));
|
||||
if (requestConfig.apiFormat === "gemini") {
|
||||
try {
|
||||
return await requestGeminiImages(requestConfig, prompt, [], n, options);
|
||||
} catch (error) {
|
||||
throw new Error(readAxiosError(error, "请求失败"));
|
||||
}
|
||||
}
|
||||
const quality = normalizeQuality(config.quality);
|
||||
const requestSize = resolveRequestSize(quality, config.size);
|
||||
try {
|
||||
@@ -402,9 +646,17 @@ export async function requestGeneration(config: AiConfig, prompt: string, option
|
||||
export async function requestEdit(config: AiConfig, prompt: string, references: ReferenceImage[], mask?: ReferenceImage, options?: RequestOptions) {
|
||||
const requestConfig = resolveModelRequestConfig(config, config.model || config.imageModel);
|
||||
const n = Math.max(1, Math.min(15, Math.floor(Math.abs(Number(config.count)) || 1)));
|
||||
const requestPrompt = buildImageReferencePromptText(prompt, references);
|
||||
if (requestConfig.apiFormat === "gemini") {
|
||||
if (mask) throw new Error("Gemini 调用格式暂不支持蒙版编辑");
|
||||
try {
|
||||
return await requestGeminiImages(requestConfig, requestPrompt, references, n, options);
|
||||
} catch (error) {
|
||||
throw new Error(readAxiosError(error, "请求失败"));
|
||||
}
|
||||
}
|
||||
const quality = normalizeQuality(config.quality);
|
||||
const requestSize = resolveRequestSize(quality, config.size);
|
||||
const requestPrompt = buildImageReferencePromptText(prompt, references);
|
||||
const formData = new FormData();
|
||||
formData.set("model", requestConfig.model);
|
||||
formData.set("prompt", withSystemPrompt(requestConfig, requestPrompt));
|
||||
@@ -433,6 +685,11 @@ export async function requestEdit(config: AiConfig, prompt: string, references:
|
||||
export async function requestImageQuestion(config: AiConfig, messages: AiTextMessage[], onDelta: (text: string) => void, options?: RequestOptions) {
|
||||
const requestConfig = resolveModelRequestConfig(config, config.model || config.textModel);
|
||||
try {
|
||||
if (requestConfig.apiFormat === "gemini") {
|
||||
const answer = (await requestGeminiStreamingResponse(requestConfig, toGeminiBody(requestConfig, messages), onDelta, options)).content || "没有返回内容";
|
||||
if (answer === "没有返回内容") onDelta(answer);
|
||||
return answer;
|
||||
}
|
||||
const answer = (await requestStreamingResponse(requestConfig, {
|
||||
model: requestConfig.model,
|
||||
input: toResponseInput(withSystemMessage(requestConfig, messages)),
|
||||
@@ -447,6 +704,9 @@ export async function requestImageQuestion(config: AiConfig, messages: AiTextMes
|
||||
export async function requestToolResponse(config: AiConfig, messages: ResponseInputMessage[], tools: ResponseFunctionTool[], toolChoice: ToolChoice = "auto", onDelta?: (text: string) => void, options?: RequestOptions): Promise<ToolResponseResult> {
|
||||
const requestConfig = resolveModelRequestConfig(config, config.model || config.textModel);
|
||||
try {
|
||||
if (requestConfig.apiFormat === "gemini") {
|
||||
return await requestGeminiStreamingResponse(requestConfig, toGeminiBody(requestConfig, messages, toGeminiToolOptions(tools, toolChoice)), onDelta, options);
|
||||
}
|
||||
return await requestStreamingResponse(requestConfig, {
|
||||
model: requestConfig.model,
|
||||
input: toResponseInput(withSystemMessage(requestConfig, messages)),
|
||||
@@ -459,8 +719,16 @@ export async function requestToolResponse(config: AiConfig, messages: ResponseIn
|
||||
}
|
||||
}
|
||||
|
||||
export async function fetchImageModels(config: Pick<AiConfig, "baseUrl" | "apiKey">) {
|
||||
export async function fetchImageModels(config: Pick<AiConfig, "baseUrl" | "apiKey" | "apiFormat">) {
|
||||
try {
|
||||
if (config.apiFormat === "gemini") {
|
||||
const response = await axios.get<GeminiPayload>(geminiApiUrl({ ...defaultGeminiConfig, ...config }), { headers: geminiHeaders({ ...defaultGeminiConfig, ...config }) });
|
||||
validateGeminiPayload(response.data);
|
||||
return (response.data.models || [])
|
||||
.map((model) => model.name?.replace(/^models\//, ""))
|
||||
.filter((id): id is string => Boolean(id))
|
||||
.sort((a, b) => a.localeCompare(b));
|
||||
}
|
||||
const response = await axios.get<{ data?: Array<{ id?: string }>; error?: { message?: string } }>(buildApiUrl(config.baseUrl, "/models"), {
|
||||
headers: {
|
||||
Authorization: `Bearer ${config.apiKey}`,
|
||||
@@ -476,5 +744,13 @@ export async function fetchImageModels(config: Pick<AiConfig, "baseUrl" | "apiKe
|
||||
}
|
||||
|
||||
export async function fetchChannelModels(channel: ModelChannel) {
|
||||
return fetchImageModels({ baseUrl: channel.baseUrl, apiKey: channel.apiKey });
|
||||
return fetchImageModels({ baseUrl: channel.baseUrl, apiKey: channel.apiKey, apiFormat: channel.apiFormat });
|
||||
}
|
||||
|
||||
const defaultGeminiConfig: Pick<AiConfig, "baseUrl" | "apiKey" | "apiFormat" | "model" | "systemPrompt"> = {
|
||||
baseUrl: "https://generativelanguage.googleapis.com",
|
||||
apiKey: "",
|
||||
apiFormat: "gemini",
|
||||
model: "",
|
||||
systemPrompt: "",
|
||||
};
|
||||
|
||||
@@ -232,6 +232,7 @@ function assertVideoConfig(config: AiConfig, model: string) {
|
||||
if (!model) throw new Error("请先配置视频模型");
|
||||
if (!config.baseUrl.trim()) throw new Error("请先配置 Base URL");
|
||||
if (!config.apiKey.trim()) throw new Error("请先配置 API Key");
|
||||
if (config.apiFormat === "gemini") throw new Error("Gemini 调用格式暂不支持视频生成,请使用 OpenAI 格式渠道");
|
||||
}
|
||||
|
||||
function normalizeVideoSeconds(value: string) {
|
||||
|
||||
@@ -5,11 +5,14 @@ import { create } from "zustand";
|
||||
import { persist } from "zustand/middleware";
|
||||
import { nanoid } from "nanoid";
|
||||
|
||||
export type ApiCallFormat = "openai" | "gemini";
|
||||
|
||||
export type ModelChannel = {
|
||||
id: string;
|
||||
name: string;
|
||||
baseUrl: string;
|
||||
apiKey: string;
|
||||
apiFormat: ApiCallFormat;
|
||||
models: string[];
|
||||
};
|
||||
|
||||
@@ -17,6 +20,7 @@ export type AiConfig = {
|
||||
channelMode: "remote" | "local";
|
||||
baseUrl: string;
|
||||
apiKey: string;
|
||||
apiFormat: ApiCallFormat;
|
||||
channels: ModelChannel[];
|
||||
model: string;
|
||||
imageModel: string;
|
||||
@@ -55,17 +59,21 @@ export type WebdavSyncConfig = {
|
||||
export const CONFIG_STORE_KEY = "infinite-canvas:ai_config_store";
|
||||
export type ModelCapability = "image" | "video" | "text" | "audio";
|
||||
const CHANNEL_MODEL_SEPARATOR = "::";
|
||||
const OPENAI_BASE_URL = "https://api.openai.com";
|
||||
const GEMINI_BASE_URL = "https://generativelanguage.googleapis.com";
|
||||
|
||||
export const defaultConfig: AiConfig = {
|
||||
channelMode: "local",
|
||||
baseUrl: "https://api.openai.com",
|
||||
baseUrl: OPENAI_BASE_URL,
|
||||
apiKey: "",
|
||||
apiFormat: "openai",
|
||||
channels: [
|
||||
{
|
||||
id: "default",
|
||||
name: "默认渠道",
|
||||
baseUrl: "https://api.openai.com",
|
||||
baseUrl: OPENAI_BASE_URL,
|
||||
apiKey: "",
|
||||
apiFormat: "openai",
|
||||
models: ["gpt-image-2", "grok-imagine-video", "gpt-5.5", "gpt-4o-mini-tts"],
|
||||
},
|
||||
],
|
||||
@@ -204,6 +212,7 @@ export const useConfigStore = create<ConfigStore>()(
|
||||
config: {
|
||||
...config,
|
||||
channelMode: "local",
|
||||
apiFormat: normalizeApiFormat(config.apiFormat),
|
||||
channels,
|
||||
models,
|
||||
imageModel: normalizeModelOptionValue(config.imageModel || config.model, channels),
|
||||
@@ -243,11 +252,13 @@ export function useEffectiveConfig() {
|
||||
}
|
||||
|
||||
export function createModelChannel(channel?: Partial<ModelChannel>): ModelChannel {
|
||||
const apiFormat = normalizeApiFormat(channel?.apiFormat);
|
||||
return {
|
||||
id: channel?.id?.trim() || nanoid(),
|
||||
name: channel?.name?.trim() || "新渠道",
|
||||
baseUrl: channel?.baseUrl?.trim() || "https://api.openai.com",
|
||||
baseUrl: channel?.baseUrl?.trim() || defaultBaseUrlForApiFormat(apiFormat),
|
||||
apiKey: channel?.apiKey || "",
|
||||
apiFormat,
|
||||
models: uniqueRawModels(channel?.models || []),
|
||||
};
|
||||
}
|
||||
@@ -297,7 +308,7 @@ export function resolveModelChannel(config: AiConfig, value: string) {
|
||||
const decoded = decodeChannelModel(value);
|
||||
const model = decoded?.model || value;
|
||||
const matched = decoded ? config.channels.find((channel) => channel.id === decoded.channelId) : config.channels.find((channel) => channel.models.includes(model));
|
||||
return matched || config.channels[0] || createModelChannel({ id: "default", name: "默认渠道", baseUrl: config.baseUrl, apiKey: config.apiKey, models: config.models.map(modelOptionName) });
|
||||
return matched || config.channels[0] || createModelChannel({ id: "default", name: "默认渠道", baseUrl: config.baseUrl, apiKey: config.apiKey, apiFormat: config.apiFormat, models: config.models.map(modelOptionName) });
|
||||
}
|
||||
|
||||
export function resolveModelRequestConfig(config: AiConfig, value: string) {
|
||||
@@ -307,6 +318,7 @@ export function resolveModelRequestConfig(config: AiConfig, value: string) {
|
||||
model: modelOptionName(value || config.model),
|
||||
baseUrl: channel.baseUrl,
|
||||
apiKey: channel.apiKey,
|
||||
apiFormat: channel.apiFormat,
|
||||
};
|
||||
}
|
||||
|
||||
@@ -327,6 +339,7 @@ function normalizeChannels(config: AiConfig) {
|
||||
name: "默认渠道",
|
||||
baseUrl: config.baseUrl || defaultConfig.baseUrl,
|
||||
apiKey: config.apiKey || "",
|
||||
apiFormat: config.apiFormat || defaultConfig.apiFormat,
|
||||
models: uniqueRawModels([
|
||||
...(config.models || []),
|
||||
config.model,
|
||||
@@ -341,6 +354,14 @@ function normalizeChannels(config: AiConfig) {
|
||||
return channels.map((channel) => ({ ...channel, models: uniqueRawModels(channel.models) }));
|
||||
}
|
||||
|
||||
export function defaultBaseUrlForApiFormat(apiFormat: ApiCallFormat) {
|
||||
return apiFormat === "gemini" ? GEMINI_BASE_URL : OPENAI_BASE_URL;
|
||||
}
|
||||
|
||||
function normalizeApiFormat(apiFormat: unknown): ApiCallFormat {
|
||||
return apiFormat === "gemini" ? "gemini" : "openai";
|
||||
}
|
||||
|
||||
function uniqueRawModels(models: string[]) {
|
||||
return Array.from(new Set((models || []).map((model) => modelOptionName(model).trim()).filter(Boolean)));
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user