feat(model): enhance model configuration and script handling for audio and video capabilities

This commit is contained in:
HouYunFei
2026-07-15 10:13:32 +08:00
parent a4dcc679c9
commit c57f7d61a7
13 changed files with 690 additions and 266 deletions
+3 -1
View File
@@ -2,11 +2,13 @@
## Unreleased
+ [新增] 支持自定义生图/视频接口调用方式以适配不同中转站。
## v0.7.0 - 2026-07-14
+ [新增] Agent 对话消息改用 streamdown 流式渲染,提升Markdown 内容展示效果。
+ [新增] Agent 新增画布、工作台、提示词库和素材等站点级工具。
+ [新增] Agent 面板改为全站常驻右侧栏,开关时同步推动顶栏和页面内容,并保留独立入口按钮
+ [新增] Agent 面板改为全站常驻右侧栏,开关时同步推动顶栏和页面内容。
+ [新增] Agent 新增 `site_navigate` 工具,支持页面跳转。
+ [新增] Agent 对话运行中支持一键停止,中断当前 Codex turn。
+ [新增] 画布节点支持统一维护名称字段,默认显示在节点上方,并可直接双击名称编辑。
@@ -146,9 +146,9 @@ function buildNodeConfig(globalConfig: AiConfig, node: CanvasNodeData, mode: Can
const defaultModel = mode === "image" ? globalConfig.imageModel : mode === "video" ? globalConfig.videoModel : mode === "audio" ? globalConfig.audioModel : globalConfig.textModel;
const fallbackModel = mode === "image" ? defaultConfig.imageModel : mode === "video" ? defaultConfig.videoModel : mode === "audio" ? defaultConfig.audioModel : defaultConfig.textModel;
const currentModel = node.metadata?.model;
const model = currentModel && modelMatchesCapability(currentModel, mode)
const model = currentModel && modelMatchesCapability(globalConfig, currentModel, mode)
? currentModel
: defaultModel && modelMatchesCapability(defaultModel, mode)
: defaultModel && modelMatchesCapability(globalConfig, defaultModel, mode)
? defaultModel
: fallbackModel;
return {
@@ -134,9 +134,9 @@ function buildNodeConfig(globalConfig: AiConfig, node: CanvasNodeData, mode: Can
const defaultModel = mode === "image" ? globalConfig.imageModel : mode === "video" ? globalConfig.videoModel : mode === "audio" ? globalConfig.audioModel : globalConfig.textModel;
const fallbackModel = mode === "image" ? defaultConfig.imageModel : mode === "video" ? defaultConfig.videoModel : mode === "audio" ? defaultConfig.audioModel : defaultConfig.textModel;
const currentModel = node.metadata?.model;
const model = currentModel && modelMatchesCapability(currentModel, mode)
const model = currentModel && modelMatchesCapability(globalConfig, currentModel, mode)
? currentModel
: defaultModel && modelMatchesCapability(defaultModel, mode)
: defaultModel && modelMatchesCapability(globalConfig, defaultModel, mode)
? defaultModel
: fallbackModel;
return {
+45 -173
View File
@@ -1,20 +1,18 @@
import { App, Button, Form, Input, Modal, Progress, Select, Tabs } from "antd";
import { CircleAlert, Cloud, Plus, RefreshCw, Trash2, Wifi } from "lucide-react";
import { Cloud, Pencil, Plus, RefreshCw, Trash2, Wifi } from "lucide-react";
import { useEffect, useState } from "react";
import { ModelPicker } from "@/components/model-picker";
import { fetchChannelModels } from "@/services/api/image";
import { ChannelEditorDrawer } from "@/components/layout/channel-editor-drawer";
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, defaultBaseUrlForApiFormat, filterModelsByCapability, modelOptionLabel, modelOptionsFromChannels, normalizeModelOptionValue, useConfigStore, type AiConfig, type ApiCallFormat, type ConfigTabKey, type ModelCapability, type ModelChannel } from "@/stores/use-config-store";
import { createModelChannel, modelOptionsFromChannels, normalizeModelOptionValue, selectableModelsByCapability, useConfigStore, type AiConfig, type ApiCallFormat, type ConfigTabKey, type ModelCapability, type ModelChannel } from "@/stores/use-config-store";
type ModelGroup = {
capability: ModelCapability;
modelKey: "imageModel" | "videoModel" | "textModel" | "audioModel";
modelsKey: "imageModels" | "videoModels" | "textModels" | "audioModels";
defaultLabel: string;
optionsLabel: string;
};
type WebdavDomainProgress = {
@@ -26,15 +24,10 @@ type WebdavDomainProgress = {
};
const modelGroups: ModelGroup[] = [
{ capability: "image", modelKey: "imageModel", modelsKey: "imageModels", defaultLabel: "默认生图模型", optionsLabel: "生图模型可选项" },
{ capability: "video", modelKey: "videoModel", modelsKey: "videoModels", defaultLabel: "默认视频模型", optionsLabel: "视频模型可选项" },
{ capability: "text", modelKey: "textModel", modelsKey: "textModels", defaultLabel: "默认文本模型", optionsLabel: "文本模型可选项" },
{ 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" },
{ capability: "image", modelKey: "imageModel", defaultLabel: "默认生图模型" },
{ capability: "video", modelKey: "videoModel", defaultLabel: "默认视频模型" },
{ capability: "text", modelKey: "textModel", defaultLabel: "默认文本模型" },
{ capability: "audio", modelKey: "audioModel", defaultLabel: "默认音频模型" },
];
const webdavDomainKeys: AppSyncDomainKey[] = ["canvas", "assets", "image-workbench", "video-workbench"];
@@ -58,7 +51,7 @@ function createWebdavDomainProgress(): Record<AppSyncDomainKey, WebdavDomainProg
export function AppConfigPanel({ showDoneButton = false, initialTab = "channels" }: { showDoneButton?: boolean; initialTab?: ConfigTabKey }) {
const { message } = App.useApp();
const [activeTab, setActiveTab] = useState<ConfigTabKey>(initialTab);
const [loadingChannelId, setLoadingChannelId] = useState("");
const [editingChannelId, setEditingChannelId] = useState("");
const [testingWebdav, setTestingWebdav] = useState(false);
const [syncingWebdav, setSyncingWebdav] = useState(false);
const [webdavSyncStatus, setWebdavSyncStatus] = useState("");
@@ -70,8 +63,8 @@ export function AppConfigPanel({ showDoneButton = false, initialTab = "channels"
const shouldPromptContinue = useConfigStore((state) => state.shouldPromptContinue);
const setConfigDialogOpen = useConfigStore((state) => state.setConfigDialogOpen);
const clearPromptContinue = useConfigStore((state) => state.clearPromptContinue);
const modelOptions = config.models.map((model) => ({ label: modelOptionLabel(config, model), value: model }));
const webdavReady = Boolean(webdav.url.trim());
const editingChannel = config.channels.find((channel) => channel.id === editingChannelId) || null;
useEffect(() => setActiveTab(initialTab), [initialTab]);
const saveConfig = (nextConfig: AiConfig) => {
@@ -86,22 +79,12 @@ export function AppConfigPanel({ showDoneButton = false, initialTab = "channels"
clearPromptContinue();
};
const updateChannels = (channels: ModelChannel[]) => {
const nextConfig = withChannels(config, channels);
saveConfig(nextConfig);
};
const updateChannel = (id: string, patch: Partial<ModelChannel>) => {
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 updateChannels = (channels: ModelChannel[]) => saveConfig(withChannels(config, channels));
const addChannel = () => {
updateChannels([...config.channels, createModelChannel({ name: `渠道 ${config.channels.length + 1}` })]);
const channel = createModelChannel({ name: `渠道 ${config.channels.length + 1}` });
updateChannels([...config.channels, channel]);
setEditingChannelId(channel.id);
};
const deleteChannel = (id: string) => {
@@ -112,46 +95,8 @@ export function AppConfigPanel({ showDoneButton = false, initialTab = "channels"
updateChannels(config.channels.filter((channel) => channel.id !== id));
};
const refreshChannelModels = async (channel: ModelChannel) => {
if (!channel.baseUrl.trim() || !channel.apiKey.trim()) {
message.error("请先填写该渠道的 Base URL 和 API Key");
return;
}
setLoadingChannelId(channel.id);
try {
const models = await fetchChannelModels(channel);
updateChannels(config.channels.map((item) => (item.id === channel.id ? { ...item, models } : item)));
message.success(`${channel.name} 模型列表已更新`);
} catch (error) {
message.error(error instanceof Error ? error.message : "读取模型失败");
} finally {
setLoadingChannelId("");
}
};
const refreshAllModels = async () => {
const runnable = config.channels.filter((channel) => channel.baseUrl.trim() && channel.apiKey.trim());
if (!runnable.length) {
message.error("请先填写至少一个渠道的 Base URL 和 API Key");
return;
}
setLoadingChannelId("all");
try {
const entries = await Promise.all(runnable.map(async (channel) => [channel.id, await fetchChannelModels(channel)] as const));
const modelMap = new Map(entries);
updateChannels(config.channels.map((channel) => (modelMap.has(channel.id) ? { ...channel, models: modelMap.get(channel.id) || [] } : channel)));
message.success("模型列表已更新");
} catch (error) {
message.error(error instanceof Error ? error.message : "读取模型失败");
} finally {
setLoadingChannelId("");
}
};
const updateCapabilityModels = (group: ModelGroup, models: string[]) => {
const next = uniqueModels(models.map((model) => normalizeModelOptionValue(model, config.channels)).filter(Boolean));
updateConfig(group.modelsKey, next);
if (!next.includes(config[group.modelKey])) updateConfig(group.modelKey, next[0] || "");
const saveChannel = (channel: ModelChannel) => {
updateChannels(config.channels.map((item) => (item.id === channel.id ? channel : item)));
};
const testWebdav = async () => {
@@ -215,107 +160,48 @@ export function AppConfigPanel({ showDoneButton = false, initialTab = "channels"
key: "channels",
label: "渠道",
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 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"></span>
<span>Tab </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")}>
</Button>
</div>
</div>
<div className="flex shrink-0 gap-2">
<Button icon={<RefreshCw className="size-4" />} loading={Boolean(loadingChannelId)} onClick={() => void refreshAllModels()}>
</Button>
<div>
<div className="mb-4 flex flex-wrap items-center justify-between gap-3">
<div className="text-xs text-stone-500">///</div>
<Button type="primary" icon={<Plus className="size-4" />} onClick={addChannel}>
</Button>
</div>
</div>
<div className="space-y-3">
<div className="space-y-2">
{config.channels.map((channel) => (
<section key={channel.id} className="rounded-lg border border-stone-200 p-3 dark:border-stone-800">
<div className="mb-3 flex items-center justify-between gap-3">
<div key={channel.id} className="flex items-center justify-between gap-3 rounded-lg border border-stone-200 px-4 py-3 dark:border-stone-800">
<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">
{apiFormatLabel(channel.apiFormat)} · {channel.models.length}
<div className="mt-1 truncate text-xs text-stone-500">
{apiFormatLabel(channel.apiFormat)} · {channel.models.length} · {channel.baseUrl || "未填写接口地址"}
</div>
</div>
<div className="flex shrink-0 gap-2">
<Button size="small" loading={loadingChannelId === channel.id} onClick={() => void refreshChannelModels(channel)}>
<Button size="small" icon={<Pencil className="size-3.5" />} onClick={() => setEditingChannelId(channel.id)}>
</Button>
<Button size="small" danger icon={<Trash2 className="size-3.5" />} onClick={() => deleteChannel(channel.id)} />
</div>
</div>
<div className="grid gap-4 md:grid-cols-2">
<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 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>
</section>
))}
</div>
</Form>
</div>
),
},
{
key: "models",
label: "模型",
key: "preferences",
label: "偏好设置",
children: (
<Form layout="vertical" requiredMark={false}>
<div className="mb-4 rounded-lg border border-stone-200 p-3 dark:border-stone-800">
<div className="text-sm font-semibold"></div>
<div className="mt-1 text-xs leading-5 text-stone-500"></div>
</div>
<div className="grid gap-4 md:grid-cols-2">
{modelGroups.map((group) => (
<Form.Item key={group.modelsKey} label={group.optionsLabel} className="mb-0">
<Select
mode="tags"
showSearch
allowClear
maxTagCount="responsive"
placeholder={config.models.length ? `请选择或输入${group.optionsLabel}` : "先到渠道里填写或拉取模型"}
value={config[group.modelsKey]}
options={modelOptions}
onChange={(models) => updateCapabilityModels(group, models)}
/>
</Form.Item>
))}
</div>
<div className="mt-4 grid gap-4 md:grid-cols-2 xl:grid-cols-4">
<div className="mb-2 text-sm font-semibold"></div>
<div className="mb-4 grid gap-4 md:grid-cols-2 xl:grid-cols-4">
{modelGroups.map((group) => (
<Form.Item key={group.modelKey} label={group.defaultLabel} className="mb-0">
<ModelPicker config={config} value={config[group.modelKey]} onChange={(model) => updateConfig(group.modelKey, model)} capability={group.capability} fullWidth />
</Form.Item>
))}
</div>
</Form>
),
},
{
key: "preferences",
label: "生成偏好",
children: (
<Form layout="vertical" requiredMark={false}>
<div className="mb-2 text-sm font-semibold"></div>
<div className="grid gap-4 md:grid-cols-4">
<Form.Item label="画布默认生图张数" extra="新建画布生图和配置节点默认使用,单个节点仍可单独覆盖。" className="mb-4">
<Input
@@ -407,6 +293,7 @@ export function AppConfigPanel({ showDoneButton = false, initialTab = "channels"
</Button>
</div>
) : null}
<ChannelEditorDrawer open={Boolean(editingChannel)} channel={editingChannel} onSave={saveChannel} onClose={() => setEditingChannelId("")} />
</>
);
}
@@ -420,7 +307,7 @@ export function AppConfigModal() {
title={
<div>
<div className="text-lg font-semibold"></div>
<div className="mt-1 text-xs font-normal text-stone-500"></div>
<div className="mt-1 text-xs font-normal text-stone-500"></div>
</div>
}
open={isConfigOpen}
@@ -436,48 +323,33 @@ export function AppConfigModal() {
}
function withChannels(config: AiConfig, channels: ModelChannel[]): AiConfig {
const models = modelOptionsFromChannels(channels);
const imageModels = keepOrSuggest(config.imageModels, filterModelsByCapability(models, "image"), models);
const videoModels = keepOrSuggest(config.videoModels, filterModelsByCapability(models, "video"), models);
const textModels = keepOrSuggest(config.textModels, filterModelsByCapability(models, "text"), models);
const audioModels = keepOrSuggest(config.audioModels, filterModelsByCapability(models, "audio"), models);
return {
const next: AiConfig = {
...config,
channels,
models,
models: modelOptionsFromChannels(channels),
baseUrl: channels[0]?.baseUrl || config.baseUrl,
apiKey: channels[0]?.apiKey || config.apiKey,
apiFormat: channels[0]?.apiFormat || config.apiFormat,
imageModels,
videoModels,
textModels,
audioModels,
imageModel: normalizeDefaultModel(config.imageModel, imageModels),
videoModel: normalizeDefaultModel(config.videoModel, videoModels),
textModel: normalizeDefaultModel(config.textModel, textModels),
audioModel: normalizeDefaultModel(config.audioModel, audioModels),
};
return {
...next,
imageModel: pickDefaultModel(next, "image", config.imageModel),
videoModel: pickDefaultModel(next, "video", config.videoModel),
textModel: pickDefaultModel(next, "text", config.textModel),
audioModel: pickDefaultModel(next, "audio", config.audioModel),
};
}
function keepOrSuggest(current: string[], suggested: string[], allModels: string[]) {
const available = new Set(allModels);
const kept = uniqueModels(current).filter((model) => available.has(model));
return kept.length ? kept : suggested;
}
function normalizeDefaultModel(value: string, options: string[]) {
if (options.includes(value)) return value;
return options[0] || value;
function pickDefaultModel(config: AiConfig, capability: ModelCapability, current: string) {
const options = selectableModelsByCapability(config, capability);
const normalized = normalizeModelOptionValue(current, config.channels);
return options.includes(normalized) ? normalized : options[0] || "";
}
function normalizeImageCount(value: string) {
return String(Math.max(1, Math.min(15, Math.floor(Math.abs(Number(value)) || 3))));
}
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";
}
@@ -0,0 +1,134 @@
import { Button, Drawer, Input, Segmented, Select, Space } from "antd";
import { ListPlus, Trash2 } from "lucide-react";
import { useEffect, useState } from "react";
import { defaultBaseUrlForApiFormat, guessCapability, normalizeChannelModels, type ApiCallFormat, type ChannelModel, type ModelCapability, type ModelChannel } from "@/stores/use-config-store";
import { ModelScriptEditor } from "./model-script-editor";
import { ModelSelectModal } from "./model-select-modal";
const apiFormatOptions: Array<{ label: string; value: ApiCallFormat }> = [
{ label: "OpenAI", value: "openai" },
{ label: "Gemini", value: "gemini" },
];
const capabilityOptions: Array<{ label: string; value: ModelCapability }> = [
{ label: "生图", value: "image" },
{ label: "视频", value: "video" },
{ label: "文本", value: "text" },
{ label: "音频", value: "audio" },
];
type ScriptTarget = { name: string; capability: ModelCapability; value: string };
export function ChannelEditorDrawer({ open, channel, onSave, onClose }: { open: boolean; channel: ModelChannel | null; onSave: (channel: ModelChannel) => void; onClose: () => void }) {
const [draft, setDraft] = useState<ModelChannel | null>(channel);
const [selectOpen, setSelectOpen] = useState(false);
const [scriptTarget, setScriptTarget] = useState<ScriptTarget | null>(null);
useEffect(() => {
if (open && channel) setDraft(channel);
}, [open, channel]);
if (!draft) return null;
const patch = (value: Partial<ModelChannel>) => setDraft((current) => (current ? { ...current, ...value } : current));
const setModels = (models: ChannelModel[]) => patch({ models });
const changeApiFormat = (apiFormat: ApiCallFormat) => {
const baseUrl = !draft.baseUrl.trim() || draft.baseUrl.trim() === defaultBaseUrlForApiFormat(draft.apiFormat) ? defaultBaseUrlForApiFormat(apiFormat) : draft.baseUrl;
patch({ apiFormat, baseUrl });
};
const applySelection = (names: string[]) => {
const map = new Map(draft.models.map((model) => [model.name, model]));
setModels(names.map((name) => map.get(name) || { name, capability: guessCapability(name) }));
};
const setCapability = (name: string, capability: ModelCapability) => setModels(draft.models.map((model) => (model.name === name ? { ...model, capability } : model)));
const setScript = (name: string, script: string) => setModels(draft.models.map((model) => (model.name === name ? { ...model, script: script || undefined } : model)));
const removeModel = (name: string) => setModels(draft.models.filter((model) => model.name !== name));
const save = () => {
onSave({ ...draft, name: draft.name.trim() || "未命名渠道", models: normalizeChannelModels(draft.models) });
onClose();
};
return (
<Drawer
open={open}
width={640}
title="编辑渠道"
onClose={onClose}
styles={{ body: { paddingTop: 16 } }}
extra={
<Space>
<Button onClick={onClose}></Button>
<Button type="primary" onClick={save}>
</Button>
</Space>
}
>
<div className="grid gap-4 md:grid-cols-2">
<label className="block">
<span className="mb-1 block text-sm font-medium"></span>
<Input value={draft.name} onChange={(event) => patch({ name: event.target.value })} />
</label>
<label className="block">
<span className="mb-1 block text-sm font-medium"></span>
<Select className="w-full" value={draft.apiFormat} options={apiFormatOptions} onChange={changeApiFormat} />
</label>
<label className="block md:col-span-2">
<span className="mb-1 block text-sm font-medium"></span>
<Input value={draft.baseUrl} onChange={(event) => patch({ baseUrl: event.target.value })} placeholder="https://api.example.com" />
</label>
<label className="block md:col-span-2">
<span className="mb-1 block text-sm font-medium">API Key</span>
<Input.Password value={draft.apiKey} onChange={(event) => patch({ apiKey: event.target.value })} placeholder="sk-..." />
</label>
</div>
<div className="mt-6 mb-3 flex flex-wrap items-center justify-between gap-2">
<div>
<div className="text-sm font-semibold"></div>
<div className="mt-0.5 text-xs text-stone-500"> {draft.models.length} </div>
</div>
<Button type="primary" icon={<ListPlus className="size-4" />} onClick={() => setSelectOpen(true)}>
</Button>
</div>
<div className="space-y-2 rounded-lg border border-stone-200 p-2 dark:border-stone-800">
{draft.models.length ? (
draft.models.map((model) => (
<div key={model.name} className="flex flex-wrap items-center gap-3 rounded-md px-2 py-1.5 hover:bg-stone-50 dark:hover:bg-stone-900/40">
<span className="min-w-0 flex-1 truncate text-sm" title={model.name}>
{model.name}
</span>
<div className="flex shrink-0 items-center gap-2">
<Segmented size="small" value={model.capability} options={capabilityOptions} onChange={(value) => setCapability(model.name, value as ModelCapability)} />
<Button size="small" type={model.script ? "primary" : "default"} ghost={Boolean(model.script)} onClick={() => setScriptTarget({ name: model.name, capability: model.capability, value: model.script || "" })}>
{model.script ? "脚本已设" : "调用脚本"}
</Button>
<Button size="small" danger type="text" icon={<Trash2 className="size-3.5" />} onClick={() => removeModel(model.name)} />
</div>
</div>
))
) : (
<div className="px-2 py-8 text-center text-sm text-stone-500"></div>
)}
</div>
<ModelSelectModal open={selectOpen} channel={draft} selectedNames={draft.models.map((model) => model.name)} onConfirm={applySelection} onClose={() => setSelectOpen(false)} />
<ModelScriptEditor
open={Boolean(scriptTarget)}
capability={scriptTarget?.capability || "text"}
modelName={scriptTarget?.name || ""}
value={scriptTarget?.value || ""}
onSave={(script) => scriptTarget && setScript(scriptTarget.name, script)}
onClose={() => setScriptTarget(null)}
/>
</Drawer>
);
}
@@ -0,0 +1,67 @@
import { Button, Input, Modal } from "antd";
import { useEffect, useState } from "react";
import { PLUGIN_TEMPLATES } from "@/services/api/model-plugin";
import type { ModelCapability } from "@/stores/use-config-store";
const capabilityLabels: Record<ModelCapability, string> = { image: "生图", video: "视频", text: "文本", audio: "音频" };
const variableHints = [
"input:本次请求的归一化输入(prompt / references / messages / params / body",
"config{ baseUrl, apiKey, model, apiFormat, systemPrompt }",
"http{ url(path), post(path, body, opts), get(path, opts) },已绑定当前渠道",
"poll(request, extract, { intervalMs, timeoutMs }):轮询直到 extract 返回真值",
"sleep(ms)、signal:延时与取消信号;onDelta(text):推送流式文本(文本模型)",
];
export function ModelScriptEditor({ open, capability, modelName, value, onSave, onClose }: { open: boolean; capability: ModelCapability; modelName: string; value: string; onSave: (script: string) => void; onClose: () => void }) {
const [draft, setDraft] = useState(value);
useEffect(() => {
if (open) setDraft(value);
}, [open, value]);
return (
<Modal
open={open}
title={
<div>
<div className="text-base font-semibold"> · {modelName}</div>
<div className="mt-1 text-xs font-normal text-stone-500">{capabilityLabels[capability]}使</div>
</div>
}
width={720}
centered
onCancel={onClose}
styles={{ body: { maxHeight: "64vh", overflowY: "auto" } }}
footer={[
<Button key="template" onClick={() => setDraft(PLUGIN_TEMPLATES[capability])}>
</Button>,
<Button key="reset" danger onClick={() => setDraft("")}>
</Button>,
<Button key="cancel" onClick={onClose}>
</Button>,
<Button
key="save"
type="primary"
onClick={() => {
onSave(draft.trim());
onClose();
}}
>
</Button>,
]}
>
<div className="mb-3 rounded-md border border-stone-200 bg-stone-50 p-3 text-xs leading-6 text-stone-600 dark:border-stone-800 dark:bg-stone-900/40 dark:text-stone-300">
<div className="mb-1 font-semibold"> return </div>
{variableHints.map((hint) => (
<div key={hint}>· {hint}</div>
))}
</div>
<Input.TextArea value={draft} onChange={(event) => setDraft(event.target.value)} autoSize={{ minRows: 12, maxRows: 24 }} spellCheck={false} placeholder="留空使用系统默认调用;点击“插入模板”查看示例。" style={{ fontFamily: "var(--font-mono, ui-monospace, SFMono-Regular, Menlo, monospace)", fontSize: 12 }} />
</Modal>
);
}
+1 -1
View File
@@ -83,7 +83,7 @@ export function ModelPicker({ config, value, onChange, capability, className, fu
function emptyModelLabel(config: AiConfig, capability?: ModelCapability) {
const label = capability === "image" ? "生图" : capability === "video" ? "视频" : capability === "text" ? "文本" : capability === "audio" ? "音频" : "";
if (capability && config.models.length) return "请先在上方配置可选模型";
if (capability && config.models.length) return `请先在渠道里为${label}指定模型`;
return config.models.length ? `暂无匹配的${label}模型` : "请先到配置里添加渠道和模型";
}
+3 -3
View File
@@ -6,7 +6,7 @@ import { imageAspectOptions, imageQualityOptions } from "@/components/image-sett
import { videoResolutionOptions, videoSecondOptions, videoSizeOptions } from "@/components/video-settings-panel";
import { useCanvasStore } from "@/stores/canvas/use-canvas-store";
import { useAssetStore } from "@/stores/use-asset-store";
import { modelOptionLabel, modelOptionName, normalizeModelOptionValue, useConfigStore } from "@/stores/use-config-store";
import { modelOptionLabel, modelOptionName, normalizeModelOptionValue, selectableModelsByCapability, useConfigStore } from "@/stores/use-config-store";
import { useWorkbenchAgentStore } from "@/stores/use-workbench-agent-store";
// 在网页端执行 Agent 的「站点级」工具(画布列表、工作台生成、提示词搜索、素材增删查等)。
@@ -87,7 +87,7 @@ function getImageConfig() {
const model = config.imageModel || config.model;
return {
current: { model, modelName: modelOptionName(model), quality: config.quality || "auto", size: config.size || "1:1", count: config.count || "1" },
models: config.imageModels.map((value) => ({ value, label: modelOptionLabel(config, value) })),
models: selectableModelsByCapability(config, "image").map((value) => ({ value, label: modelOptionLabel(config, value) })),
qualityOptions: imageQualityOptions,
sizeOptions: imageAspectOptions,
countRange: { min: 1, max: 15 },
@@ -135,7 +135,7 @@ function getVideoConfig() {
generateAudio: config.videoGenerateAudio !== "false",
watermark: config.videoWatermark === "true",
},
models: config.videoModels.map((value) => ({ value, label: modelOptionLabel(config, value) })),
models: selectableModelsByCapability(config, "video").map((value) => ({ value, label: modelOptionLabel(config, value) })),
sizeOptions: videoSizeOptions,
secondsOptions: videoSecondOptions,
resolutionOptions: videoResolutionOptions,
+35 -2
View File
@@ -2,7 +2,8 @@ import axios from "axios";
import { audioMimeType, normalizeAudioFormatValue, normalizeAudioSpeedValue, normalizeAudioVoiceValue } from "@/lib/audio-generation";
import { uploadMediaFile, type UploadedFile } from "@/services/file-storage";
import { buildApiUrl, resolveModelRequestConfig, type AiConfig } from "@/stores/use-config-store";
import { buildApiUrl, resolveModelRequestConfig, resolveModelScript, type AiConfig } from "@/stores/use-config-store";
import { runModelPlugin } from "./model-plugin";
type RequestOptions = { signal?: AbortSignal };
@@ -20,8 +21,26 @@ function aiHeaders(config: AiConfig) {
export async function requestAudioGeneration(config: AiConfig, prompt: string, options?: RequestOptions): Promise<Blob> {
const requestConfig = resolveModelRequestConfig(config, config.model || config.audioModel);
const model = requestConfig.model.trim();
assertAudioConfig(requestConfig, model);
const format = normalizeAudioFormatValue(config.audioFormat);
const script = resolveModelScript(config, config.model || config.audioModel);
if (script) {
if (!model) throw new Error("请先配置音频模型");
if (!requestConfig.baseUrl.trim()) throw new Error("请先配置 Base URL");
if (!requestConfig.apiKey.trim()) throw new Error("请先配置 API Key");
try {
const result = await runModelPlugin({
capability: "audio",
script,
config: requestConfig,
input: { prompt, params: { voice: normalizeAudioVoiceValue(config.audioVoice), format, speed: normalizeAudioSpeedValue(config.audioSpeed), instructions: config.audioInstructions.trim() } },
signal: options?.signal,
});
return await audioPluginBlob(result, format);
} catch (error) {
throw new Error(readAxiosError(error, "音频生成失败"));
}
}
assertAudioConfig(requestConfig, model);
const instructions = config.audioInstructions.trim();
try {
@@ -44,6 +63,20 @@ export async function requestAudioGeneration(config: AiConfig, prompt: string, o
}
}
async function audioPluginBlob(result: unknown, format: string): Promise<Blob> {
if (result instanceof Blob) return result.type.startsWith("audio/") ? result : new Blob([result], { type: audioMimeType(format) });
let source = "";
if (typeof result === "string") source = result;
else if (result && typeof result === "object") {
const record = result as Record<string, unknown>;
source = typeof record.b64_json === "string" ? record.b64_json : typeof record.data === "string" ? record.data : typeof record.url === "string" ? record.url : "";
}
if (!source) throw new Error("模型调用脚本没有返回音频");
const url = source.startsWith("data:") || /^https?:/i.test(source) ? source : `data:${audioMimeType(format)};base64,${source}`;
const blob = await (await fetch(url)).blob();
return blob.type.startsWith("audio/") ? blob : new Blob([blob], { type: audioMimeType(format) });
}
export async function storeGeneratedAudio(blob: Blob, format = "mp3"): Promise<UploadedFile> {
const audio = blob.type.startsWith("audio/") ? blob : new Blob([blob], { type: audioMimeType(format) });
return uploadMediaFile(audio, "audio");
+65 -1
View File
@@ -1,6 +1,7 @@
import axios from "axios";
import { buildApiUrl, resolveModelRequestConfig, type AiConfig, type ModelChannel } from "@/stores/use-config-store";
import { buildApiUrl, resolveModelRequestConfig, resolveModelScript, type AiConfig, type ModelChannel } from "@/stores/use-config-store";
import { normalizePluginImages, runModelPlugin } from "./model-plugin";
import { nanoid } from "nanoid";
import { dataUrlToFile } from "@/lib/image-utils";
import { buildImageReferencePromptText } from "@/lib/image-reference-prompt";
@@ -656,6 +657,28 @@ function parseGeminiImagePayload(payload: GeminiPayload) {
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)));
const script = resolveModelScript(config, config.model || config.imageModel);
if (script) {
const quality = normalizeQuality(config.quality);
const requestSize = resolveRequestSize(quality, config.size);
try {
const result = await runModelPlugin({
capability: "image",
script,
config: requestConfig,
input: {
prompt: withSystemPrompt(requestConfig, prompt),
references: [],
params: { size: requestSize, quality, count: n },
body: { model: requestConfig.model, n, ...(quality ? { quality } : {}), ...(requestSize ? { size: requestSize } : {}), response_format: "b64_json", output_format: IMAGE_OUTPUT_FORMAT },
},
signal: options?.signal,
});
return normalizePluginImages(result).map((dataUrl) => ({ id: nanoid(), dataUrl }));
} catch (error) {
throw new Error(readAxiosError(error, "请求失败"));
}
}
if (requestConfig.apiFormat === "gemini") {
try {
return await requestGeminiImages(requestConfig, prompt, [], n, options);
@@ -693,6 +716,29 @@ export async function requestEdit(config: AiConfig, prompt: string, references:
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);
const script = resolveModelScript(config, config.model || config.imageModel);
if (script) {
const quality = normalizeQuality(config.quality);
const requestSize = resolveRequestSize(quality, config.size);
const refs = await Promise.all(references.map((image) => imageToDataUrl(image)));
try {
const result = await runModelPlugin({
capability: "image",
script,
config: requestConfig,
input: {
prompt: withSystemPrompt(requestConfig, requestPrompt),
references: refs,
params: { size: requestSize, quality, count: n },
body: { model: requestConfig.model, n, ...(quality ? { quality } : {}), ...(requestSize ? { size: requestSize } : {}), response_format: "b64_json", output_format: IMAGE_OUTPUT_FORMAT },
},
signal: options?.signal,
});
return normalizePluginImages(result).map((dataUrl) => ({ id: nanoid(), dataUrl }));
} catch (error) {
throw new Error(readAxiosError(error, "请求失败"));
}
}
if (requestConfig.apiFormat === "gemini") {
if (mask) throw new Error("Gemini 调用格式暂不支持蒙版编辑");
try {
@@ -730,6 +776,24 @@ 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);
const script = resolveModelScript(config, config.model || config.textModel);
if (script) {
try {
const answer = await runModelPlugin<string>({
capability: "text",
script,
config: requestConfig,
input: { messages: withSystemMessage(requestConfig, messages), body: { model: requestConfig.model } },
signal: options?.signal,
onDelta,
});
const text = String(answer ?? "").trim() || "没有返回内容";
if (text === "没有返回内容") onDelta(text);
return text;
} catch (error) {
throw new Error(readAxiosError(error, "请求失败"));
}
}
try {
if (requestConfig.apiFormat === "gemini") {
const answer = (await requestGeminiStreamingResponse(requestConfig, toGeminiBody(requestConfig, messages), onDelta, options)).content || "没有返回内容";
+204
View File
@@ -0,0 +1,204 @@
import axios from "axios";
import { buildApiUrl, type AiConfig, type ModelCapability } from "@/stores/use-config-store";
type RequestOptions = { signal?: AbortSignal };
export type PluginHttpOptions = {
headers?: Record<string, string>;
params?: Record<string, unknown>;
responseType?: "json" | "blob" | "text" | "arraybuffer";
};
export type PluginHttp = {
url: (path: string) => string;
post: (path: string, body?: unknown, options?: PluginHttpOptions) => Promise<unknown>;
get: (path: string, options?: PluginHttpOptions) => Promise<unknown>;
};
export type PluginPollOptions = { intervalMs?: number; timeoutMs?: number };
export type PluginConfigView = {
baseUrl: string;
apiKey: string;
model: string;
apiFormat: string;
systemPrompt: string;
};
export type RunPluginArgs = {
capability: ModelCapability;
script: string;
config: AiConfig;
input: Record<string, unknown>;
signal?: AbortSignal;
onDelta?: (text: string) => void;
};
function pluginHeaders(config: AiConfig, extra?: Record<string, string>, hasJsonBody = false): Record<string, string> {
const headers: Record<string, string> = {};
if (config.apiFormat === "gemini") headers["x-goog-api-key"] = config.apiKey;
else headers.Authorization = `Bearer ${config.apiKey}`;
if (hasJsonBody) headers["Content-Type"] = "application/json";
return { ...headers, ...extra };
}
function pluginUrl(config: AiConfig, path: string) {
if (/^https?:/i.test(path)) return path;
return buildApiUrl(config.baseUrl, path.startsWith("/") ? path : `/${path}`);
}
function createPluginHttp(config: AiConfig, options?: RequestOptions): PluginHttp {
const request = async (method: "get" | "post", path: string, body: unknown, opts?: PluginHttpOptions) => {
const isForm = typeof FormData !== "undefined" && body instanceof FormData;
const response = await axios.request({
method,
url: pluginUrl(config, path),
data: method === "post" ? body : undefined,
params: opts?.params,
headers: pluginHeaders(config, opts?.headers, method === "post" && !isForm && body !== undefined),
responseType: opts?.responseType || "json",
signal: options?.signal,
});
return response.data;
};
return {
url: (path) => pluginUrl(config, path),
post: (path, body, opts) => request("post", path, body, opts),
get: (path, opts) => request("get", path, undefined, opts),
};
}
function sleep(ms: number, signal?: AbortSignal) {
return new Promise<void>((resolve, reject) => {
if (signal?.aborted) {
reject(new DOMException("Aborted", "AbortError"));
return;
}
const timer = setTimeout(resolve, ms);
signal?.addEventListener(
"abort",
() => {
clearTimeout(timer);
reject(new DOMException("Aborted", "AbortError"));
},
{ once: true },
);
});
}
function createPoll(signal?: AbortSignal) {
return async function poll<T, R>(request: () => Promise<T>, extract: (value: T) => R | null | undefined | false, options?: PluginPollOptions): Promise<R> {
const intervalMs = options?.intervalMs ?? 2500;
const timeoutMs = options?.timeoutMs ?? 300000;
const deadline = performance.now() + timeoutMs;
for (;;) {
if (signal?.aborted) throw new DOMException("Aborted", "AbortError");
const result = extract(await request());
if (result !== null && result !== undefined && result !== false) return result;
if (performance.now() >= deadline) throw new Error("插件轮询超时,请检查调用脚本或稍后重试");
await sleep(intervalMs, signal);
}
};
}
/**
* Run a user-authored model call script. The script body runs as an async function with these locals:
* input —— normalized request input for this capability (prompt / references / messages / params / body)
* config —— { baseUrl, apiKey, model, apiFormat, systemPrompt }
* http —— { url(path), post(path, body, opts), get(path, opts) } bound to the model's channel
* poll —— poll(request, extract, { intervalMs, timeoutMs }) resolves with the first truthy extract result
* sleep —— sleep(ms)
* signal —— AbortSignal for cancellation
* onDelta —— (text) => void, push streaming text (text capability only)
* The script must `return` the result; each caller normalizes it to its capability's shape.
*/
export async function runModelPlugin<T = unknown>(args: RunPluginArgs): Promise<T> {
const configView: PluginConfigView = {
baseUrl: args.config.baseUrl,
apiKey: args.config.apiKey,
model: args.config.model,
apiFormat: args.config.apiFormat,
systemPrompt: args.config.systemPrompt,
};
const http = createPluginHttp(args.config, { signal: args.signal });
const poll = createPoll(args.signal);
const runner = new Function(
"input",
"config",
"http",
"poll",
"sleep",
"signal",
"onDelta",
`"use strict"; return (async () => {\n${args.script}\n})();`,
) as (input: unknown, config: PluginConfigView, http: PluginHttp, poll: ReturnType<typeof createPoll>, sleep: (ms: number) => Promise<void>, signal: AbortSignal | undefined, onDelta?: (text: string) => void) => Promise<T>;
try {
return await runner(args.input, configView, http, poll, (ms: number) => sleep(ms, args.signal), args.signal, args.onDelta);
} catch (error) {
if (error instanceof DOMException && error.name === "AbortError") throw error;
if (axios.isCancel(error)) throw error;
const message = error instanceof Error ? error.message : String(error);
throw new Error(`模型调用脚本执行失败:${message}`);
}
}
export const PLUGIN_TEMPLATES: Record<ModelCapability, string> = {
image: `// 输入:input.prompt / input.references(dataURL[]) / input.body(默认请求体) / input.params
// 返回:dataURL 或 URL 字符串,或它们的数组,或 [{ dataUrl }]
const data = await http.post("/images/generations", {
...input.body,
model: config.model,
prompt: input.prompt,
});
return (data.data || []).map((item) => item.b64_json ? \`data:image/png;base64,\${item.b64_json}\` : item.url);`,
video: `// 输入:input.prompt / input.references(dataURL[]) / input.params
// 返回:{ url } 或 { blob } 或视频 URL 字符串
const task = await http.post("/videos", {
model: config.model,
prompt: input.prompt,
seconds: input.params.seconds,
});
return await poll(
() => http.get(\`/videos/\${task.id}\`),
(state) => state.status === "completed" ? { url: state.video_url || state.url } : null,
{ intervalMs: 2500, timeoutMs: 300000 },
);`,
audio: `// 输入:input.prompt / input.params(voice/format/speed/instructions)
// 返回:Blob,或 base64/dataURL 字符串
return await http.post("/audio/speech", {
model: config.model,
input: input.prompt,
voice: input.params.voice,
response_format: input.params.format,
speed: Number(input.params.speed),
}, { responseType: "blob" });`,
text: `// 输入:input.messages([{role,content}]) / input.body
// 用 onDelta(text) 推送流式文本;返回最终完整文本
const data = await http.post("/chat/completions", {
model: config.model,
messages: input.messages,
});
const text = data.choices?.[0]?.message?.content || "";
onDelta(text);
return text;`,
};
/** Normalize whatever an image script returns into the app's generated-image shape. */
export function normalizePluginImages(result: unknown): string[] {
const items = Array.isArray(result) ? result : [result];
const urls = items
.map((item) => {
if (typeof item === "string") return item;
if (item && typeof item === "object") {
const record = item as Record<string, unknown>;
if (typeof record.dataUrl === "string") return record.dataUrl;
if (typeof record.url === "string") return record.url;
if (typeof record.b64_json === "string") return `data:image/png;base64,${record.b64_json}`;
}
return "";
})
.filter(Boolean);
if (!urls.length) throw new Error("模型调用脚本没有返回图片");
return urls;
}
+54 -2
View File
@@ -1,10 +1,12 @@
import axios from "axios";
import { nanoid } from "nanoid";
import { dataUrlToFile } from "@/lib/image-utils";
import { getMediaBlob, uploadMediaFile, type UploadedFile } from "@/services/file-storage";
import { imageToDataUrl } from "@/services/image-storage";
import { boolConfig, buildSeedancePromptText, isSeedanceVideoConfig, normalizeSeedanceDuration, normalizeSeedanceRatio, normalizeSeedanceResolution, seedanceVideoReferenceError, SEEDANCE_REFERENCE_LIMITS } from "@/lib/seedance-video";
import { buildApiUrl, modelOptionName, resolveModelRequestConfig, type AiConfig } from "@/stores/use-config-store";
import { buildApiUrl, modelOptionName, resolveModelRequestConfig, resolveModelScript, type AiConfig } from "@/stores/use-config-store";
import { runModelPlugin } from "./model-plugin";
import type { ReferenceImage } from "@/types/image";
import type { ReferenceAudio, ReferenceVideo } from "@/types/media";
@@ -23,9 +25,12 @@ type ApiEnvelope<T> = T | { code?: number | string; data?: T | null; msg?: strin
type RequestOptions = { signal?: AbortSignal };
export type VideoGenerationResult = { blob?: Blob; url?: string; mimeType?: string };
export type VideoGenerationTask = { id: string; provider: "openai" | "seedance"; model: string };
export type VideoGenerationTask = { id: string; provider: "openai" | "seedance" | "plugin"; model: string };
export type VideoGenerationTaskState = { status: "pending" } | { status: "completed"; result: VideoGenerationResult } | { status: "failed"; error: string };
/** Results for scripted (plugin) video models, which run their own create+poll in one shot at task creation. */
const pluginVideoResults = new Map<string, VideoGenerationResult>();
function aiApiUrl(config: AiConfig, path: string) {
return buildApiUrl(config.baseUrl, path);
}
@@ -54,6 +59,8 @@ export async function requestVideoGeneration(config: AiConfig, prompt: string, r
export async function createVideoGenerationTask(config: AiConfig, prompt: string, references: ReferenceImage[] = [], videoReferences: ReferenceVideo[] = [], audioReferences: ReferenceAudio[] = [], options?: RequestOptions): Promise<VideoGenerationTask> {
const selectedModel = (config.model || config.videoModel).trim();
const requestConfig = resolveModelRequestConfig(config, selectedModel);
const script = resolveModelScript(config, selectedModel);
if (script) return createPluginVideoTask(requestConfig, selectedModel, script, prompt, references, options);
assertVideoConfig(requestConfig, requestConfig.model);
if (isSeedanceVideoConfig(requestConfig)) {
return createSeedanceTask(requestConfig, selectedModel, prompt, references, videoReferences, audioReferences, options);
@@ -65,11 +72,56 @@ export async function createVideoGenerationTask(config: AiConfig, prompt: string
}
export async function pollVideoGenerationTask(config: AiConfig, task: VideoGenerationTask, options?: RequestOptions): Promise<VideoGenerationTaskState> {
if (task.provider === "plugin") {
const result = pluginVideoResults.get(task.id);
return result ? { status: "completed", result } : { status: "failed", error: "插件视频任务已失效,请重新生成" };
}
const requestConfig = resolveModelRequestConfig(config, task.model);
assertVideoConfig(requestConfig, requestConfig.model);
return task.provider === "seedance" ? pollSeedanceTask(requestConfig, task, options) : pollOpenAIVideoTask(requestConfig, task, options);
}
async function createPluginVideoTask(config: AiConfig, model: string, script: string, prompt: string, references: ReferenceImage[], options?: RequestOptions): Promise<VideoGenerationTask> {
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");
const refs = await Promise.all(references.map((image) => imageToDataUrl(image)));
const result = videoPluginResult(
await runModelPlugin({
capability: "video",
script,
config,
input: {
prompt,
references: refs,
params: {
seconds: normalizeVideoSeconds(config.videoSeconds),
size: normalizeVideoSize(config.size),
resolution: normalizeVideoResolution(config.vquality),
ratio: config.size,
generateAudio: boolConfig(config.videoGenerateAudio, true),
watermark: boolConfig(config.videoWatermark, false),
},
},
signal: options?.signal,
}),
);
const id = nanoid();
pluginVideoResults.set(id, result);
return { id, provider: "plugin", model };
}
function videoPluginResult(result: unknown): VideoGenerationResult {
if (result instanceof Blob) return { blob: result };
if (typeof result === "string") return { url: result, mimeType: "video/mp4" };
if (result && typeof result === "object") {
const record = result as Record<string, unknown>;
if (record.blob instanceof Blob) return { blob: record.blob };
const url = [record.url, record.video_url, record.result_url].find((value) => typeof value === "string" && value) as string | undefined;
if (url) return { url, mimeType: "video/mp4" };
}
throw new Error("模型调用脚本没有返回视频");
}
export async function storeGeneratedVideo(result: VideoGenerationResult): Promise<UploadedFile> {
if (result.blob) return uploadMediaFile(result.blob, "video");
if (result.url) {
+66 -70
View File
@@ -4,6 +4,13 @@ import { persist } from "zustand/middleware";
import { nanoid } from "nanoid";
export type ApiCallFormat = "openai" | "gemini";
export type ModelCapability = "image" | "video" | "text" | "audio";
export type ChannelModel = {
name: string;
capability: ModelCapability;
script?: string;
};
export type ModelChannel = {
id: string;
@@ -11,7 +18,7 @@ export type ModelChannel = {
baseUrl: string;
apiKey: string;
apiFormat: ApiCallFormat;
models: string[];
models: ChannelModel[];
};
export type AiConfig = {
@@ -35,10 +42,6 @@ export type AiConfig = {
videoWatermark: string;
systemPrompt: string;
models: string[];
imageModels: string[];
videoModels: string[];
textModels: string[];
audioModels: string[];
quality: string;
size: string;
count: string;
@@ -52,10 +55,9 @@ export type WebdavSyncConfig = {
directory: string;
lastSyncedAt: string;
};
export type ConfigTabKey = "channels" | "models" | "preferences" | "webdav";
export type ConfigTabKey = "channels" | "preferences" | "webdav";
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";
@@ -72,7 +74,12 @@ export const defaultConfig: AiConfig = {
baseUrl: OPENAI_BASE_URL,
apiKey: "",
apiFormat: "openai",
models: ["gpt-image-2", "grok-imagine-video", "gpt-5.5", "gpt-4o-mini-tts"],
models: [
{ name: "gpt-image-2", capability: "image" },
{ name: "grok-imagine-video", capability: "video" },
{ name: "gpt-5.5", capability: "text" },
{ name: "gpt-4o-mini-tts", capability: "audio" },
],
},
],
model: "default::gpt-image-2",
@@ -90,10 +97,6 @@ export const defaultConfig: AiConfig = {
videoWatermark: "false",
systemPrompt: "",
models: ["default::gpt-image-2", "default::grok-imagine-video", "default::gpt-5.5", "default::gpt-4o-mini-tts"],
imageModels: ["default::gpt-image-2"],
videoModels: ["default::grok-imagine-video"],
textModels: ["default::gpt-5.5"],
audioModels: ["default::gpt-4o-mini-tts"],
quality: "auto",
size: "1:1",
count: "1",
@@ -122,44 +125,44 @@ type ConfigStore = {
clearPromptContinue: () => void;
};
function isVideoModelName(model: string) {
const value = modelOptionName(model).toLowerCase();
return value.includes("seedance") || value.includes("video") || value.includes("sora") || value.includes("veo") || value.includes("kling") || value.includes("wan") || value.includes("hailuo");
const VIDEO_KEYWORDS = ["seedance", "video", "sora", "veo", "kling", "wan", "hailuo"];
const AUDIO_KEYWORDS = ["audio", "tts", "speech", "voice", "music", "sound"];
const IMAGE_KEYWORDS = ["seedream", "gpt-image", "image", "dall-e", "dalle", "imagen", "flux", "sdxl", "stable-diffusion", "midjourney"];
/** Best-effort default capability for a freshly fetched model name; user can override in the channel editor. */
export function guessCapability(name: string): ModelCapability {
const value = name.toLowerCase();
if (VIDEO_KEYWORDS.some((keyword) => value.includes(keyword))) return "video";
if (AUDIO_KEYWORDS.some((keyword) => value.includes(keyword))) return "audio";
if (IMAGE_KEYWORDS.some((keyword) => value.includes(keyword))) return "image";
return "text";
}
function isImageModelName(model: string) {
const value = modelOptionName(model).toLowerCase();
return !isVideoModelName(model) && !isAudioModelName(model) && (value.includes("seedream") || value.includes("gpt-image") || value.includes("image") || value.includes("dall-e") || value.includes("dalle") || value.includes("imagen") || value.includes("flux") || value.includes("sdxl") || value.includes("stable-diffusion") || value.includes("midjourney"));
function findChannelModel(config: AiConfig, value: string): { channel: ModelChannel; model: ChannelModel } | null {
const decoded = decodeChannelModel(value);
const name = decoded?.model || value;
const channel = decoded ? config.channels.find((item) => item.id === decoded.channelId) : config.channels.find((item) => item.models.some((model) => model.name === name));
const model = channel?.models.find((item) => item.name === name);
return channel && model ? { channel, model } : null;
}
function isAudioModelName(model: string) {
const value = modelOptionName(model).toLowerCase();
return value.includes("audio") || value.includes("tts") || value.includes("speech") || value.includes("voice") || value.includes("music") || value.includes("sound");
export function modelCapabilityOf(config: AiConfig, value: string): ModelCapability | undefined {
return findChannelModel(config, value)?.model.capability;
}
function isTextModelName(model: string) {
return !isImageModelName(model) && !isVideoModelName(model) && !isAudioModelName(model);
}
export function modelMatchesCapability(model: string, capability?: ModelCapability) {
export function modelMatchesCapability(config: AiConfig, value: string, capability?: ModelCapability) {
if (!capability) return true;
if (capability === "image") return isImageModelName(model);
if (capability === "video") return isVideoModelName(model);
if (capability === "audio") return isAudioModelName(model);
return isTextModelName(model);
}
export function filterModelsByCapability(models: string[], capability?: ModelCapability) {
return capability ? models.filter((model) => modelMatchesCapability(model, capability)) : models;
return modelCapabilityOf(config, value) === capability;
}
export function selectableModelsByCapability(config: AiConfig, capability?: ModelCapability) {
if (!capability) return config.models;
return config[modelListKey(capability)];
return config.channels.flatMap((channel) => channel.models.filter((model) => model.capability === capability).map((model) => encodeChannelModel(channel.id, model.name)));
}
function modelListKey(capability: ModelCapability) {
return `${capability}Models` as "imageModels" | "videoModels" | "textModels" | "audioModels";
/** The user script (if any) attached to a model; empty string means use the system default call. */
export function resolveModelScript(config: AiConfig, value: string) {
return findChannelModel(config, value)?.model.script?.trim() || "";
}
function isAiConfigReady(config: AiConfig, model: string) {
@@ -215,7 +218,7 @@ export const useConfigStore = create<ConfigStore>()(
channels,
models,
imageModel: normalizeModelOptionValue(config.imageModel || config.model, channels),
videoModel: normalizeModelOptionValue(config.videoModel || "grok-imagine-video", channels),
videoModel: normalizeModelOptionValue(config.videoModel, channels),
textModel: normalizeModelOptionValue(config.textModel || config.model, channels),
audioModel: normalizeModelOptionValue(config.audioModel || defaultConfig.audioModel, channels),
audioVoice: config.audioVoice || defaultConfig.audioVoice,
@@ -227,10 +230,6 @@ export const useConfigStore = create<ConfigStore>()(
videoGenerateAudio: config.videoGenerateAudio || "true",
videoWatermark: config.videoWatermark || "false",
canvasImageCount: config.canvasImageCount || "3",
imageModels: Array.isArray(persistedConfig.imageModels) ? normalizeModelList(config.imageModels, channels) : filterModelsByCapability(models, "image"),
videoModels: Array.isArray(persistedConfig.videoModels) ? normalizeModelList(config.videoModels, channels) : filterModelsByCapability(models, "video"),
textModels: Array.isArray(persistedConfig.textModels) ? normalizeModelList(config.textModels, channels) : filterModelsByCapability(models, "text"),
audioModels: Array.isArray(persistedConfig.audioModels) ? normalizeModelList(config.audioModels, channels) : filterModelsByCapability(models, "audio"),
},
};
},
@@ -238,18 +237,26 @@ export const useConfigStore = create<ConfigStore>()(
),
);
function normalizeModelList(models: string[], channels: ModelChannel[]) {
const allModelOptions = channels.flatMap((channel) => channel.models.map((model) => encodeChannelModel(channel.id, model)));
return Array.from(new Set((models || []).map((model) => model.trim()).filter(Boolean)))
.map((model) => normalizeModelOptionValue(model, channels))
.filter((model) => !allModelOptions.length || allModelOptions.includes(model) || !isChannelModelValue(model));
}
export function useEffectiveConfig() {
const config = useConfigStore((state) => state.config);
return useMemo(() => ({ ...config, channelMode: "local" as const }), [config]);
}
/** Normalize a mixed list of raw model names or model objects into deduped ChannelModel entries. */
export function normalizeChannelModels(models: Array<string | ChannelModel> | undefined): ChannelModel[] {
const seen = new Set<string>();
const result: ChannelModel[] = [];
for (const item of models || []) {
const name = (typeof item === "string" ? item : item?.name || "").trim();
if (!name || seen.has(name)) continue;
seen.add(name);
const capability = typeof item === "string" ? guessCapability(name) : item.capability || guessCapability(name);
const script = typeof item === "string" ? undefined : item.script?.trim() || undefined;
result.push({ name, capability, script });
}
return result;
}
export function createModelChannel(channel?: Partial<ModelChannel>): ModelChannel {
const apiFormat = normalizeApiFormat(channel?.apiFormat);
return {
@@ -258,7 +265,7 @@ export function createModelChannel(channel?: Partial<ModelChannel>): ModelChanne
baseUrl: channel?.baseUrl?.trim() || defaultBaseUrlForApiFormat(apiFormat),
apiKey: channel?.apiKey || "",
apiFormat,
models: uniqueRawModels(channel?.models || []),
models: normalizeChannelModels(channel?.models),
};
}
@@ -288,7 +295,7 @@ export function modelOptionLabel(config: AiConfig, value: string) {
}
export function modelOptionsFromChannels(channels: ModelChannel[]) {
return uniqueModelOptions(channels.flatMap((channel) => channel.models.map((model) => encodeChannelModel(channel.id, model))));
return uniqueModelOptions(channels.flatMap((channel) => channel.models.map((model) => encodeChannelModel(channel.id, model.name))));
}
export function normalizeModelOptionValue(value: string | undefined, channels: ModelChannel[]) {
@@ -297,17 +304,17 @@ export function normalizeModelOptionValue(value: string | undefined, channels: M
const decoded = decodeChannelModel(model);
if (decoded) {
const channel = channels.find((item) => item.id === decoded.channelId);
return channel && channel.models.includes(decoded.model) ? model : "";
return channel && channel.models.some((item) => item.name === decoded.model) ? model : "";
}
const channel = channels.find((item) => item.models.includes(model)) || channels[0];
return channel && channel.models.includes(model) ? encodeChannelModel(channel.id, model) : model;
const channel = channels.find((item) => item.models.some((entry) => entry.name === model)) || channels[0];
return channel && channel.models.some((item) => item.name === model) ? encodeChannelModel(channel.id, model) : model;
}
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, apiFormat: config.apiFormat, models: config.models.map(modelOptionName) });
const matched = decoded ? config.channels.find((channel) => channel.id === decoded.channelId) : config.channels.find((channel) => channel.models.some((item) => item.name === model));
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).map((name) => ({ name, capability: guessCapability(name) })) });
}
export function resolveModelRequestConfig(config: AiConfig, value: string) {
@@ -328,7 +335,7 @@ function normalizeChannels(config: AiConfig) {
...channel,
id: channel.id || (index === 0 ? "default" : `channel-${index + 1}`),
name: channel.name || (index === 0 ? "默认渠道" : `渠道 ${index + 1}`),
models: uniqueRawModels(channel.models || []),
models: normalizeChannelModels(channel.models),
}),
);
if (!channels.length) {
@@ -339,18 +346,11 @@ function normalizeChannels(config: AiConfig) {
baseUrl: config.baseUrl || defaultConfig.baseUrl,
apiKey: config.apiKey || "",
apiFormat: config.apiFormat || defaultConfig.apiFormat,
models: uniqueRawModels([
...(config.models || []),
config.model,
config.imageModel,
config.videoModel,
config.textModel,
config.audioModel,
]),
models: normalizeChannelModels([config.model, config.imageModel, config.videoModel, config.textModel, config.audioModel].map(modelOptionName)),
}),
);
}
return channels.map((channel) => ({ ...channel, models: uniqueRawModels(channel.models) }));
return channels;
}
export function defaultBaseUrlForApiFormat(apiFormat: ApiCallFormat) {
@@ -361,10 +361,6 @@ 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)));
}
function uniqueModelOptions(models: string[]) {
return Array.from(new Set((models || []).map((model) => model.trim()).filter(Boolean)));
}