System configuration

System configuration is maintained by administrators; creators never need to touch it. The configuration lives on the server, and the workspace never shows channel keys or storage credentials.

Open the configuration dialog

• Sidebar gear: the gear at the bottom of the narrow left sidebar, visible to administrators only.
• Canvas toolbar: the configuration icon in the canvas top-right toolbar.
• Missing-model hints: when the Image, Video, or Canvas page reports no default model, an administrator clicking the hint opens the dialog directly.

The dialog is titled "配置与用户偏好" and has five tabs: 我的渠道, 我的模型, 积分设置, 邀请奖励, 对象存储. It always opens on the channels tab. Non-administrator accounts do not see the entry point, and the dialog closes itself if the signed-in role is not administrator.

How saving works: drafts first, then per-tab save

• There is no global save button. Every tab has its own "保存", greyed out when nothing changed or while a save is running.
• Adding a channel or editing a model only writes to a draft; you must click that tab's "保存" to send it to the server. Success reports "渠道配置已保存", "模型配置已保存", "积分设置已保存", and so on; failures show the concrete reason (for example "保存对象存储失败").
• Closing the dialog with unsaved changes asks first: "放弃未保存修改？" / "关闭后会丢弃所有 Tab 中未保存的修改。", with "放弃修改" and "继续编辑".
• Saved changes apply immediately for the administrator. Creators load configuration when they sign in, so they need a page refresh to see new models and prices.
• Tasks read configuration when they run (the queue only stores task ids), so new tasks use the latest settings while existing tasks keep their creation-time model snapshot.

My channels

A channel is an entry point for calling an AI service: one base URL plus one set of credentials.

Default base URLs per call format:

Fetching models: after filling in the base URL and API key, click "拉取模型". The result is written into the model list with the hint "<channel> 模型列表已更新，请点击保存"; missing inputs report "请先填写该渠道的 Base URL 和 API Key". For MiniMax and 自定义 the button is disabled ("MiniMax 请手动配置 MiniMax-H3", "自定义格式请手动配置模型列表"), so you maintain the model names yourself.

Deleting a channel: the confirmation reads "该渠道会在点击保存后从服务端删除。". If models still reference it you get "请先在“我的模型”中删除关联模型并保存，再删除渠道".

Channels have no enable/disable switch and no standalone connectivity test — the quickest way to validate a key is "拉取模型". The key is masked in the form, but the list API returns it in plain text, so avoid opening this dialog on a shared screen.

My models

Models come from your channels: first type or fetch the model names in "我的渠道", then come back here to add capabilities and pricing. With no models configured the tab warns "请先在“我的渠道”中添加模型，再配置可选模型".

1. Choose which models to configure

The three "可选项" dropdowns at the top select the models you want to configure under image, video, and text. Each model is one row showing its display name and a "默认" marker, with an "编辑" button opening the configuration dialog.

Three default selectors decide what the generation pages start with: the default image model, the default video model, and the default text model.

2. Mark capabilities

Capabilities gate the UI: only models with the matching capability appear on that generation page, and a video model without "支持音频输入" cannot reference audio on the canvas.

3. Display name and icon

• Display name: what the generation page shows; leave it empty (or identical to the real model name) to use the real name.
• Display icon: 自动匹配, 默认（Clapperboard）, Agnes, Seedance（字节跳动）, Claude, OpenAI, Gemini, DeepSeek, GLM, Grok, or MiniMax. Auto-matching guesses from the model name.

4. Billing and pricing

• Text and image models: a single "每次积分" field, charged per run.
• Video models: pick a "计费方式" first — per run or per second. Per run only needs an "积分/次" value per resolution tier; per second additionally needs "最短生成时长（秒）" (minimum 1, default 3). Prices are entered per generation mode × resolution (480P / 720P / 768P / 1080P / 2K / 4K); empty tiers show as "未配置" and users cannot pick them.
• Formulas: per run = unit price × count; per second = unit price × seconds × count. The dialog has no quote preview — quotes appear on the generation pages and the canvas.

5. Request and response templates for custom channels

Models on a channel whose call format is "自定义" need custom JSON here (request template and response parsing). Available variables: {{prompt}}, {{model}}, {{references}}, {{videoReferences}}, {{audioReferences}}, {{size}}, {{resolution}}, {{seconds}}, {{count}}, {{taskId}}; the request method can be POST or GET. Invalid JSON reports "模型自定义JSON必须是对象" or "模型自定义JSON格式不正确".

6. Common blocks

• "模型配置引用了未保存的渠道或模型，请先保存“我的渠道”": save the channels tab first, then save the models.
• Video billing requires a per-run or per-second choice, prices must be integers no smaller than 0, and only the modes you ticked and resolutions the model supports can be priced — the server rejects anything else.

Credits

A single field, 初始积分 (default 100), sets the starting balance for users registered, created in the console, or signing in through a third party afterwards. The help text reads "只影响之后注册的用户，已有用户积分不会自动变更".

To adjust an existing user's balance, use "调整积分" in the console's user management, which requires a reason and writes a ledger entry.

Note that per-task charging is not configured here — it lives on each model in "我的模型".

Invitation rewards

A single field, 每名新用户奖励积分. There is no switch; set it to 0 to pause rewards. The help text reads "新用户通过有效邀请链接完成邮箱注册或首次第三方登录后，奖励发放给邀请人。设置为 0 表示暂停发放；只影响之后注册的用户，不追溯历史邀请。"

• It is independent of 初始积分: the starting balance goes to the new user, the reward goes to the inviter.
• The invite link lives on the Personal information page under "专属邀请链接".

Object storage

Object storage holds uploads, avatars, and generated results. You can configure several entries, but one must be the default: saving without a default reports "请设置默认对象存储", and the first storage created on a site becomes the default automatically.

Shared fields

Provider differences

Switching provider clears the credentials, bucket, region, endpoint, and public base URL you had entered.

Testing and defaults

• Test upload: fill in the required fields and click "测试上传". If you have not saved yet it reports "请先保存对象存储配置，再执行测试上传" — save first, then test. Success writes a small text file and reports "对象存储测试上传成功，请点击保存记录测试时间" with an "打开测试文件" link; failures show the provider message directly, for example "腾讯云COS SecretId不能为空", "阿里云OSS Endpoint必须是HTTP或HTTPS地址", or "七牛云Kodo区域ID不受支持".
• Set as default: an inline "设为默认" button, disabled when the entry already is the default. The title gains a "默认 · " prefix and shows "上次测试 …" or "尚未测试".
• Delete: the default entry cannot be deleted ("默认对象存储不能删除，请先设置其他默认项"), and the last remaining entry reports "至少保留一个对象存储". Non-default entries are deleted physically with no "still referenced" protection, so make sure no assets depend on them first.
• An incomplete configuration blocks saving with messages such as "请完整填写访问密钥、密钥、Bucket 和 Region", "请填写阿里云OSS Endpoint", or "请填写七牛云Kodo公开访问地址".

Scope and troubleshooting

• After saving channels or models, the generation pages immediately show the new models and prices; creators see them after a page refresh.
• If the Image or Video page reports that no model is available, or uploads and saved results fail, come back here and check the default model and default object storage.
• Never commit channel keys or storage credentials, and never paste them into issues or screenshots.
• Users, skills, prompts, and announcements are managed in the Admin console; how these settings show up in the product is covered by Image creation, Video creation, and Infinite canvas.