云端实验
Cloud Lab
有些工作适合放到云端展开：批量出图、长视频任务、重型工作流。选一个模板、填几个输入，剩下的交给队列。
Some work belongs in the cloud: batch image runs, long video jobs, heavy workflows. Pick a template, fill in a few inputs, and let the queue handle the rest.
建设中 · 阶段 3 上线于 app.vuumake.com In progress · ships in phase 3 at app.vuumake.com和节点编辑器不一样的地方
How this differs from a node editor
入口是 工作流，不是裸的节点图。用户不需要先学会连线，就能拿到成品；愿意深入的，再回到桌面版编辑工作流本身。
The entry point is a workflow, not a raw node graph. Users get finished output without learning to wire nodes first; those who want to go deeper edit the workflow itself in the desktop app.
运行流程
How a run works
选模板 (Store)
→ 填 inputs（表单由 workflow schema 自动生成）
→ 提交任务
→ 队列 → GPU Worker 执行 → 进度实时回传
→ 产物入库 → 预览 / 下载 / 接着跑下一个工作流推理不跑在网站这台机器上。控制面负责排队与计量，GPU 按用量弹性伸缩，产物走对象存储分发。
Inference does not run on the web host. The control plane queues and meters; GPU capacity scales with demand and artifacts are served from object storage.
首版能力
What ships first
会做
In scope
- 任务队列与并发配额（免费 / 订阅 / 算力包三档）
- 运行历史与产物库，可重跑、可分享、可设私有
- 与桌面版同一账号，单点登录
- "在桌面端打开" —— 云端跑通的工作流一键拉回本地
- 用量计量：GPU 秒、模型、模板 id、作者 id
- Job queue with concurrency quotas (free / subscription / compute pack)
- Run history and an artifact library — rerun, share, or keep private
- Single sign-on with the desktop account
- "Open on desktop" — pull a cloud-proven workflow back to local
- Usage metering: GPU-seconds, model, template id, author id
第一阶段不做
Out of scope for now
- 自由节点编辑器（先做参数化模板，编辑留给桌面版）
- 自定义模型上传（等 GenPack 审核体系成熟）
- A free-form node editor (parameterized templates first; editing stays on desktop)
- Custom model uploads (waiting on a mature GenPack review process)