本地优先的 AI 创作工具（Electron + Vue 3 + TypeScript），以 DeepSeek Harness 作为应用内 AI 运行时，并内置 MCP Server 供 Claude Code / Codex 调用——把短剧、广告与成片素材生成为本地工程文件。
DSH 适配
原生运行时
作者声明
安全审计
未审计
最后核验
2026-09-02
许可证
NOASSERTION
01它能帮你完成什么？
Produce short-drama, ad and finished-film assets from a brief in the AI chat panel
Storyboards, node-graph workflows, images, video, 3D models, voice and finished timelines saved as local project files
Short-drama / ad / film creators and AI art studios using DeepSeek Harness as the in-app AI runtime
Let Claude Code / Codex and other external AI agents drive your project
Planned and persisted workflows, generation runs, and asset / node-graph read-write operations
Developers and AI agents that want to operate AI Art Engine programmatically over stdio or HTTP
02如何接入 DeepSeek Harness？
前置条件
- A desktop app install for your platform — Windows .exe, macOS .dmg (x64 / arm64) or Linux .AppImage
- A model provider API key (e.g. DeepSeek, OpenAI) — the app calls models through your own key
安装步骤
- 01
Download the installer for your platform from GitHub Releases
- 02
Install and launch, then create a new project (新建工程)
- 03
In Settings add a model provider and paste your API key; optionally configure object storage
- 04
Open the AI chat panel (left ◈ icon) and @-reference assets to generate, or follow the MCP guide to connect Claude Code
- 05
Or build from source: git clone, cd ai-art-engine, npm install, npm run dev (needs Node.js 22+)
验证接入成功
- After install, the left narrow bar shows the ◈ icon that opens the AI chat panel (点左侧窄栏「◈」打开 AI 对话)
- Settings → 模型 lets you add a model provider and paste a key (设置里添加模型提供商并填写密钥)
03DSH 适配与能力边界
Runs DeepSeek Harness as the in-app AI assistant runtime; loads dsh SKILL.md-format custom skills; exposes a built-in MCP Server for external agents (Claude Code / Codex)
In-app AI chat (DeepSeek Harness runtime)
project assets @-referenced in the chat→images / video / 3D / voice generation, node-graph edits, workflow runs and status
generated media and node-graph changes are written to the local project directoryBuilt-in MCP Server for external agents
stdio bridge or HTTP connection from Claude Code / Codex / etc.→planned & persisted workflows, generation runs, asset & node-graph read-write
external agents can read and write your project assets and node graphs over MCPNode-graph generation
text / image / video / audio / 3D nodes with model parameters→reusable host assets, storyboards, animations, layered PSD exports
Custom skills in dsh SKILL.md format
.md files matching dsh SKILL.md (frontmatter name/description + body)→auto-loaded skill instructions injected as <available_skills> in context
04适合谁？何时不该用？
适合
- Short-drama / ad / film creators and AI art studios using DeepSeek Harness as the in-app AI runtime
- Developers and AI agents that want to operate AI Art Engine programmatically over stdio or HTTP
不适合
- Local ComfyUI requires installing comfy-api-proxy (default 8189) first and must not connect directly to 8188.
05兼容性、维护与安全提示
- The app is local-first and calls models through your own API key; generation needs network access to your configured provider endpoint and an account/billing for that provider.
- Packaged installers ship for Windows (.exe), macOS (.dmg x64/arm64) and Linux (.AppImage); other platforms must build from source.
- Local ComfyUI requires installing comfy-api-proxy (default 8189) first and must not connect directly to 8188.
GPL-3.0 per README (derivative works must be open-sourced under the same license); latest release v5.0.7 (2026-08-31)
06常见问题
AI Art Engine 是如何使用 DeepSeek Harness 的？
应用内的 AI 对话面板运行在 DeepSeek Harness 运行时之上：你在对话里用 @ 引用工程资产，助手就会调用 MCP 工具生成图片、视频、3D 与语音，编辑节点图并运行工作流。
Claude Code 或其他 Agent 能驱动这个应用吗？
可以——内置的 MCP Server 让 Claude Code / Codex 等外部 Agent 通过 stdio 桥或 HTTP 连接，用来规划并落盘工作流、执行生成、读写资产与节点图。
怎么添加自己的模型提供商？
在「设置 → 模型」中添加提供商（DeepSeek、OpenAI 等）并填入 API Key；模型调用走你自己的密钥，工程本地优先、数据不出本机。
支持 dsh SKILL.md 格式的自定义技能吗？
支持——在「设置 → 自定义技能」把符合 dsh SKILL.md 格式（frontmatter 的 name/description + Markdown 正文）的 .md 文件放进技能目录，下次对话即自动生效。
如何安装？
从 GitHub Releases 下载对应平台安装包（Windows .exe、macOS .dmg、Linux .AppImage），安装后新建工程并填入模型密钥；也可从源码构建（需 Node.js 22+）。
07相关的 DSH 工作流
dsh-plugin-radar
作者 adamplatin123
DSH Plugin Radar — 开源 DSH 插件生态雷达：自动发现 15900+ 候选、k8s 运行级实测 10000+、15 分钟快照管线；插件目录是其自动生成的 artifact
picgo-core
作者 picgo
强大的图片上传引擎，同时支持 CLI 与 API。
dsh-infinite-gen-3
作者 minglink
DeepSeek 专用破甲插件「无限三代」dsh-infinite-gen-3 — armor-breaking plugin for DeepSeek，破甲版：稳定化破甲，求 Star 收藏 ⭐
helloagents
作者 hellowind777
一个自主的高级智能伙伴，不仅分析问题，更持续工作直到完成实现和验证。
08数据与来源
**AI 对话面板** — 应用内 AI 助手（DeepSeek Harness 运行时）：在对话里
**自定义技能** — 设置 → **自定义技能**：把符合 dsh SKILL.md 格式（frontmatter `name` / `description` + Markdown 正文）的 `.md` 文件放进技能目录
**MCP 工具服务** — 内置 MCP Server，Claude Code / Codex 等外部 AI Agent 可经 stdio 桥或 HTTP 直连操作你的工程
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-09-02。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。