AI chat panel
@-reference project assets and the agent calls MCP tools to generate image / video / 3D / speech, edit node graphs and run workflows with live status. Multi-session history; any configured text provider.
NEWv7.3 is out · custom answers in AI chat question cards →
The agent-nativeAI art asset engine
Images, video, voice, 3D models and explorable worlds — generated, organized and reused in one desktop app. Node-graph pipelines, one-click workflows that become reusable assets, and a built-in MCP server that lets the in-app AI or Claude Code do the work. Local projects, your own keys.
30+Model providers
70+MCP tools
12+One-click templates
8Generation modalities
100%Local projects
Works with your favorite models
OpenRouterOpenAIGoogle GeminiDeepSeekxAI GrokVolcengine ArkKlingMiniMaxDashScopeZhipu GLMKimiElevenLabsComfyUIWorld LabsMeshyTripoRodinLuma AILux3DModelScopeMagicRouterNewAPIvLLMOllamaLM Studio
Features
One toolkit, from script to final cut
AI that actually does the work. Assets, shots, node graph, director stage and timeline share one local project.
MCP tool server
Claude Code / Codex connect over a stdio bridge or HTTP to plan and commit workflows, run generation and read/write assets and graphs. Stable token, audit log, concurrency gate.
MCP setup guideSkill system
Built-in storyboard, director-review and creation skills load automatically; drop in your own SKILL.md. Skills set the process, tools execute.
Manual §5.6Plugin marketplace
Workflows, skill bundles and MCP servers in one place. Script paths are listed for consent; going past the sandbox is Allow once / Reject only.
Manual §5.7Blender MCP
9 tools close the inspect → bpy → screenshot → export GLB loop, connecting outbound to the addon — no uv, no Python.
Blender toolsetNode graph + one-click workflows
Text / image / video / voice / music / SFX / 3D / spatial-world nodes. Short-drama, game UI, product ad and e-commerce templates become reusable host assets — double-click to Dive in and tweak.
Manual §73D director stage
Pose characters and cameras, AI blockout, 20+ primitives; capture blocking shots and record action clips into downstream nodes. Accepts 3D models, panoramas and Gaussian splats.
Manual §12World models
World Labs Marble turns text or reference images into walkable 3D worlds; export a mesh or splat straight to the stage.
Portrait retouch
PixCake-style retouching with five intensity steps; only the requested area changes, everything else keeps original pixels.
Final timeline
Arrange imported and generated clips, picture-in-picture and transitions, preview a selection or the whole cut, then export.
Manual §10Agent Native
One command and Claude Code
runs your production
While running, the app serves a local tool server on 127.0.0.1 with a Bearer token. External agents and the in-app chat share the same tools.
- Plan and commit workflows — natural language to a plan preview, committed as a host asset that appears in the UI
- Run generation and track tasks — image, video, speech, music, SFX, 3D and spatial worlds
- Safe by design — persistent token, audit log, concurrency gate; Ask / Plan modes narrow writes per request
$ claude mcp add aiartengine -- node ./resources/mcp-bridge.mjs ✓ Added stdio MCP server aiartengine > Make a mystery short with the short-drama template, lead from @heroine-sheet ● workflow_list_presets matched: short-drama storyboard ● workflow_plan plan preview ready ● workflow_commit host asset committed, shown in UI ● task_run running mcpTaskId=7f3a… ● task_status script ✓ beats ✓ 9-grid ✓ video …
Demo
From node wiring to a finished cut
One chain: reference image → character sheet → select image → video generation.
Workflow
Seven steps to a finished piece
Build each step by hand, or let an agent run the whole thing.
-
01
One-Click Workflow
Pick a preset (including the short-drama storyboard template) or describe what you need in one sentence, preview the topology and create a reusable host asset. See Manual §7 and the short-drama guide.
-
02
Asset library
Images, videos and voices referenced by GUID, with asset-package import and export.
-
03
Episode pipeline
Beat breakdown, 9-grid / 4-grid storyboard sheets and director review put the narrative rhythm on screen.
-
04
Node generation
Text / image / video / voice / music / SFX / spatial-world / 3D model nodes across 30+ providers, with multi-cloud object storage. Ports must match; select nodes accept list ports only.
-
05
Director stage
3D posing and cameras; capture blocking shots and record actions into downstream generation.
-
06
Final timeline
Arrange imported and node clips, preview a selection or the whole cut, export.
-
07
Agent pipeline
The short-drama template builds script → beats → 9-grid → collage → 4-grid → motion prompts → 36 videos; the Episode Pipeline window regenerates a stage and invalidates downstream consistently.
Get Started
Pick the path that fits your goal
New here
Get your first clip in 5 minutes
Install → connect a model → pick a template → get a clip, no manual node wiring.
Short drama
Run the full short-drama pipeline
Storyboards, collage, director review and motion video, with all four review nodes explained.
Reference
Complete manual
Core concepts, every screen, shortcuts and troubleshooting.
Download
Free download, ready out of the box
Installers bundle their own runtime — no Node.js needed. Add your model keys and start creating.
- Windows NSIS installer · .exe Go to Releases
- macOS .dmg · x64 for Intel, arm64 for Apple Silicon Go to Releases
- Linux AppImage · chmod +x then run Go to Releases
Pushing a v* tag builds every platform on GitHub Actions. Or run from source: npm install && npm run dev (Node.js 22+).
Community
Community & feedback
Join the QQ groups, watch the video tutorials or open an issue.
QQ groups346340389647306826
Video tutorialsBilibili space →
IssuesGitHub Issues →
Email284139554@qq.com
Ready to let AI roll camera for you?
Open source, free and local-first. Install and get your first clip in five minutes.