本文由 AI 分析生成
Summary
OpenBlender is a paid Blender addon (€12.99+) by CRT (pgcrt) that integrates generative AI workflows directly into Blender without leaving the application. It supports image-to-image, image-to-video, image-to-3D model generation, image remixing, LoRA support, HDRI generation from text, AI chat, and MCP server control — all backed by ComfyUI running locally with models like Zimage Turbo, Flux Klein, and LTX2.
OpenBlender 是整合生成式 AI 至 Blender 的付費外掛，支援圖像生成、影片生成、3D 模型生成、HDRI 生成、AI 聊天以及 MCP Server 遠端控制，底層使用本地 ComfyUI 執行推理。
Key Points
- 8 core functions: IMG to IMG, IMG to VID (3 keyframe interpolation), IMG to 3D (Trellis 2), Remix, LoRA support, TXT to HDRI (360° panorama), AI Chat, MCP Server
- Non-blocking design: all inference runs in background threads so Blender viewport remains responsive
- Run-on-Change mode: viewport changes automatically trigger regeneration — real-time AI-assisted iteration
- Custom ComfyUI workflow support: any ComfyUI JSON workflow can be loaded and executed from within Blender
- MCP Server (OpenClaw): exposes HTTP/SSE endpoints on port 9876 so Claude Desktop, Cursor, or other MCP clients can control Blender remotely — scene manipulation, material editing, animation, rendering
- Hardware requirement: RTX 4090 or equivalent recommended for simultaneous EEVEE rendering + local AI inference
- Chat backends: OpenRouter (Claude 3.5 Sonnet, GPT-4, Llama 3) or LM Studio (local models)
Insights
The MCP server integration is the most architecturally interesting feature. It converts Blender into an MCP-compliant tool server, enabling LLM agents to “see” and manipulate 3D scenes through structured API calls rather than code generation. This is a different paradigm from “generate Python scripts for Blender” — instead, the agent issues high-level commands directly to the scene graph.
The Run-on-Change mode blurs the line between traditional 3D modeling and AI-guided generation: the artist blocks out geometry and lighting, then AI continuously regenerates the visual result. The feedback loop is tighter than text-to-image prompting.
The RTX 4090 requirement highlights a real constraint: local AI-in-creative-tools requires high-VRAM GPUs that most artists don’t have. Cloud inference alternatives (via OpenRouter) exist for chat but not for image/video generation in this addon.
Connections
Raw Excerpt
外部のAIアシスタントやプログラムが、リモートでBlenderを直接制御できるようにするMCP Serverにも対応しています。Claude DesktopやCursorなどのMCP対応ツールと繋ぐことで、「チャットで指示して自動でシーンを構築させる」といった次世代のワークフローを実現します。