Guide 05 · 7 min read
AI connectors
Let Claude, ChatGPT, Grok or a local model build rooms with you.
VTube Pro can work with an AI of your choice to create things, for example a new 3D room from a sentence like "a cosy neon ramen bar with 4 stools". The AI plans it, Blender builds it on your PC, VTube Pro checks it, and you install it with a click.
Everything starts on one page: Settings → AI connectors.
Which way should I connect?
| You use | Do this | It can |
|---|---|---|
| Claude Desktop | Click Add to Claude Desktop, then quit and reopen Claude Desktop. | Use every tool. Risky ones wait for your Approve click. |
| Claude Code or Grok Build | Copy the one-line command from the page. | Use every tool, same approvals. |
| A local model (Ollama, LM Studio) or an API key (OpenAI, xAI, Anthropic, OpenRouter) | Use Create with AI right on the page. | Use every tool, same approvals. |
| ChatGPT or grok.com | Use the Instruction pack (no setup), or the optional Remote connector. | Write a creation package you paste back. |
| Any other chat AI | Use the Instruction pack. | Write a creation package. |
VTube Pro must be running for any of these.
You stay in control
Every action an AI can take has a permission level:
- Always fine: reading guides and your avatar list, rendering previews, checking a package.
- Asks first: changing your avatar library.
- Always asks, every time: anything that changes what's on your stream, and anything that runs Blender scripts.
When an AI asks, the AI connectors page shows exactly what it wants to do, with Approve once, Approve for 30 min and Deny. If you don't answer within 5 minutes, it's denied.
The Instruction pack (works with any AI)
- Click Copy instruction pack.
- Paste it into any AI chat together with your idea.
- The AI answers with one block of text: a creation package.
- Paste it back into the page, click Validate, fix anything it reports, and click Build + install.
- Open Scene → Room to see your new room.
Fully local, no cloud, no keys
You can do all of this without any online service:
- Install Ollama and pull a Qwen model that fits your graphics card (see below).
- Optionally install ComfyUI with a free SDXL model to paint textures.
- Install Blender (the page has a one-click install button).
- Press Re-check on the AI connectors page, pick your model in Create with AI, and describe your room.
| Model | Graphics memory |
|---|---|
| qwen2.5:7b | 6–8 GB |
| qwen2.5:14b (recommended) | 10–12 GB |
| qwen3:30b | 18–20 GB |
| qwen2.5:32b | 22–24 GB |
Leave 2–3 GB free for VTube Pro's own tracking, and don't run a big model and SDXL together on an 8–12 GB card.
Optional: the Remote connector
ChatGPT and grok.com can only talk to a public web address. If you want them connected live, switch on Remote connector and run a tunnel (Cloudflare Tunnel or ngrok). Only safe tools are exposed remotely: reading, rendering, checking, and sending packages to your inbox. It's off by default. Keep the address private and switch it off when you're done.
Keys and privacy
API keys stay in a settings file on your PC and are sent only to the provider they belong to. If you use a paid AI service, you pay that provider directly for what you use. Local models cost nothing.