AI for SketchUp: what each kind of tool can actually do

AI rendering vs AI modeling in SketchUp

Most of the confusion about AI in SketchUp comes from mixing up two different jobs.

Image tools take a view of your model and produce a picture. They are good at mood, materials and lighting for a client presentation. They don't change a single edge in the model, and what they draw can drift from your actual design.

Geometry tools create or change the model itself: faces, groups, components, tags, materials. The result is something you can measure, dimension and build from.

If someone says "I use AI in SketchUp", they usually mean renders. If you want help with the modeling work, you need one of the geometry tools below.

SketchUp's own AI features

Trimble ships several AI features under the name SketchUp AI. On desktop they come preinstalled as extensions. Open them from SketchUp AI, or use Extensions › AI Assistant. On Windows, AI Assistant has its own toolbar; on macOS, add its button through Customize Toolbar. If you haven't updated SketchUp, you can get AI Assistant from the Extension Warehouse.

• AI Assistant › Help: ask how a tool or workflow works and get an answer inside SketchUp.
• AI Assistant › Generate Object: make a single 3D object, with photoreal materials, from a text prompt or an image. Good for props and entourage.
• AI Render: combines the current viewport with a prompt or a preset style and produces an image. It started out as the Labs feature SketchUp Diffusion.
• Generate Textures: in the Materials panel, creates the extra PBR maps (roughness, normal and so on) for a material that lacks them.

These run on AI credits. Paid subscriptions (Go, Pro, Studio) include a monthly allowance, SketchUp Free includes none, and Help costs no credits. An add-on subscription buys more. Check help.sketchup.com for current numbers.

Strengths: nothing to set up, supported by Trimble, and some of them also run in SketchUp for web and iPad. Limits: each request makes one new thing. None of them edit the objects already in your model.

The SketchUp connector for Claude

In April 2026 Trimble also released a connector that lets Claude build SketchUp models from a conversation. You enable Trimble SketchUp in Claude's connector settings and sign in with a Trimble ID. Claude builds the model in a cloud SketchUp session and hands you a thumbnail and a .skp download link.

It needs no SketchUp install, so it works from a browser, and it handles groups, components and curved surfaces. Trimble's own help page is clear about the main limit: it creates new .skp files and cannot edit an existing one. A free SketchUp entitlement covers 30 models.

AI render tools

Beyond AI Render, there are third-party tools that do the same job with more control. Veras (made by EvolveLAB, now owned by Chaos) restyles a view of your model from a prompt and keeps your massing. Real-time renderers such as D5 Render have added AI features like style transfer and AI material matching on top of their normal rendering.

These are the right choice when the deliverable is a picture: early concept images, mood options, a quick client render. They are the wrong choice when you need the model itself to change. A beautiful render of a deck with the wrong joist spacing still has the wrong joist spacing in the .skp.

Chatbots that write Ruby for you to paste

SketchUp has a Ruby API, and anything you can do by hand, a script can do. ChatGPT, Claude and Gemini can all write that Ruby. You paste it into Extensions › Developer › Ruby Console (Window › Ruby Console before SketchUp 2021.1) and press Enter.

This is free if you already pay for a chatbot, and it's fine for short jobs: count the components, list the materials, rename a handful of groups.

The catch is that the chatbot can't see your model. It guesses at names and structure, can't run the code, and only learns about an error when you paste it back. A script that worked on one object can quietly fail on the next 200. Scripts pasted into the console also don't group their changes into one undo step unless the code does it. See using ChatGPT to write SketchUp Ruby and how to run a Ruby script for how to do this safely.

Agents that work in your open model (MCP)

MCP (Model Context Protocol) is a standard way for an AI app to use outside tools. A SketchUp MCP server is a small extension that lets an agent like Claude Code, Cursor or Codex run Ruby in your open model, read the result and try again.

That loop is the difference from pasting code. The agent looks up the real component names, runs the change, reads any error and fixes its own code. With a screenshot tool it can also look at the viewport to check what it built. It works on the model you already have, which none of the generators above do.

The options today:

• Free open-source servers on GitHub, such as mhyrr/sketchup-mcp and its more active fork zinin/sketchup-mcp2. Free and MIT licensed; setup needs a Python tool (uv) and some config editing. Features vary between projects.
• MakeIt4Me, a paid extension. We make it, so weigh what we say here accordingly; it gets its own section below.

Trimble's Claude connector is also built on MCP, but it runs in the cloud and makes new files, so it belongs with the generators. How the local servers work and how to set one up is in SketchUp MCP servers. The differences between a free server and MakeIt4Me are in this comparison.

Limits: you need SketchUp Pro or Studio on desktop (Free and Go don't run extensions), an agent app on the same computer, and a subscription with an AI provider. Prompting is also less predictable than clicking a button in a finished extension.

Which AI tool to use for what

• A client image or mood options: AI Render, or a tool like Veras or D5 if you need more control.
• One prop for the scene (a plant, a chair, a sculpture): Generate Object. To build a whole building from pictures, see modeling from a photo.
• A quick first massing, with no desktop SketchUp at hand: the Claude connector.
• "How do I..." questions: AI Assistant Help, which is free.
• A small one-off script you can read: a chatbot plus the Ruby Console.
• Changes to an existing model (replace a material everywhere, rename 300 components, tag by type, block out framing in your real file): an MCP agent.
• A repeat task you want as a button: an agent can write and test the tool; see making your own extension.

Many people end up using two: an agent or script for the modeling chores, and an AI renderer for the pictures at the end. To compare an agent with doing the same chores by hand or with pasted code, see the comparison table.

Questions