Guides
How to get ChatGPT to write a Blender add-on that actually installs
How to get ChatGPT to write a Blender add-on that installs: prompts, the extension manifest, debugging errors, and where AI-written add-ons fail.
6 min read

Quick answer
Can ChatGPT write a Blender add-on?
Yes, for small tools: a panel, a few buttons, operators that rename, tidy or export. It works best if you start from a script that already runs, tell it your Blender version, and ask for the extension format with a blender_manifest.toml (Blender 4.2 and later). Large add-ons with custom UI and state still need a developer.
Why most ChatGPT add-ons fail on the first try
Ask ChatGPT for "a Blender add-on that does X" in one go and you'll often get code that looks right and doesn't install. Usually one of three things is wrong:
- The wrong packaging for your Blender version. Since Blender 4.2, add-ons are distributed as extensions, described by a
blender_manifest.tomlfile. Older add-ons used abl_infodictionary at the top of the Python file. Code trained on years of tutorials mixes the two. - Operators used out of context. Many Blender operators only work in a particular editor or mode. A script that works when you click Run in the Text Editor can fail when the same code runs from a button.
- Too much at once. A 400-line first draft has a dozen small bugs, and you can't tell which one broke the install.
The fix is a method, not a better prompt. Build it in three stages.
Stage 1: get a plain script working
Before anything about add-ons, get the behaviour right in the Scripting workspace.
- Tell ChatGPT: "Blender 5.3. Write a bpy script that, for every selected mesh, applies rotation and scale, sets the origin to the bottom centre, and renames it to
Prop_plus a two-digit number sorted by X position." - Paste it into the Text Editor and press Run Script.
- If it fails, open the system console (Window ‣ Toggle System Console on Windows; start Blender from a terminal on macOS and Linux) and paste the whole traceback back.
- Repeat until it runs cleanly on a test scene. Save that version.
Only now does it become an add-on. You've separated "does the logic work?" from "does the packaging work?", which halves the debugging.
Tip: do the action once by hand and copy the line Blender's Info editor logs. Pasting that real operator call into ChatGPT ("make this run on every selected object") is often more reliable than describing the action in words.
Stage 2: wrap it as an operator and a panel
Next prompt, with the working script pasted in:
"Turn this working script into a Blender add-on for Blender 5.3. Put the logic in an Operator class with a bl_idname of object.tidy_props. Add a Panel in the 3D Viewport's N sidebar, in a tab called Tidy, with one button that runs the operator. Include register() and unregister(). Use the extension format with a blender_manifest.toml. Show the manifest and the __init__.py separately."
What to check in the answer:
- The panel's
bl_space_typeisVIEW_3D, itsbl_region_typeisUI, andbl_categoryis the tab name. Get those wrong and the panel quietly doesn't appear. - The operator returns
{'FINISHED'}, and defines apollmethod if it needs selected objects. register()registers every class andunregister()removes them in reverse order.- The manifest has an
id,version,name,maintainer,type = "add-on", a minimum Blender version, and a licence. Compare it with the extensions page of the Blender manual for your version rather than trusting the model's memory.

Stage 3: install, break, fix
- Put
__init__.pyandblender_manifest.tomlin one folder and zip it, with the files at the top level of the zip, not inside an extra folder. - In Blender: Edit ‣ Preferences, open the extensions/add-ons section, use the drop-down menu to Install from Disk, and pick the zip.
- Enable it, press N in the 3D Viewport and look for your tab.
- Press the button on a test scene.
When it fails, paste the exact error and say when it happened: on install, on enable, or on button press. That one sentence of context saves several rounds of guessing.
Prompts that produce good add-ons
Small, clear tools work best. Some that work well with ChatGPT (and equally with Claude):
- "A panel with a button that exports each selected object as its own GLB into an
exportsfolder next to the .blend, with a checkbox for Apply Modifiers." - "An operator that finds every material whose image texture file is missing on disk and lists them in the Info log."
- "A button that adds a Bevel (2 cm, 3 segments) and a Weighted Normal modifier to every selected mesh, skipping objects that already have them."
- "A panel showing the triangle count of the selected objects and turning red above 5,000."
- "A one-click setup for 3D printing: apply scale, check for non-manifold edges with Select All by Trait, and report the count."
Avoid, at least for a first add-on: custom drawing in the viewport, modal tools that track the mouse, add-ons that keep state across sessions, or anything that downloads from the internet.
Use ChatGPT to review, not just write
Once it works, paste the whole add-on back and ask:
- "Review this for Blender 5.3 API issues and anything deprecated."
- "What happens if nothing is selected? If the .blend is unsaved? If an object isn't a mesh?"
- "Add docstrings and a short README."
The edge-case question alone usually turns up two or three crashes you'd have hit later.
Want it to drive Blender instead?
An add-on is a fixed tool you click. If what you really want is to talk to Blender ("tidy these, then export"), that's MCP, not an add-on. With a Blender MCP server, Codex or Claude runs Python in your open Blender. With ChatGPT on the web, you need a workspace plan with full MCP and a tunnel, because Blender is local. Comparison of the routes: can ChatGPT control Blender?. If you're choosing an assistant for this, see ChatGPT vs Claude for Blender.
What AI-written add-ons can't do (well)
- Big, stateful tools. Rigging systems, node-based UIs or anything with many interacting panels needs real architecture. AI drafts become hard to maintain quickly.
- Guarantee API accuracy. Blender's Python API changes between versions. Always state your version, and check unfamiliar calls against the API docs.
- Make a 3D model appear. An add-on can only build from primitives, modifiers and geometry nodes. Detailed meshes come from modelling or a generator.
- Be safe by default. You're installing code. Read what it does, especially anything touching files, the network or
oscalls.
Pair your add-on with real models
Many useful add-ons are about tidying and exporting assets that already exist. Vi3W makes those assets: text, sketch or image to 3D, then Add bones, Make it move, New paint job and Clean shape, with GLB, FBX, OBJ or STL ready to import. Try Vi3W.
Frequently asked questions
chatgpt blender addon
There's no official ChatGPT add-on inside Blender. You can have ChatGPT write your own add-on. To have it drive Blender live instead, use a Blender MCP server with Codex, or with ChatGPT in a workspace that supports full MCP.
how to add addon blender
Zip the add-on folder, then in Blender open Edit, Preferences, Add-ons (Get Extensions in recent versions), use the drop-down to Install from Disk, pick the zip, and tick the add-on to enable it.
blender addon not showing up
Usually it's the wrong format for your version (bl_info only versus a blender_manifest.toml) or a zip with an extra folder level. Open the system console to read the import error, paste it back to ChatGPT, and check the panel's category tab in the N sidebar.
Last reviewed

