I tried Claude and Chatgpt for Blender. My aim is to produce 100% photorealistic still images with Blender + Photoshop. In order to do this I asked Chatgpt to give me a script that will work inside the open File-01.blend that will: 1) Add a medium res cylinder C 2) Add new empty GN to C. 3) Plug the “Muscle Generator” GN on the empty GN 4) Add some lights with specifications and constraints. 5) Add reflectors with specifications and constraints. Here I do my things. Then I asked a second script that will: 1) Save File-01 2) Save it again as File-02 3) Clean up all the unused data blocks and so on from the Blender Files tab. 4) Apply all the modifiers 5) Import the material “Skin 2026-06-30” Here I do my things. Then I asked to write a 3rd script that should have: 1) Save File-02 2) Save it as File-03 3) Apply all modifiers and clean up. 4) Import the “Create Body Hair” node 5) Import the collection “Body hair” 6) Create an empty mesh and by some means add the hair so that they are on the object but are separated form the object. 7) Create a hair lamp that will light only the hair. 8) Import an apply the material “Hair Material” on the hair. It is done by the node Create Body Hair. Now you will notice it is something I could do by myself and I did. Chatgpt made an astonishing number of blatant mistakes at almost every step. In the end I got a “Your initial intuition (intuition???) was better”. I must also say I had to come with all the ideas to build the nodes and for the pipeline’s architecture. Then I asked Chatgpt to make a seamless texture out of an already good skin texture. I got a black image. My take on all this is that Chatgpt is good for simple tasks, very good to come up with highbrow bullshit but very poor to implement it. In short, waste of time and not enticing to spending money on it. Claude on the other end is more reliable, doesn’t give me suggestions that it is not able to implement and produced very very good seamless textures from the same images than Chatgpt.

Hey everyone! 👋 I just created a video showing how to set up free coding AI models in VS Code using OpenRouter. Video can be found here: https://www.skool.com/cgpython/classroom/3eeec317?md=63db098b9fbf42fba6441a24e17e1c99 In the video, I walk through: - Creating an OpenRouter account - Generating an API key - Signing in with GitHub - Connecting OpenRouter models inside VS Code - Selecting and pinning free models This is a great option if you want to experiment with AI-assisted coding in VS Code without paying for premium models right away. There are some daily limits on the free models, but they're still useful for testing, learning, and smaller coding tasks. If you have any questions while setting it up drop them here, and I'll be happy to help!

In the Blender Python for Artists course section on functions Victor points out that displaying the python console is different for Mac and Linux, there isn't the "toggle system console" menu entry provided in Windows. Instead you must start Blender from a text console and that window will display console messages as you use Blender as in the image below In kubuntu you can edit the Blender entry in the Application Launcher to start in a console by setting the "Run in terminal" option in the Advanced properties as shown below

I plan to take a preliminary 3-hour course first and then take this one, as I think this approach makes more sense for me.

You don't need to pay for anything to start coding with AI right now. Here's what you can try in the free tier on OpenRouter. 1. 𝐐𝐰𝐞𝐧𝟑 𝐂𝐨𝐝𝐞𝐫 𝟒𝟖𝟎𝐁 Use this when you want to actually code. Paste the script, paste the error, and let it help you get unstuck. 2. 𝐆𝐞𝐦𝐦𝐚 𝟒 𝟑𝟏𝐁 Use this when the problem is visual. Screenshot your Blender UI, render, node setup, or error, and ask what’s going wrong. 3. 𝐍𝐞𝐦𝐨𝐭𝐫𝐨𝐧 𝟑 𝐔𝐥𝐭𝐫𝐚 Use this when you need to read a lot before coding. Great for understanding docs or a large codebase, but I’d be careful letting it write final code. Which of these have you already used?