Every 3D Artist Should Be Using Claude Code (My Setup)

Summary

Stef argues that AI coding agents such as Claude Code and Codex can help 3D artists and other creative professionals—not just programmers—by organizing project knowledge and automating repetitive work. He presents his agent setup and demonstrates how he uses it for concept development, research, image generation, video tasks, and 3D workflows.

Setup and workflow

• Stef keeps his agent’s instructions, process notes, and accumulated knowledge in a local project folder. He launches Claude Code in that folder so it can use those files as context.
• He gives the agent a named identity, “Mr. Mark,” and built a simple local website-style interface for viewing information and generated results. Requests can produce visual cards, such as image-generation or research cards.
• An inbox folder lets him drop in files and reference them in chats. He recommends keeping the agent’s knowledge and project information in a private repository.
• He uses Cursor, but says VS Code, Claude’s desktop app, Codex, and other IDEs or CLIs can also work. He suggests starting with basic personal and project information, then documenting processes and automating them gradually.

Creative work and automation

• For video concepts, Stef can describe an idea to the agent, review generated options, give feedback, and request further iterations. He documents his preferences and workflow so the agent can follow them without repeating every instruction.
• Using a fal.ai API key and model instructions, the agent can generate images through supported services and models. Stef says generating around 60 images took about five minutes in one example. He also shows the agent producing a set of potential assets for a city environment.
• He uses agent skills and APIs for research, including YouTube analytics, X post parsing, and video transcription or analysis. He also describes using FFmpeg-related skills for tasks such as trimming time-lapse footage and removing unwanted screen-recording sections.
• Stef uses voice input with Whisper rather than typing for many prompts. He says a friend’s local, open-source Whisper setup works for him without a paid subscription.

3D software and MCP

Stef highlights the growing availability of MCP connections for creative software, mentioning Blender, Unreal Engine, and Figma. These connections can let an agent work with software rather than only provide advice in a chat window.

• For Blender, he suggests starting with practical tasks such as cleaning up a scene, applying sensible names, or creating scripts and add-ons.
• He describes using Claude Code on a commercial project to help build, adjust, and debug a Geometry Nodes setup.
• With Blender and Unreal Engine connected, an agent can help move assets and textures through a workflow—for example, baking a texture in Blender and bringing it into Unreal Engine, positioning objects, or setting up materials and shaders.
• In his tests, Stef says Codex seems better at interpreting screenshots and spatial positioning. Claude Code’s relative performance can depend on which models and capabilities are available. He notes that these tools are changing quickly.

Cost and practical advice

• Stef says the $20 Claude subscription was not enough for his use of Claude Code, while he considers the $100 tier more suitable. He says he has not hit its limits while running several research and image-generation tasks. He also mentions using higher-tier subscriptions while working with a team.
• Image generation through fal.ai is an additional, usage-based cost. Stef reports spending about $68 that month and suggests considering alternatives with unlimited image generation when producing large volumes.
• He recommends beginning with a small, private agent setup and expanding it with documented knowledge, skills, and MCP connections as useful tasks emerge. He estimates that his setup helps him produce more video content by reducing time spent on repetitive creative work.

Guide and demonstrations

• Agent setup guide: Create a project folder, launch an agent within it, add instructions and knowledge, and optionally build a simple visual interface.
• Concept and asset workflow: Prompt for a concept, review and refine image iterations, then generate a collection of supporting assets.
• Blender automation examples: Use MCP for scene cleanup, naming, scripting, add-ons, and Geometry Nodes assistance.
• Blender-to-Unreal workflow: Test using an agent to transfer baked textures or objects and assist with Unreal materials and placement.
• Voice prompting and video research: Use speech-to-text and agent skills for research, transcription, and selected editing tasks.

Main speaker and source

• Stef (Stefan 3D AI): The video’s presenter, sharing his Claude Code and Codex setup, experiments, cost estimates, and 3D workflow examples.