A creator recently experimented with GPT-6 Astra together with Blender to produce a new Fabergé-style egg inspired by popular culture. The workflow began with image generation: using ChatGPT Images 2.5 and the prompt “Generate a photo of a faberge egg that's themed after the TV show Pluribus - research first.”
The resulting image was, in the author's words, a solid first attempt. They pasted that image into Codex running GPT-6 Astra (high) and instructed the agent to use its local Blender capability with the prompt: “Use your blender local skill to create a blender model of this faverge egg.”
Process and output
The Codex-driven process ran for 17 minutes 51 seconds and produced several .blend files. The author already had a "vibe-coded Blender viewing" experiment available, so they added it to their tools collection and enabled a .blend URL Viewer so others can view the Pluribus-themed Blender model in a web browser.
Key details of the workflow:
- Image generation: ChatGPT Images 2.5 with the prompt to create a Pluribus-themed Fabergé egg photo.
- Model generation: Codex using GPT-6 Astra (high), instructed to create a Blender model from the generated image.
- Processing time: 17 minutes 51 seconds.
- Output: multiple .blend files.
- Viewing: .blend URL Viewer for in-browser inspection.
Why this matters
The experiment demonstrates an automated path from AI-generated imagery to editable 3D assets by combining image generation with an LLM-based code agent that can produce native Blender files. Such a pipeline can reduce manual modeling work and quickly yield multiple, editable iterations in .blend format for further refinement.
Related tags
The project is associated with: 3d, javascript, tools, ai, generative-ai, llms, blender, coding-agents, codex, gpt-6-astra.
The author has shared their files and tools so others can try viewing and building upon the generated Blender models in the browser.
