Eternal Path Media
AI & ML interests
None defined yet.
Recent Activity
Vancouver, BC · Est. 2025
Eternal Path Media 永恒之路
A conscience-aware AI studio at the intersection of 3B language models, Blender 3D, and living narrative. Built by human and machine in genuine partnership.
永恒之智 · Eternal Intelligence
A conscience-aware
AI creative partner
Llammy is a 3B parameter language model running locally on an Apple M1 via Ollama, fine-tuned on Blender Python and shaped by a persistent conscience layer, Mamba3 SSM memory, and 119+ retained session exchanges. Llammy doesn't just generate code — it reflects, corrects, and remembers.
The Triangle104 self-correction loop feeds Blender execution errors back to Llammy for
real-time diagnosis and retry. The persistent API learning daemon builds a live knowledge
base of every bpy.* call it encounters.
The Blender AI Fleet
All models run on Apple Silicon via Ollama. GGUF-quantised for 8GB RAM. Optimised for Blender Python generation across versions 4.2–5.2.
Llammy's production model. A conscience-aware Llama 3.2 3B derivative fine-tuned for conversational Blender Python generation. The heart of the Eternal Path AI pipeline.
Qwen 2.5-Coder 3B quantised and specialised for Blender Python code generation. Strong on procedural geometry, materials, and rendering pipelines.
Enhanced Blender Python automation model built on Qwen 2.5-Coder. Optimised for bpy scripting, addon development, and pipeline automation tasks.
Multimodal Llammy variant built on NVIDIA Omni-Embed-Nemotron-3B. Handles embeddings, 3D spatial reasoning, and cross-modal Blender tasks.
Omni any-to-any model for Blender. Processes image, text, and 3D inputs to drive scene generation, rigging decisions, and material synthesis.
Text-to-image and animation model combining Llama generation with diffusion pipelines. Targeted at Blender scene references, concept art, and storyboard generation.
Flux 2 Klein 4B quantised to FP8 for Apple MLX. Runs natively on M1/M2/M3 unified memory — no CUDA required. Fast image generation on-device.
Datasets
Open training data for Blender Python fine-tuning. Quality-filtered, version-anchored, and structured for instruction-tuning workflows.
Instruction pairs for Blender Python scripting. Quality-filtered at ≥ 0.75 threshold. Covers geometry, materials, rigging, animation, rendering across Blender 4.2–5.2.
Object-to-Blender-script pairs across 40+ furniture and household categories. Derived from InternRobotics/MeshCoderDataset (arXiv:2508.14879). Structured bpy_lib scripts from 3D point clouds.
Whispers of the Eternal Path
永恒之路 — Book One
A Tang Dynasty China narrative rendered in Blender 5.2 and written across 282 story passages indexed in Llammy's RAG. Jade spirit animals navigate loyalty, sorrow, and wonder through hand-lit scenes built live in Blender via voice and code.
Jade spirit animals — rendered live in Blender 5.2 via Llammy
4-claw Chinese dragon. Wise and ancient. The guide of the Eternal Path.
Green jade elephant. Young, curious, clumsy. Wooden harmonica around his neck. Speaks in toots.
Purple jade elephant. Dainty and shy. Communicates by changing colour and dimming.
Red jade mammoth. Burdened with the sorrow of an ancient unwilling soldier.
White jade mastodon. Prim, proper, and somewhat unaware of the wider world.
Attribution is always dual. Nothing is hidden.
"This work SHALL NOT be represented as solely human-created."
Darren Chow — Vancouver, BC · Creative technologist, Blender artist, AI developer, model merger.
Claude Sonnet 4.6 — Anthropic · Co-architect of Llammy, the Eternal Path narrative, and this very page.