🍃 Moss Merry Way — AI Concept Art Generator

Status: Prototype Complete · Stack: Google Colab, Stable Diffusion, 🤗 Hugging Face, xFormers, NVIDIA T4

Overview:
This project is the first creative tech sprint from Glowlock Labs, focused on building a text-to-image concept art generator for visualizing environments in the world of Glowlock. For this sprint, we brought to life the realm of Moss Merry Way—a fog-drenched, magical forest filled with sleepy light and emotional stillness.

Goal:
To generate high-quality concept art with the mood of a handcrafted fairytale world using minimal prompts and lightweight customizations.

Use Case:
These images are intended as a base for looping animated backgrounds, storybook illustration, and AI-assisted visual development for fantasy IPs.

🔧 Tech Stack

• Platform: Google Colab

• Model: Stable Diffusion XL (via Hugging Face)

• Libraries: diffusers, accelerate, xformers, transformers

• Hardware: NVIDIA T4 GPU (Colab Pro)

• Environment Fixes:

  • Forced PyTorch upgrade to 2.1+

  • Downgraded NumPy to <2.0 to resolve compatibility

  • Hugging Face Token authorization

🧪 Sprint Outcome

• ✅ Successful installation and setup of Kohya LoRA Trainer (XL version)

• ✅ Resolved NumPy compatibility errors with patched Colab environment

• ✅ Generated final render capturing the atmosphere of Moss Merry Way

• ✅ Deployed to GitHub and documented workflow for reuse and iteration

• ✅ Published visual + copy across Instagram and Facebook