EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hugging Face Blog·Sep 3·5 min read
Training a coding model to paint watercolours with TRL and OpenEnv
TL;DR
A coding model was trained to create watercolor paintings using JavaScript and the p5.brush library, resulting in a viral video showcasing its capabilities. The project aims to open-source the training process and models for broader access.
✦ Why It Matters
Explore the open-source resources to replicate or build upon this innovative watercolor painting model.
Key Takeaways
- Explore using TRL and OpenEnv for creative AI projects.
- Utilize aesthetic preferences as a reward mechanism in model training.
- Curate a reference pool to guide model outputs effectively.
- Experiment with different weightings in reward systems to optimize results.
How It Works
The model generates watercolor paintings by writing JavaScript code that utilizes the p5.brush library, which simulates natural drawing techniques. It employs reinforcement learning to optimize its outputs based on aesthetic preferences, comparing generated images against a curated pool of references.
The training process involves adjusting reward weights between two judges to refine the model's artistic style.