EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago

TL;DR
A new AI model allows users to create images by writing code instead of just using prompts, enhancing control over the creative process. This approach utilizes reinforcement learning to evaluate aesthetic quality through a structured reward system.
✦ Why It Matters
Experiment with coding-based image generation to enhance your creative projects today.
Key Takeaways
How It Works
The system operates through a four-step loop: it receives a prompt, generates a JavaScript sketch, renders it to a PNG, and evaluates it against reference images. The evaluation is conducted by a judge model that selects the better image, providing a reward signal that updates the model for future iterations.
Related