Reinforcement Learning for SVG Painting

Definition

The training of neural networks to generate expressive visual art by emitting executable vector code (SVG/Canvas) optimized through visual feedback reward functions.

Key Characteristics

• Formulates mark-making as sequential discrete token generation.
• Optimized against raster similarity (CLIP loss, perceptual metric) without pixel diffusion.
• Produces resolution-independent, editable, highly compact vector artifacts.

Applications

• Generative vector illustration
• Autonomous artistic agents
• Font design

Mentions in Source

• “Formulates painting as a discrete sequential code generation task optimized via visual reward functions.” — surya-narreddi-training-ai-to-paint-with-code