arXiv cs.CV
· Papers
SymbOmni: Evolving Agentic Omni Models via Symbolic Concept Learning
arXiv:2607.12042v1 Announce Type: new Abstract: Visual generation is increasingly ubiquitous in diverse domains, from text-to-image/video synthesis to multimodal interactive creation. Yet prevailing monolithic models remain fundamentally constrained by their inability to learn cumulatively and evolve autonomously, whic