Liquid: Scalable Multi-modal Generation

Unified multimodal generation framework that combines visual understanding and image generation within a single LLM without external pretrained visual tools. Demonstrates favorable scaling laws where joint visual-language training degradation diminishes with model size.