What is GLM-Image GLM-Image is a discrete auto-regressive image generation model designed for knowledge-dense prompts and high-fidelity output. A hybrid architecture lets the auto-regressive generator focus on semantic structure while the diffusion decoder restores fine details.
Hybrid AR + Diffusion A 9B auto-regressive generator produces semantic tokens, while a 7B DiT diffusion decoder reconstructs fine-grained details. Dense-Knowledge Rendering Strong semantic alignment and text rendering enable complex, knowledge-rich visuals. Text-to-Image + Image-to-Image Supports generation, editing, style transfer, identity preservation, and multi-subject consistency.
