RenderFormer is a transformer-based neural rendering pipeline that directly renders images from triangle-based scene representations with full global illumination effects. It formulates rendering as a sequence-to-sequence transformation problem where tokens representing triangles are processed by a transformer model. The system operates without per-scene training or fine-tuning, making it generalize across different scenes. It supports complex lighting effects including specular reflections, shadows, diffuse indirect lighting, and multiple light sources.