This paper presents an end-to-end EEG-based visual reconstruction framework for zero-shot image classification, retrieval, and reconstruction. The framework consists of a tailored brain encoder called the Adaptive Thinking Mapper (ATM), which projects neural signals from different sources into a shared subspace as clip embeddings, and a two-stage multi-pipe EEG-to-image generation strategy. In the first stage, EEG is embedded to align with high-level clip embeddings, and a prior diffusion model refines the EEG embedding into image priors. A blurry image is also decoded from EEG to maintain low-level features. In the second stage, both the high-level clip embedding, the blurry image, and the caption from EEG latent are input into a pre-trained diffusion model. The framework is evaluated on both EEG and MEG data, demonstrating superior performance in classification, retrieval, and reconstruction. The results show that the EEG-based framework achieves state-of-the-art performance, highlighting the portability, low cost, and high temporal resolution of EEG, enabling a wide range of brain-computer interface (BCI) applications. The code is available at https://github.com/ncclab-sustech/EEG_Image_decode.This paper presents an end-to-end EEG-based visual reconstruction framework for zero-shot image classification, retrieval, and reconstruction. The framework consists of a tailored brain encoder called the Adaptive Thinking Mapper (ATM), which projects neural signals from different sources into a shared subspace as clip embeddings, and a two-stage multi-pipe EEG-to-image generation strategy. In the first stage, EEG is embedded to align with high-level clip embeddings, and a prior diffusion model refines the EEG embedding into image priors. A blurry image is also decoded from EEG to maintain low-level features. In the second stage, both the high-level clip embedding, the blurry image, and the caption from EEG latent are input into a pre-trained diffusion model. The framework is evaluated on both EEG and MEG data, demonstrating superior performance in classification, retrieval, and reconstruction. The results show that the EEG-based framework achieves state-of-the-art performance, highlighting the portability, low cost, and high temporal resolution of EEG, enabling a wide range of brain-computer interface (BCI) applications. The code is available at https://github.com/ncclab-sustech/EEG_Image_decode.