This study introduces Concept2Brain, an artificial intelligence system that predicts how brain activity unfolds when someone views an image or reads a short passage using event-related potentials from EEG that measure perception, attention, and emotion. It first captures the meaning of the input with a vision and language encoder and then translates that concept signal into an EEG representation learned from human recordings through a conditioned variational autoencoder, which organizes activity into a compact latent space and allows the model to generate full multi-channel waveforms for any virtual participant that resemble real responses. The synthetic signals recover familiar patterns from the literature, including early sensory peaks, a classic attention-related positivity, and a late positive potential that increases with emotional content. The model also reproduces a face-selective response with the expected right posterior scalp distribution that aligns with the N170 benchmark in face perception studies. The system maintains realistic differences across individuals rather than collapsing to a single average, and it works for both images and short text prompts that describe affective scenes. The team provides a cloud platform where users can upload images or enter text, select virtual participants, visualize scalp activity over time, and download data, allowing open and reproducible in silico EEG datasets without running a full experiment for each new stimulus set.