The microstructure of phenolic resin undergoes significant transformation under high-temperature pyrolysis, affecting its mechanical performance and fracture behavior. By combining Convolutional Neural Networks (CNN) and Conditional Variational Autoencoders (CVAE), a generative modeling framework was proposed. Then its suitability to predict microstructure evolution of phenolic resin under varying pyrolysis temperatures was studied. Finite Element Method (FEM) simulations were conducted to analyze stress distributions. Results indicate a significant increase in the area of high stress concentration zones with rising pyrolysis temperature, with pore bridges, sharp edges, and clustered porosity identified as potential fracture initiation sites.