Intent modeling has demonstrated significant potential in sequential recommendation research. However, existing intent modeling methods suffer from issues such as intent noise pollution, one-sided modeling, and the separation of intent modeling from the recommendation task. To address these limitations, we propose a sequential recommendation method based on intent modeling, termed Diff-EISR. This method incorporates a diffusion model to achieve sequence denoising. Additionally, an end-to-end intent modeling module is constructed, and an intent disentanglement loss is introduced to further drive the model to learn comprehensive user intents. Extensive experiments on four public datasets, including Yelp and Sports, demonstrate the superiority of Diff-EISR over existing models.