The growing demand for efficient textile manufacturing and the urgent need to reduce environmental waste from overproduction has fueled global interest in the application of artificial intelligence to predicting fashion trends, predicting demand, and selecting sustainable materials. In this paper, we present a proposed comprehensive framework, Textile Intelligence Stack, that integrates four AI model components, including a fine-tuned convolutional neural network for visual trend extraction, a BERT-based natural language processing model for social media signal mining, a long-term memory network for time-series demand forecasting, and an XGBoost classifier for clothing fabric recommendation. Each component is examined in terms of architectural design, data requirements, training methods, and contribution to the overall prediction pipeline. Traditional statistical forecasting methods, such as ARIMA, provide interpretable input data but are subject to significant forecast errors and fail to capture the non-linear, trend-driven nature of fashion demand. The proposed multi-model architecture addresses these limitations by integrating heterogeneous data sources, including social media platforms, e-commerce transaction records, and fashion trend images, into a single manufacturing analytics pipeline. The fabric recommendation module is particularly novel in the existing literature by providing environmentally sensitive material recommendations at the manufacturer level, a functionality that has not been implemented in any previous unified system. Beyond the performance of individual models, this article highlights the critical role of multi-source data fusion, transfer learning, and explainable AI in improving forecast accuracy, recommendation transparency, and practical adoption in textile manufacturing companies. By integrating multidisciplinary insights from deep learning, natural language processing, and sustainable manufacturing research, this study demonstrates that smart textile production requires a unified AI pipeline supported by diverse data streams and manufacturer-oriented decision support rather than relying on a single forecasting model or data source.