Long-read transcriptome sequencing (long-RNA-seq) revolutionizes transcriptome research by enabling full-length transcript analysis for comprehensive exploration of isoform diversity. We developed IFDlong, a probabilistic framework and software suite for detecting isoform and fusion transcripts from bulk or single-cell long-RNA-seq data. IFDlong annotates each long read, identifies novel isoforms, quantifies expression via an expectation-maximization algorithm, and profiles fusion transcripts. In large-scale simulation and real data analyses, IFDlong outperforms existing tools and demonstrated high accuracy and robustness across multiple in-house and public datasets, including healthy tissues, cell lines, and different diseases.