This classroom action research investigates how a music-based, AI-supported CALL approach can improve grammar learning and reduce affective barriers among Thai primary EFL learners. Grounded in Krashen’s Affective Filter Hypothesis, the study integrates LyricsTraining for interactive song-based grammar tasks, Task-Based Language Teaching (TBLT) for structured pre-task/task/post-task cycles, and Project-Based Learning (PBL) for creative multimodal output. Participants are 45 Grade 5 students (CEFR A1–A2) at La-orutis Demonstration School, Lampang, Thailand, implementing a 10-week intervention across two cycles: Cycle 1 focuses on grammar learning through LyricsTraining and collaborative analysis of target forms (present passive, will-future, first conditional, and present perfect with for/since). Cycle 2 culminates in a group mini project where students compose English songs using SUNO AI and design promotional posters with Canva for Education. Data will be collected through 20-item grammar tests (pre, post-1, post-2), an affective filter questionnaire, observation checklists, student reflection logs, and a song–poster rubric. Quantitative analyses (means, SD, paired-sample t-tests) and qualitative thematic analysis will examine changes in grammar performance, motivation, confidence, and anxiety. Expected outcomes include improved grammar accuracy, reduced anxiety, increased confidence and motivation, and high engagement through AI-assisted creative production.