Research on speech perception suggests that directing attention to a speaker’s mouth can strengthen visual–auditory associations that support comprehension. This study investigates whether such benefits extend to online second-language (L2) learning by examining the effectiveness of feedback delivered through 3D avatars that reproduce articulatory movements. Sixteen Chinese-speaking learners of Japanese (beginner to intermediate levels) were assigned to either a 3D-avatar feedback condition or an audio-only condition within an online training environment. Participants completed a pre-test, a feedback session, and immediate and delayed post-tests targeting the acquisition of Japanese particles. The 3D avatars, created with Unity and DiDiMo, were presented as short video clips, while the audio-only group received the sound extracted from those same clips. Descriptive analyses showed that although the 3D-avatar group did not outperform the audio-only group at any testing point, they exhibited larger gains from pre-test to both post-tests. This pattern may reflect the 3D-avatar group’s lower initial proficiency; nonetheless, the consistent improvement suggests that visually enriched feedback may support L2 development even when it does not lead to higher absolute performance. Further research with larger samples and methods such as eye-tracking is needed to determine whether increased attention to articulatory cues drives these effects and to clarify the potential pedagogical value of 3D-modeled facial movement in online L2 instruction.