Interactive collaborative video is now a common part of remote work. Despite its prevalence, traditional video conferencing can be challenging, sometimes causing social discomforts that undermine process and outcomes. Avatars on 2D displays offer a promising alternative for enhancing self-representation, bridging the gap between virtual reality (VR) and traditional non-immersive video. However, the use of such avatars in activity-oriented group settings remains underexplored. To address this gap, we conducted a mixed-methods, within-subject study investigating the impacts of avatar-mediated versus traditional video representations on collaboration satisfaction and self-esteem. 32 participants (8 groups of 4 with pre-established relationships) engaged in goal-directed activities, followed by group interviews. Results indicate that avatars significantly enhance self-esteem and collaboration satisfaction, while qualitative insights reveal the dynamic perceptions and experiences of avatars, including benefits, challenges, and factors influencing adoption likelihood. Our study contributes to understanding and implications of avatars as a camera-driven representation in video-mediated collaborative interactions.