This paper focuses on corrective feedback and examines how trainee-teachers use different semiotic resources to soften feedback sequences during synchronous online interactions. The ISMAEL corpus of webconferencing-supported L2 interactions in French provided data for this qualitative study. Using multimodal transcriptions, the analysis describes multimodal strategies used both in the error and resolution phases of the corrective sequences and in face work during these. These include combining visual and verbal modes in mitigation/remedial strategies and in role-switching cues. [For the complete volume, see ED571330.]