Background Violence and other harms that result from conflict in forensic inpatient mental health settings are an international problem. De-escalation approaches for reducing conflict are recommended, yet the evidence-base for their use is limited. For the first time, the present study uses implementation science and behaviour change approaches to identify the specific organisational and individual behaviour change targets for enhanced de-escalation in low and medium secure forensic inpatient settings. The primary objective of this study was to identify and describe individual professional, cultural and system-level barriers and enablers to the implementation of de-escalation in forensic mental health inpatient settings. The secondary objective was to identify the changes in capabilities, opportunities and motivations required to enhance de-escalation behaviours in these settings. Methods Qualitative design with data collection and analysis informed by the Theoretical Domains Framework (TDF). Two medium secure forensic mental health inpatient wards and one low secure mental health inpatient ward participated. 12 inpatients and 18 staff participated across five focus groups and one individual interview (at participant preference) guided by a semi-structured interview schedule informed by the TDF domains. Data were analysed via Framework Analysis, organised into the 14 TDF domains then coded inductively within each domain. Results The capabilities required to enhance de-escalation comprised relationship-building, emotional regulation and improved understanding of patients. Staff opportunities for de-escalation are limited by shared beliefs within nursing teams stigmatising therapeutic intimacy in nurse-patient relationships and emotional vulnerability in staff. These beliefs may be modified by ward manager role-modelling. Increased opportunity for de-escalation may be created by increasing service user involvement in antipsychotic prescribing and modifications to the physical environment (sensory rooms and limiting restrictions on patient access to ward spaces). Staff motivation to engage in de-escalation may be increased through reducing perceptions of patient dangerousness via post-incident debriefing and advanced de-escalation planning. Conclusions Interventions to enhance de-escalation in forensic mental health settings should enhance ward staff’s understanding of patients and modify beliefs about therapeutic boundaries which limit the quality of staff-patient relationships. The complex interactions within the capabilities-opportunities-motivation configuration our novel analysis generated, indicates that de-escalation behaviour is unlikely to be changed through knowledge and skills-based training alone. De-escalation training should be implemented with adjunct interventions targeting: collaborative antipsychotic prescribing; debriefing and de-escalation planning; modifications to the physical environment; and ward manager role-modelling of emotional vulnerability and therapeutic intimacy in nurse-patient relationships.