Reliable fish passage monitoring is difficult in turbid lowland rivers, where visual observation is limited and continuous manual sampling is rarely practical. This challenge is central to evaluating temporary fish passage technologies being trialled near Menindee to restore native fish movement in the Darling-Baaka system. During the 2025–26 migration season, a temporarily installed Fishheart hydraulic fishway moved close to 30,000 fish, creating a need to validate the AI-facilitated workflow used to monitor passage through the system. Using a small subsample of passage events collected over two weeks of the season, we validated workflow counts against fish captured in an outlet net across matched net-sampling intervals. The workflow combined automated image-based detection and preliminary AI classification with fisheries staff review, species verification and count generation. Overall, the workflow performed well under challenging conditions, counting 278 fish compared with 315 fish captured in the net, suggesting it provided a useful and likely conservative estimate of passage. However, species-specific biases were evident. Golden perch and Carp were undercounted, most likely because some individuals passed through the system without being detected or imaged under turbid, restricted-visibility conditions. In contrast, Bony herring detections exceeded net captures, with supplementary review suggesting some small fish may have passed the camera but escaped the net. These results show that AI-facilitated monitoring can support fish passage assessment, but validation must account for both detection limits and imperfections in the physical reference method.