LeafTip-RN: Generative AI-powered temporal interpolation for continuous phenotypic analysis of seedling establishment traits in wheat.

Seedling establishment represents a critical phase in early crop growth and development, directly influencing biomass accumulation and yield potential. To characterise early growth dynamics under field conditions, both growth rate and uniformity of emergence need to be assessed continuously; however, manual quantification of these dynamic traits in large-scale trials remains impractical. Here, we present LeafTip-RN, an open-source and deep learning (DL)-powered pipeline for dynamically measuring wheat (Triticum aestivum L.) early establishment in the field. To enable flexible and scalable data collection, ultralow-altitude drone phenotyping was employed, followed by the development of an optimised DL model to automate leaf-tip-related feature extraction from complex backgrounds. Notably, to address data sparsity arising from eight phenotyping timepoints, we integrated an image-to-video generative AI (GenAI) module into the pipeline to interpolate keyframes between early and late seedling stages (i.e. 18-40 days after sowing), resulting in a training library comprising 353,019 labelled leaf tips. Using the pipeline, we successfully quantified multiple agronomically important establishment-related traits (e.g. plot-level leaf tips and seedling spatial uniformity), followed by deriving their growth curves for 51 wheat varieties across two growing seasons (2024-2026). After validating these LeafTip-RN-derived traits, we further computed varietal relative growth rates and uniformity indices, based on which the 51 varieties were classified into high-, medium-, and low-performance groups, revealing discrepancies between LeafTip-RN-derived classification (18-40 DAS) and manual assessment at 40 DAS when dynamic early performance was considered. Finally, to facilitate broad adoption by the plant research community, we developed an openly accessible graphical user interface (GUI) for non-expert users to visualise and analyse rapid seedling developmental changes. Taken together, our study provides a scalable GenAI-powered solution for evaluating seedling establishment in wheat, offering valuable tools for breeders and researchers to identify varieties with enhanced early growth vigour and emergence dynamics that are extensible to other cereal crops.