This paper introduces AlphaFLOW and ESMFLOW, generative models that combine AlphaFold and ESMFold with flow matching to sample protein ensembles. The method repurposes existing single-state predictors to generate diverse protein structures, capturing conformational flexibility and ensemble properties. By fine-tuning AlphaFold and ESMFold under a custom flow matching framework, the models produce sequence-conditioned generative models that outperform traditional methods like MSA subsampling in precision and diversity. When trained on molecular dynamics (MD) ensembles, the models accurately capture structural flexibility, positional distributions, and higher-order ensemble observables. AlphaFLOW and ESMFLOW also demonstrate faster convergence to equilibrium properties compared to MD simulations, making them a promising alternative for expensive simulations. The models are evaluated on PDB and MD ensembles, showing improved performance in predicting conformational flexibility, atomic position distributions, and ensemble observables. The results highlight the potential of flow matching for generating diverse protein structures and improving structural biology research.This paper introduces AlphaFLOW and ESMFLOW, generative models that combine AlphaFold and ESMFold with flow matching to sample protein ensembles. The method repurposes existing single-state predictors to generate diverse protein structures, capturing conformational flexibility and ensemble properties. By fine-tuning AlphaFold and ESMFold under a custom flow matching framework, the models produce sequence-conditioned generative models that outperform traditional methods like MSA subsampling in precision and diversity. When trained on molecular dynamics (MD) ensembles, the models accurately capture structural flexibility, positional distributions, and higher-order ensemble observables. AlphaFLOW and ESMFLOW also demonstrate faster convergence to equilibrium properties compared to MD simulations, making them a promising alternative for expensive simulations. The models are evaluated on PDB and MD ensembles, showing improved performance in predicting conformational flexibility, atomic position distributions, and ensemble observables. The results highlight the potential of flow matching for generating diverse protein structures and improving structural biology research.