We compute the elliptic flow v2 of thermal photons in a strongly coupled plasma with constant magnetic field via gauge/gravity duality. The D3/D7 embedding is applied to generate the contributions from massive quarks. By considering the cases in 2+1 flavor SYM analogous to the photon production in QGP, we obtain the thermal-photon v2, which is qualitatively consistent with the direct-photon v2 measured in RHIC at intermediate energy. However, due to the simplified setup, the thermal-photon v2 in our model should be regarded as the upper bound for the v2 generated by solely magnetic field in the strongly coupled scenario.The elliptic flow v 2 characterizes the momentum anisotropy of produced particles in heavy ion collisions. The recent observations from RHIC and LHC revealed surprising results, where the large elliptic flow of direct photons has been measured [1,2]. Unlike the hadronic flow, the large flow of direct photons is unexpected since the high-energy photons are presumed to be generated in early times, where the initial flow should be relatively small compared to the flow built up by hydrodynamics. The anisotropy flow of thermal photons with viscous hydrodynamics has been recently reported in [3,4]. In theory, novel mechanisms should be introduced to break the azimuthal symmetry of photon production. The magnetic field led by colliding nuclei has been recently considered as one of possible candidates to bring about the large flow. In the weakly coupled scenario, the photon production with magnetic field has been studied in a variety of approaches [5][6][7][8][9]. Other mechanism such as the synchrotron radiation from the interaction of escaping quarks with the collective confining color field has been proposed in [10].However, in the strongly coupled scenario, the perturbative approaches may not be applied. The AdS/CFT correspondence [11][12][13][14][15], a holographic duality between a strongly coupled N = 4 Super Yang-Mills(SYM) theory and a classical supergravity in the asymptotic AdS 5 × S 5 background in the limit of large N c and strong t'Hooft