ABSTRACT Giant planets are expected to form at orbital radii that are relatively large compared to transit and radial velocity detections (>1 au). As a result, giant planet formation is best observed through direct imaging. By simulating the formation of giant (0.3–5MJ) planets by core accretion, we predict planet magnitude in the near-infrared (2–4 μm) and demonstrate that, once a planet reaches the runaway accretion phase, it is self-luminous and is bright enough to be detected in near-infrared wavelengths. Using planet distribution models consistent with existing radial velocity and imaging constraints, we simulate a large sample of systems with the same stellar and disc properties to determine how many planets can be detected. We find that current large (8–10 m) telescopes have at most a 0.2 per cent chance of detecting a core-accretion giant planet in the L’ band and 2 per cent in the K band for a typical solar-type star. Future instruments such as METIS and VIKiNG have higher sensitivity and are expected to detect exoplanets at a maximum rate of 2 and 8 per cent, respectively.