The faster light-curve evolution of low-luminosity Type Ia supernovae (SNe Ia) suggests that they could result from the explosion of white dwarf (WD) progenitors below the Chandrasekhar mass (M_Ch). Here we present 1D non-local thermodynamic equilibrium time-dependent radiative transfer simulations of pure central detonations of carbon–oxygen WDs with a mass (M_tot) between 0.88 and 1.15 M_⊙ and a ^56Ni yield between 0.08 and 0.84 M_⊙. Their lower ejecta density compared to M_Ch models results in a more rapid increase of the luminosity at early times and an enhanced γ-ray escape fraction past maximum light. Consequently, their bolometric light curves display shorter rise times and larger post-maximum decline rates. Moreover, the higher M(^56Ni)/M_tot ratio at a given ^56Ni mass enhances the temperature and ionization level in the spectrum-formation region for the less luminous models, giving rise to bluer colours at maximum light and a faster post-maximum evolution of the B − V colour. For sub-M_Ch models fainter than M_B ≈ −18.5 mag at peak, the greater bolometric decline and faster colour evolution lead to a larger B-band post-maximum decline rate, ΔM_15(B). In particular, all of our previously published M_Ch models (standard and pulsational delayed detonations) are confined to ΔM_15(B) < 1.4 mag, while the sub-M_Ch models with M_tot ≲ 1 M_⊙ extend beyond this limit to ΔM_15(B) ≈ 1.65 mag for a peak M_B ≈ −17 mag, in better agreement with the observed width–luminosity relation (WLR). Regardless of the precise ignition mechanism, these simulations suggest that fast-declining SNe Ia at the faint end of the WLR could result from the explosion of WDs whose mass is significantly below the Chandrasekhar limit.