Total solar irradiance (TSI) has been monitored from space since 1978, i.e. for about four solar cycles. The measurements show a prominent variability in phase with the solar cycle, as well as fluctuations on timescales shorter than a few days. However, the measurements were done by multiple and usually relatively short-lived missions. The different absolute calibrations of the individual instruments and the unaccounted for instrumental trends make estimates of the possible long-term trend in the TSI highly uncertain. Furthermore, both the variability and the uncertainty are strongly wavelength-dependent. While the variability in the UV irradiance is clearly in-phase with the solar cycle, the phase of the variability in the visible range has been debated. In this paper, we aim at getting an insight into the long-term trend of TSI since 1996 and the phase of the solar irradiance variations in the visible part of the spectrum. We use independent ground-based full-disc photometric observations in Ca II K and continuum from the Rome and San Fernando observatories to compute the TSI since 1996. We follow the empirical San Fernando approach based on the photometric sum index. We find a weak declining trend in the TSI of $ {-7.8}_{-0.8}^{+4.9}\times 1{0}^{-3}$ Wm−2 y−1 between the 1996 and 2008 activity minima, while between 2008 and 2019 the reconstructed TSI shows no trend to a marginally decreasing (but statistically insignificant) trend of $ {-0.1}_{-0.02}^{+0.25}\times 1{0}^{-3}$ Wm−2 y−1. The reference TSI series used for the reconstruction does not significantly affect the determined trend. The variation in the blue continuum (409.2 nm) is rather flat, while the variation in the red continuum (607.1 nm) is marginally in anti-phase, although this result is extremely sensitive to the accurate assessment of the quiet Sun level in the images. These results provide further insights into the long-term variation of the TSI. The amplitude of the variations in the visible is below the uncertainties of the processing, which prevents an assessment of the phase of the variations.