We address novel view synthesis from RGB-D images with a limited number of input views. Existing methods often struggle under such conditions and are sensitive to photometric inconsistencies and noisy depth measurements. We propose FaDeN, a fast method that estimates absolute scale, enhances sensor depth maps for supervision, corrects photometric variations, and synthesizes virtual views for improved scene coverage. FaDeN reduces training time by more than 200 \(\times \) compared to similar NeRF-based methods while producing novel views of comparable or superior quality. Experiments on public and custom datasets demonstrate improvements in novel view synthesis, enhancing both image quality and geometric accuracy. In addition, we show that FaDeN improves the accuracy and completeness of 3D surface reconstructions.