Paper-based assays for HIV diagnosis are cheap, portable, and require little to no skilled training, electricity, or expensive equipment to perform successfully. However, there is a need for tests with lower limits of detection (LOD) for diagnosing infection at the earliest stages. Effective labeling to increase signal-to-noise ratio is critical to achieving comprehensive early stage HIV detection and would also enable a reduction in the required patient sample volume. Our aim is to assess the LOD of sandwich-format lateral flow assays with localized geometric manipulations along the fluidic flow path and with various labeling schemes, including gold nanoparticles, black cellulose nanoparticles, and enzymatic amplification via horseradish peroxidase (HRP) with colorimetric substrate diaminobenzidine. Premixed and sequential reagent additions produced comparable results, so the premixed mode was selected as preferred due to its shorter time to result. Geometric constrictions offered marginal signal gains with a model system but did not offer similar improvements with our HIV system. The LOD achieved with the enzyme-based labeling scheme (approximately 10 pg/mL) and with cellulose nanoparticles (150 pg/mL) were substantially lower than the LOD achieved using conventional gold nanoparticles (19 ng/mL).