Tuberculosis continues to kill millions of people each year. The main difficulty in eradication of the disease is the prolonged duration of treatment, which takes at least 6 months. Persister cells have long been associated with failed treatment and disease relapse because of their phenotypical, though transient, tolerance to drugs. By targeting these persisters, the duration of treatment could be shortened, leading to improved tuberculosis treatment and a reduction in transmission. The unique in vivo environment drives the generation of persisters; however, appropriate in vivo mycobacterial persister models enabling optimized drug screening are lacking. To set up a persister infection model that is suitable for this, we infected zebrafish embryos with in vitro-starved Mycobacterium marinum. In vitro starvation resulted in a persister-like phenotype with the accumulation of stored neutral lipids and concomitant increased tolerance to ethambutol. However, these starved wild-type M. marinum organisms rapidly lost their persister phenotype in vivo. To prolong the persister phenotype in vivo, we subsequently generated and analyzed mutants lacking functional resuscitation-promoting factors (Rpfs). Interestingly, the ΔrpfAB mutant, lacking two Rpfs, established an infection in vivo, whereas a nutrient-starved ΔrpfAB mutant did maintain its persister phenotype in vivo. This mutant was, after nutrient starvation, also tolerant to ethambutol treatment in vivo, as would be expected for persisters. We propose that this zebrafish embryo model with ΔrpfAB mutant bacteria is a valuable addition for drug screening purposes and specifically screens to target mycobacterial persisters.