Previous watermarking schemes supporting collusion resistance required the collusion bound q to be declared at setup time, which implied q would be fixed for the system’s entire lifetime. To improve this, we introduce a new flexible corruption model called “dynamic collusion model”, which dynamically decides the collusion bound for each encryption. Based on this, we construct the first public-key watermarking pseudorandom function (PRF) with dynamic collusion bounded. Technically, we first construct an intermediate PRF called “hinting watermarkable PRF” which requires a “hint” about the watermarked PRF key to help extract the embedding message. In our work, we construct a public-key hinting watermarkable PRF from dynamic collusion bounded functional encryption which supports the dynamic collusion model. Applying Yang et al.’s framework in [YYAS22], we can upgrade hinting watermarkable PRF to standard watermarkable PRF.