You have a path model with multiple outcome (dependent/endogenous) variables and also mediator variables. G*Power can only handle regression models with a single dependent variable. You either have to break your path model down in "submodels" (individual regression models; one for each outcome/DV) if possible and run G*Power separately for each DV (with its own set of predictors).

This isn't ideal though because you would want to run your sample size analysis for the entire model with all paths/regression coefficients combined. I would recommend using a more general and comprehensive way of determining the optimal sample size that relies on simulation. With a Monte Carlo simulation, you can assess the entire path analytic model with all paths in a single step. This allows you to not only to determine power for all paths but to also examine other key aspects of sample size planning such as whether model goodness of fit statistics, parameter estimates, and standard errors are unbiased and whether you have adequate coverage (proper confidence interval estimation) for a given sample size. For a reference, see

Article How to Use a Monte Carlo Study to Decide on Sample Size and ...

I offer a free mini-course on sample size planning for path analysis using simulation that you can find here: https://www.goquantfish.com/courses/sample-size-planning-in-mplus

Thanks Prof Christian Geiser for your insightful sharing. I never know that G*Power cannot handle regression models with more than one dependent variable (DV) as I have seen some studies with few DVs used G*Power.