Heterogeneity in linear model

I'm fitting a generalized linear model. The response variable has a gamma distribution. I have 1 fixed (categorical) factor with three levels. The residuals show a quite strong heterogeneity. I'd like to avoid log-transforming the data. Is there any way to allow for different variances for each level? I know FGLS, but that function only returns coefficient values, not the full model specification (pvalues, residuals, AIC, etc.). I've checked the 'Covariancepattern' option in fitglme, but as I understand that applies only to random factors and I don't have any. Is there something similar to the 'weights' argument in R's lme? I can provide the actual data if necessary. Thanks