Seems like image generation models are not widely supported yet. Any plans for doing so? How feasible it will be?