Stable Diffusion, not a web service. There’s community resources for training models. Some of the combinations I’m currently making use a combined 60gb of training models, require multiple reimaging passes, and lately I’ve been expanding into pre-processing.