I think the super-AGI should be able to figure this one out. This is the kind of things I would expect to have to think about when prompting a local 4B model, not the latest Sol or whatever is driving the image model here.