They have it but they are not returning. I think you’re doing some of that. In mid-journey, you can give them a picture and say give me the prompt that will generate it. And if the prompt is very, very long, then it means that the model is not tuned to that picture. But it’s not likely for us to just keep paying the API fees just to reverse engineer the logits. It’s much easier to work just with lava or any of this open source.