I think that given that context, the idea they used it as a base image and generated the rest on top of it does actually seem quite likely. Especially if they’d used a tool like ControlNet, that would explain how the text was still as consistent as it is, albeit not perfect.
I think that given that context, the idea they used it as a base image and generated the rest on top of it does actually seem quite likely. Especially if they’d used a tool like ControlNet, that would explain how the text was still as consistent as it is, albeit not perfect.