Implementation of conditioning with multiple images.

[Mixomo]

I propose that, similar to what TRELLIS recently implemented, conditioning with support for multiple images (if the architecture allows it) should be implemented. This would provide information from different angles of the subject and/or object, thereby improving reconstruction.

Thanks.