quenstion about training from scrash.

[pekinghk]

Noting that the training method of VACE's Context Adapter Tuning is somewhat similar to that of ControlNet, and the block design of VACE closely resembles the structure of the main network's block design, when VACE is trained from scratch, does it copy the corresponding weights from the corresponding blocks of the main network and perform operations similar to zero-conv? I look forward to your response.