Rebatch latent usage issues

I used comfyui to redraw hundreds of pictures, using two loras and two controlnets (using openpose and lineart models). Unfortunately, the graphics card memory OOM (torch.cuda.OutOfMemoryError: Allocation on device 0 would exceed allowed memory . (out of memory)), and then I found a way to use Rebatch Latent (batch_size is set to 1) to redraw one by one like sd without reporting OOM. I was happy that this solved the problem. Later I discovered that all the redrawn pictures had the same posture. I was a little confused. Did I use it wrong? The following is my workflow chart: