Fix error with different prompt sizes

[antoinedelplace]

Proposed solution for fixing #4

[laksjdjf]

Thanks!
I have two queations.

1. Does the context repeat affect the generation?
2. Doesn't maxi_prompt_size need to be the least common multiple of the all prompt size?

[antoinedelplace]

You're welcome! :)

1. No, the repeat does not affect the generation when the context size is the same. It fixes the bug when the context sizes are different.
2. Yes, maxi_prompt_size needs to be the least common multiple of all prompt sizes. In practice, CLIP takes no more than 77 tokens and send multiple of 77 token according to the size of the prompt.

[condac]

Applying this fix did not fully solve the issue, Still get error when writing (not that long) inputs.

  File "/opt/ComfyUI/custom_nodes/attention-couple-ComfyUI/attention_couple.py", line 123, in patch
    context_cond = torch.cat([cond.repeat(1, maxi_prompt_size_cond//cond.shape[1], 1) for cond in self.negative_positive_conds[1]], dim=0)
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
RuntimeError: Sizes of tensors must match except in dimension 0. Expected size 231 but got size 154 for tensor number 2 in the list.

[seanphan]

Make sure the size if the width/height is a 64 step (512, 576, 640...)