Skip to content

The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.

Let hunyuan dit work with all prompt lengths.

comfyanonymous8328a2d8cdab
diff --git a/comfy/ldm/hydit/poolers.py b/comfy/ldm/hydit/poolers.py
index 3470041b5..2c6e46e67 100644
--- a/comfy/ldm/hydit/poolers.py
+++ b/comfy/ldm/hydit/poolers.py
@@ -16,6 +16,7 @@ class AttentionPool(nn.Module):
self.embed_dim = embed_dim
def forward(self, x):
+ x = x[:,:self.positional_embedding.shape[0] - 1]
x = x.permute(1, 0, 2) # NLC -> LNC
x = torch.cat([x.mean(dim=0, keepdim=True), x], dim=0) # (L+1)NC
x = x + self.positional_embedding[:, None, :].to(dtype=x.dtype, device=x.device) # (L+1)NC