The most powerful and modular diffusion model GUI, api and backend with a graph/nodes interface. The fastest local inference engine in the world.
Make CLIP set last layer node work with t5 models.
comfyanonymouse2fafe0686a5
diff --git a/comfy/text_encoders/t5.py b/comfy/text_encoders/t5.pyindex a1420c6cd..f88ed9cf8 100644--- a/comfy/text_encoders/t5.py+++ b/comfy/text_encoders/t5.py@@ -209,6 +209,11 @@ class T5Stack(torch.nn.Module):intermediate = Noneoptimized_attention = optimized_attention_for_device(x.device, mask=attention_mask is not None, small_input=True)past_bias = None++ if intermediate_output is not None:+ if intermediate_output < 0:+ intermediate_output = len(self.block) + intermediate_output+for i, l in enumerate(self.block):x, past_bias = l(x, mask, past_bias, optimized_attention)if i == intermediate_output: