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.

Pad ace step 1.5 ref audio if not long enough. (#12341)

comfyanonymous17e7df43d19b
diff --git a/comfy/model_base.py b/comfy/model_base.py
index 3aa345254..858789b30 100644
--- a/comfy/model_base.py
+++ b/comfy/model_base.py
@@ -1577,6 +1577,10 @@ class ACEStep15(BaseModel):
else:
out['is_covers'] = comfy.conds.CONDConstant(False)
+ if refer_audio.shape[2] < noise.shape[2]:
+ pad = comfy.ldm.ace.ace_step15.get_silence_latent(noise.shape[2], device)
+ refer_audio = torch.cat([refer_audio.to(pad), pad[:, :, refer_audio.shape[2]:]], dim=2)
+
out['refer_audio'] = comfy.conds.CONDRegular(refer_audio)
return out