wangbo/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-12-16 09:42:29 +08:00
Code Issues Actions 11 Packages Projects Releases Wiki Activity

Do padding of audio embed in model for humo for more flexibility. (#9935)

Browse Source
This commit is contained in:
comfyanonymous 2025-09-18 16:54:16 -07:00 committed by GitHub
parent 1ea8c54064
commit 24b0fce099
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 3 additions and 4 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
comfy/ldm/wan/model.py
View File

@ -1551,6 +1551,9 @@ class HumoWanModel(WanModel):
context_img_len = None
if audio_embed is not None:
if reference_latent is not None:
zero_audio_pad = torch.zeros(audio_embed.shape[0], reference_latent.shape[-3], *audio_embed.shape[2:], device=audio_embed.device, dtype=audio_embed.dtype)
audio_embed = torch.cat([audio_embed, zero_audio_pad], dim=1)
audio = self.audio_proj(audio_embed).permute(0, 3, 1, 2).flatten(2).transpose(1, 2)
else:
audio = None

4
comfy_extras/nodes_wan.py
View File

@ -1095,10 +1095,6 @@ class WanHuMoImageToVideo(io.ComfyNode):
audio_emb = torch.stack([feat0, feat1, feat2, feat3, feat4], dim=2)[0] # [T, 5, 1280]
audio_emb, _ = get_audio_emb_window(audio_emb, length, frame0_idx=0)
# pad for ref latent
zero_audio_pad = torch.zeros(ref_latent.shape[2], *audio_emb.shape[1:], device=audio_emb.device, dtype=audio_emb.dtype)
audio_emb = torch.cat([audio_emb, zero_audio_pad], dim=0)
audio_emb = audio_emb.unsqueeze(0)
audio_emb_neg = torch.zeros_like(audio_emb)
positive = node_helpers.conditioning_set_values(positive, {"audio_embed": audio_emb})