xinyun/ComfyUI
1
0
Fork 0
mirror of https://git.datalinker.icu/comfyanonymous/ComfyUI synced 2026-08-16 01:36:41 +08:00
Code Issues Packages Projects Releases Wiki Activity

Revert "Wan2.2 fun control ref image."

Browse Source 
This reverts commit c2981b397b3a36e6ea16cffc7dbf828be8399120.
This commit is contained in:
comfyanonymous 2025-08-12 22:50:51 -04:00
parent c2981b397b
commit 91b9a6fb34
3 changed files with 0 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
comfy/ldm/wan/model.py
View File

@ -391,7 +391,6 @@ class WanModel(torch.nn.Module):
cross_attn_norm=True, cross_attn_norm=True,
eps=1e-6, eps=1e-6,
flf_pos_embed_token_number=None, flf_pos_embed_token_number=None,
in_dim_ref_conv=None,
image_model=None, image_model=None,
device=None, device=None,
dtype=None, dtype=None,
@ -485,11 +484,6 @@ class WanModel(torch.nn.Module):
else: else:
self.img_emb = None self.img_emb = None
if in_dim_ref_conv is not None:
self.ref_conv = operations.Conv2d(in_dim_ref_conv, dim, kernel_size=patch_size[1:], stride=patch_size[1:], device=operation_settings.get("device"), dtype=operation_settings.get("dtype"))
else:
self.ref_conv = None
def forward_orig( def forward_orig(
self, self,
x, x,
@ -532,13 +526,6 @@ class WanModel(torch.nn.Module):
e = e.reshape(t.shape[0], -1, e.shape[-1]) e = e.reshape(t.shape[0], -1, e.shape[-1])
e0 = self.time_projection(e).unflatten(2, (6, self.dim)) e0 = self.time_projection(e).unflatten(2, (6, self.dim))
full_ref = None
if self.ref_conv is not None:
full_ref = kwargs.get("reference_latent", None)
if full_ref is not None:
full_ref = self.ref_conv(full_ref).flatten(2).transpose(1, 2)
x = torch.concat((full_ref, x), dim=1)
# context # context
context = self.text_embedding(context) context = self.text_embedding(context)
@ -565,9 +552,6 @@ class WanModel(torch.nn.Module):
# head # head
x = self.head(x, e) x = self.head(x, e)
if full_ref is not None:
x = x[:, full_ref.shape[1]:]
# unpatchify # unpatchify
x = self.unpatchify(x, grid_sizes) x = self.unpatchify(x, grid_sizes)
return x return x
@ -586,9 +570,6 @@ class WanModel(torch.nn.Module):
x = torch.cat([x, time_dim_concat], dim=2) x = torch.cat([x, time_dim_concat], dim=2)
t_len = ((x.shape[2] + (patch_size[0] // 2)) // patch_size[0]) t_len = ((x.shape[2] + (patch_size[0] // 2)) // patch_size[0])
if self.ref_conv is not None and "reference_latent" in kwargs:
t_len += 1
img_ids = torch.zeros((t_len, h_len, w_len, 3), device=x.device, dtype=x.dtype) img_ids = torch.zeros((t_len, h_len, w_len, 3), device=x.device, dtype=x.dtype)
img_ids[:, :, :, 0] = img_ids[:, :, :, 0] + torch.linspace(0, t_len - 1, steps=t_len, device=x.device, dtype=x.dtype).reshape(-1, 1, 1) img_ids[:, :, :, 0] = img_ids[:, :, :, 0] + torch.linspace(0, t_len - 1, steps=t_len, device=x.device, dtype=x.dtype).reshape(-1, 1, 1)
img_ids[:, :, :, 1] = img_ids[:, :, :, 1] + torch.linspace(0, h_len - 1, steps=h_len, device=x.device, dtype=x.dtype).reshape(1, -1, 1) img_ids[:, :, :, 1] = img_ids[:, :, :, 1] + torch.linspace(0, h_len - 1, steps=h_len, device=x.device, dtype=x.dtype).reshape(1, -1, 1)

4
comfy/model_base.py
View File

@ -1140,10 +1140,6 @@ class WAN21(BaseModel):
if time_dim_concat is not None: if time_dim_concat is not None:
out['time_dim_concat'] = comfy.conds.CONDRegular(self.process_latent_in(time_dim_concat)) out['time_dim_concat'] = comfy.conds.CONDRegular(self.process_latent_in(time_dim_concat))
reference_latents = kwargs.get("reference_latents", None)
if reference_latents is not None:
out['reference_latent'] = comfy.conds.CONDRegular(self.process_latent_in(reference_latents[-1])[:, :, 0])
return out return out

5
comfy/model_detection.py
View File

@ -373,11 +373,6 @@ def detect_unet_config(state_dict, key_prefix, metadata=None):
flf_weight = state_dict.get('{}img_emb.emb_pos'.format(key_prefix)) flf_weight = state_dict.get('{}img_emb.emb_pos'.format(key_prefix))
if flf_weight is not None: if flf_weight is not None:
dit_config["flf_pos_embed_token_number"] = flf_weight.shape[1] dit_config["flf_pos_embed_token_number"] = flf_weight.shape[1]
ref_conv_weight = state_dict.get('{}ref_conv.weight'.format(key_prefix))
if ref_conv_weight is not None:
dit_config["in_dim_ref_conv"] = ref_conv_weight.shape[1]
return dit_config return dit_config
if '{}latent_in.weight'.format(key_prefix) in state_dict_keys: # Hunyuan 3D if '{}latent_in.weight'.format(key_prefix) in state_dict_keys: # Hunyuan 3D