mirrors/ComfyUI
1
0
Fork 0
mirror of https://github.com/comfyanonymous/ComfyUI.git synced 2025-09-14 13:35:05 +00:00
Code Issues Packages Projects Releases Wiki Activity

USO style reference. (#9677)

Browse Source 
Load the projector.safetensors file with the ModelPatchLoader node and use
the siglip_vision_patch14_384.safetensors "clip vision" model and the
USOStyleReferenceNode.
This commit is contained in:
comfyanonymous
2025-09-02 12:36:22 -07:00
committed by GitHub
parent e2d1e5dad9
commit 3412d53b1d
5 changed files with 222 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
comfy/clip_model.py
View File

@@ -61,8 +61,12 @@ class CLIPEncoder(torch.nn.Module):
def forward(self, x, mask=None, intermediate_output=None):
optimized_attention = optimized_attention_for_device(x.device, mask=mask is not None, small_input=True)
all_intermediate = None
if intermediate_output is not None:
if intermediate_output < 0:
if intermediate_output == "all":
all_intermediate = []
intermediate_output = None
elif intermediate_output < 0:
intermediate_output = len(self.layers) + intermediate_output
intermediate = None
@@ -70,6 +74,12 @@ class CLIPEncoder(torch.nn.Module):
x = l(x, mask, optimized_attention)
if i == intermediate_output:
intermediate = x.clone()
if all_intermediate is not None:
all_intermediate.append(x.unsqueeze(1).clone())
if all_intermediate is not None:
intermediate = torch.cat(all_intermediate, dim=1)
return x, intermediate
class CLIPEmbeddings(torch.nn.Module):