Skip to content

fix: freeze shared transformer modules in ZImageControlNet.from_transformer #13136

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
jscaldwell55 wants to merge 1 commit into
base: main
Choose a base branch
from
+17 −0
Open

fix: freeze shared transformer modules in ZImageControlNet.from_transformer #13136

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 17 additions & 0 deletions src/diffusers/models/controlnets/controlnet_z_image.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -517,7 +517,12 @@ def __init__(

@classmethod
def from_transformer(cls, controlnet, transformer):
# Scalar value — immutable, direct assignment is safe
controlnet.t_scale = transformer.t_scale

# Shared modules from transformer — these are intentionally shared references.
# They should remain frozen during controlnet training; only control_* modules
# are meant to be trained.
controlnet.t_embedder = transformer.t_embedder
controlnet.all_x_embedder = transformer.all_x_embedder
controlnet.cap_embedder = transformer.cap_embedder
Expand All @@ -526,6 +531,18 @@ def from_transformer(cls, controlnet, transformer):
controlnet.context_refiner = transformer.context_refiner
controlnet.x_pad_token = transformer.x_pad_token
controlnet.cap_pad_token = transformer.cap_pad_token

# Freeze all shared modules/parameters to prevent accidental training.
# Only control_* modules (control_layers, control_all_x_embedder,
# control_noise_refiner) should be trained.
controlnet.t_embedder.requires_grad_(False)
controlnet.all_x_embedder.requires_grad_(False)
controlnet.cap_embedder.requires_grad_(False)
controlnet.noise_refiner.requires_grad_(False)
controlnet.context_refiner.requires_grad_(False)
controlnet.x_pad_token.requires_grad_(False)
controlnet.cap_pad_token.requires_grad_(False)

return controlnet

@staticmethod
Expand Down