Skip to content

fix: Handle ControlNet conditioning args correctly in do_controlnet_step (closes #9444) #9469

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
botbikamordehai2-sketch wants to merge 1 commit into
base: main
Choose a base branch
from
+18 −7
Open

fix: Handle ControlNet conditioning args correctly in do_controlnet_step (closes #9444) #9469

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
25 changes: 18 additions & 7 deletions invokeai/backend/stable_diffusion/diffusion/shared_invokeai_diffusion.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -68,6 +68,8 @@
# control_data should be type List[ControlNetData]
# this loop covers both ControlNet (one ControlNetData in list)
# and MultiControlNet (multiple ControlNetData in list)
if control_data is None:
return sample
for _i, control_datum in enumerate(control_data):
control_mode = control_datum.control_mode
# soft_injection and cfg_injection are the two ControlNet control_mode booleans
Expand All @@ -83,15 +85,24 @@
last_control_step = math.ceil(control_datum.end_step_percent * total_step_count)
# only apply controlnet if current step is within the controlnet's begin/end step range
if step_index >= first_control_step and step_index <= last_control_step:
if cfg_injection:
sample_model_input = sample
else:
# expand the latents input to control model if doing classifier free guidance
# (which I think for now is always true, there is conditional elsewhere that stops execution if
# classifier_free_guidance is <= 1.0 ?)
sample_model_input = torch.cat([sample] * 2)
# Build cross_attention_kwargs with controlnet residuals
cross_attention_kwargs = {
"down_block_additional_residuals": down_block_res_samples,
"mid_block_additional_residual": mid_block_res_sample,
"controlnet_conditioning_scale": control_datum.control_weight,
"soft_injection": soft_injection,
"cfg_injection": cfg_injection,
}
# Call the model forward callback with the correct signature
# The callback may accept either (x, t, conditioning) or (x, t, conditioning, cross_attention_kwargs)
try:
sample = self.model_forward_callback(sample, timestep, conditioning_data, cross_attention_kwargs)
except TypeError:
# Fallback for older callback signature without cross_attention_kwargs
sample = self.model_forward_callback(sample, timestep, conditioning_data)
return samplet = torch.cat([sample] * 2)

Check failure on line 103 in invokeai/backend/stable_diffusion/diffusion/shared_invokeai_diffusion.py

View workflow job for this annotation

GitHub Actions / python-checks

Ruff 

invokeai/backend/stable_diffusion/diffusion/shared_invokeai_diffusion.py:103:24: SyntaxError: Expected a statement

added_cond_kwargs = None

Check failure on line 105 in invokeai/backend/stable_diffusion/diffusion/shared_invokeai_diffusion.py

View workflow job for this annotation

GitHub Actions / python-checks

Ruff 

invokeai/backend/stable_diffusion/diffusion/shared_invokeai_diffusion.py:105:1: SyntaxError: Unexpected indentation

if cfg_injection: # only applying ControlNet to conditional instead of in unconditioned
if conditioning_data.is_sdxl():
Expand Down Expand Up @@ -494,3 +505,3 @@
scaled_delta = (conditioned_next_x - unconditioned_next_x) * guidance_scale
combined_next_x = unconditioned_next_x + scaled_delta
return combined_next_x
Loading