Skip to content

denormalize latents with the mean and std if available #7111

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

Merged
merged 4 commits into from
Feb 27, 2024
Merged

denormalize latents with the mean and std if available #7111

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
7d94a9e
denormalize latents with the mean and std if available
patil-suraj Feb 27, 2024
b9398c4
fix denormalize
patil-suraj Feb 27, 2024
bf7a7ec
add latent mean and std in vae config
patil-suraj Feb 27, 2024
6f5ce08
address sayak's comment
patil-suraj Feb 27, 2024
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
2 changes: 2 additions & 0 deletions src/diffusers/models/autoencoders/autoencoder_kl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -80,6 +80,8 @@ def __init__(
norm_num_groups: int = 32,
sample_size: int = 32,
scaling_factor: float = 0.18215,
latents_mean: Optional[Tuple[float]] = None,
latents_std: Optional[Tuple[float]] = None,
force_upcast: float = True,
):
super().__init__()
Expand Down
17 changes: 16 additions & 1 deletion src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1313,7 +1313,22 @@ def __call__(
self.upcast_vae()
latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)

image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
# unscale/denormalize the latents
# denormalize with the mean and std if available and not None
has_latents_mean = hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None
has_latents_std = hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None
if has_latents_mean and has_latents_std:
latents_mean = (
torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1).to(latents.device, latents.dtype)
)
latents_std = (
torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1).to(latents.device, latents.dtype)
)
latents = latents * latents_std / self.vae.config.scaling_factor + latents_mean
else:
latents = latents / self.vae.config.scaling_factor

image = self.vae.decode(latents, return_dict=False)[0]

# cast back to fp16 if needed
if needs_upcasting:
Expand Down