Skip to content

[wan2.2] fix vae patches #12041

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 1 commit into from
Aug 1, 2025
Merged

[wan2.2] fix vae patches #12041

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
78 changes: 25 additions & 53 deletions src/diffusers/models/autoencoders/autoencoder_kl_wan.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -913,38 +913,21 @@ def patchify(x, patch_size):
if patch_size == 1:
return x

if x.dim() == 4:
# x shape: [batch_size, channels, height, width]
batch_size, channels, height, width = x.shape

# Ensure height and width are divisible by patch_size
if height % patch_size != 0 or width % patch_size != 0:
raise ValueError(f"Height ({height}) and width ({width}) must be divisible by patch_size ({patch_size})")

# Reshape to [batch_size, channels, height//patch_size, patch_size, width//patch_size, patch_size]
x = x.view(batch_size, channels, height // patch_size, patch_size, width // patch_size, patch_size)

# Rearrange to [batch_size, channels * patch_size * patch_size, height//patch_size, width//patch_size]
x = x.permute(0, 1, 3, 5, 2, 4).contiguous()
x = x.view(batch_size, channels * patch_size * patch_size, height // patch_size, width // patch_size)

elif x.dim() == 5:
# x shape: [batch_size, channels, frames, height, width]
batch_size, channels, frames, height, width = x.shape

# Ensure height and width are divisible by patch_size
if height % patch_size != 0 or width % patch_size != 0:
raise ValueError(f"Height ({height}) and width ({width}) must be divisible by patch_size ({patch_size})")
if x.dim() != 5:
raise ValueError(f"Invalid input shape: {x.shape}")
# x shape: [batch_size, channels, frames, height, width]
batch_size, channels, frames, height, width = x.shape

# Reshape to [batch_size, channels, frames, height//patch_size, patch_size, width//patch_size, patch_size]
x = x.view(batch_size, channels, frames, height // patch_size, patch_size, width // patch_size, patch_size)
# Ensure height and width are divisible by patch_size
if height % patch_size != 0 or width % patch_size != 0:
raise ValueError(f"Height ({height}) and width ({width}) must be divisible by patch_size ({patch_size})")

# Rearrange to [batch_size, channels * patch_size * patch_size, frames, height//patch_size, width//patch_size]
x = x.permute(0, 1, 4, 6, 2, 3, 5).contiguous()
x = x.view(batch_size, channels * patch_size * patch_size, frames, height // patch_size, width // patch_size)
# Reshape to [batch_size, channels, frames, height//patch_size, patch_size, width//patch_size, patch_size]
x = x.view(batch_size, channels, frames, height // patch_size, patch_size, width // patch_size, patch_size)

else:
raise ValueError(f"Invalid input shape: {x.shape}")
# Rearrange to [batch_size, channels * patch_size * patch_size, frames, height//patch_size, width//patch_size]
x = x.permute(0, 1, 6, 4, 2, 3, 5).contiguous()
x = x.view(batch_size, channels * patch_size * patch_size, frames, height // patch_size, width // patch_size)

return x

Expand All @@ -953,29 +936,18 @@ def unpatchify(x, patch_size):
if patch_size == 1:
return x

if x.dim() == 4:
# x shape: [b, (c * patch_size * patch_size), h, w]
batch_size, c_patches, height, width = x.shape
channels = c_patches // (patch_size * patch_size)

# Reshape to [b, c, patch_size, patch_size, h, w]
x = x.view(batch_size, channels, patch_size, patch_size, height, width)

# Rearrange to [b, c, h * patch_size, w * patch_size]
x = x.permute(0, 1, 4, 2, 5, 3).contiguous()
x = x.view(batch_size, channels, height * patch_size, width * patch_size)

elif x.dim() == 5:
# x shape: [batch_size, (channels * patch_size * patch_size), frame, height, width]
batch_size, c_patches, frames, height, width = x.shape
channels = c_patches // (patch_size * patch_size)
if x.dim() != 5:
raise ValueError(f"Invalid input shape: {x.shape}")
# x shape: [batch_size, (channels * patch_size * patch_size), frame, height, width]
batch_size, c_patches, frames, height, width = x.shape
channels = c_patches // (patch_size * patch_size)

# Reshape to [b, c, patch_size, patch_size, f, h, w]
x = x.view(batch_size, channels, patch_size, patch_size, frames, height, width)
# Reshape to [b, c, patch_size, patch_size, f, h, w]
x = x.view(batch_size, channels, patch_size, patch_size, frames, height, width)

# Rearrange to [b, c, f, h * patch_size, w * patch_size]
x = x.permute(0, 1, 4, 5, 2, 6, 3).contiguous()
x = x.view(batch_size, channels, frames, height * patch_size, width * patch_size)
# Rearrange to [b, c, f, h * patch_size, w * patch_size]
x = x.permute(0, 1, 4, 5, 3, 6, 2).contiguous()
x = x.view(batch_size, channels, frames, height * patch_size, width * patch_size)

return x

Expand Down Expand Up @@ -1044,7 +1016,6 @@ def __init__(
patch_size: Optional[int] = None,
scale_factor_temporal: Optional[int] = 4,
scale_factor_spatial: Optional[int] = 8,
clip_output: bool = True,
) -> None:
super().__init__()

Expand Down Expand Up @@ -1244,10 +1215,11 @@ def _decode(self, z: torch.Tensor, return_dict: bool = True):
out_ = self.decoder(x[:, :, i : i + 1, :, :], feat_cache=self._feat_map, feat_idx=self._conv_idx)
out = torch.cat([out, out_], 2)

if self.config.clip_output:
out = torch.clamp(out, min=-1.0, max=1.0)
if self.config.patch_size is not None:
out = unpatchify(out, patch_size=self.config.patch_size)

out = torch.clamp(out, min=-1.0, max=1.0)

self.clear_cache()
if not return_dict:
return (out,)
Expand Down
Loading