Skip to content

[Wan] Standardize vae.encode() sampling mode in WanVideoToVideoPipeline #11639

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 2 commits into from
Jun 11, 2025
Merged

[Wan] Standardize vae.encode() sampling mode in WanVideoToVideoPipeline #11639

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6ee6b53
fix: vae sampling mode
tolgacangoz May 30, 2025
40243c8
fix a typo
tolgacangoz May 30, 2025
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
9 changes: 2 additions & 7 deletions src/diffusers/pipelines/wan/pipeline_wan_video2video.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -419,12 +419,7 @@ def prepare_latents(
)

if latents is None:
if isinstance(generator, list):
init_latents = [
retrieve_latents(self.vae.encode(video[i].unsqueeze(0)), generator[i]) for i in range(batch_size)
]
else:
init_latents = [retrieve_latents(self.vae.encode(vid.unsqueeze(0)), generator) for vid in video]
init_latents = [retrieve_latents(self.vae.encode(vid.unsqueeze(0)), sample_mode="argmax") for vid in video]

init_latents = torch.cat(init_latents, dim=0).to(dtype)

Expand All @@ -441,7 +436,7 @@ def prepare_latents(
if hasattr(self.scheduler, "add_noise"):
latents = self.scheduler.add_noise(init_latents, noise, timestep)
else:
latents = self.scheduelr.scale_noise(init_latents, timestep, noise)
latents = self.scheduler.scale_noise(init_latents, timestep, noise)
else:
latents = latents.to(device)

Expand Down