storytold/FineTrainers-Conditioning
2
0
Fork 0
mirror of https://github.com/storytold/FineTrainers-Conditioning.git synced 2026-10-22 08:05:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Make a fix to the video tensors.

Browse Source
This commit is contained in:
CrossProduct
2025-01-28 23:03:04 +00:00
parent 8ec35d58fb
commit 4ab3b25c13
2 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
finetrainers/conditioning/conditioned_pipeline.py
+1 -1
View File
@@ -198,7 +198,7 @@ class LTXConditionedPipeline(LTXPipeline):
img_ref_latent = condition_latents_prepare.prepare_latents_for_conditioning( img_ref_latent = condition_latents_prepare.prepare_latents_for_conditioning(
vae=self.vae, vae=self.vae,
image_or_video=pose_video, image_or_video=img_ref_video,
patch_size=self.transformer.config.patch_size, patch_size=self.transformer.config.patch_size,
patch_size_t=self.transformer.config.patch_size_t, patch_size_t=self.transformer.config.patch_size_t,
device=device, device=device,
finetrainers/trainer.py
+4
View File
@@ -1359,6 +1359,8 @@ class Trainer:
frames = frames[: max_num_frames].float() frames = frames[: max_num_frames].float()
frames = frames.permute(0, 3, 1, 2).contiguous() frames = frames.permute(0, 3, 1, 2).contiguous()
frames = torch.stack([frame for frame in frames], dim=0) frames = torch.stack([frame for frame in frames], dim=0)
# add singleton dimension for downstream inference since no preprocessing took place
frames = frames.unsqueeze(0)
return frames return frames
@staticmethod @staticmethod
@@ -1381,4 +1383,6 @@ class Trainer:
# nearest_res = self._find_nearest_resolution(frames.shape[2], frames.shape[3]) # nearest_res = self._find_nearest_resolution(frames.shape[2], frames.shape[3])
# frames_resized = torch.stack([frame for frame in frames], dim=0) # frames_resized = torch.stack([frame for frame in frames], dim=0)
frames = torch.stack([frame for frame in frames], dim=0) frames = torch.stack([frame for frame in frames], dim=0)
# adding singleton dimension for downstream inference.
frames = frames.unsqueeze(0)
return frames return frames