storytold/FineTrainers-Conditioning
2
0
Fork 0
mirror of https://github.com/storytold/FineTrainers-Conditioning.git synced 2026-10-29 09:07:10 +00:00
Code Issues Packages Projects Releases Wiki Activity

Simple conditioning sorta merges but checking inputs now.

Browse Source
This commit is contained in:
CrossProduct
2025-02-06 17:25:13 +00:00
parent d65958b181
commit 93e5069dc0
3 changed files with 9 additions and 82 deletions
Download Patch File Download Diff File Expand all files Collapse all files
finetrainers/conditioning/conditioned_pipeline.py
+5 -13
View File
@@ -213,31 +213,23 @@ class LTXConditionedPipeline(LTXPipeline):
# residual x latent
# add pose information to both channels
noisy_latents = noise_latent + pose_latent
pose_img_ref_latents = img_ref_latent + pose_latent
# expand channel information # B x 2C latent will be projected to adapter to scale it back to 128d using adapter.
condition_latent = torch.cat([pose_img_ref_latents,noisy_latents], dim=1)
noisy_latents = noise_latent + pose_latent + img_ref_latent
noise_latent_tokens = post_conditioned_latent_patchify(latents=noise_latent, num_frames=num_frames,
height=height,
width=width,
patch_size = 1,
patch_size_t = 1)["latents"]
# # expand channel information # B x 2C latent will be projected to adapter to scale it back to 128d using adapter.
# condition_latent = torch.cat([pose_img_ref_latents,noisy_latents], dim=1)
condition_tokens = post_conditioned_latent_patchify(latents=condition_latent,
condition_tokens = post_conditioned_latent_patchify(latents=noisy_latents,
num_frames=num_frames,
height=height,
width=width,
patch_size = 1,
patch_size_t = 1)["latents"]
# target video as a input residual
# noisy_residual_tokens = post_conditioned_latent_patchify(latents=noisy_latents,
# num_frames=num_frames,
# height=height,
# width=width,
# patch_size = 1,
# patch_size_t = 1)["latents"]
# Need to change the latents to
# 5. Prepare timesteps