comi-logic/Wan2GP
7
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Support for Hunyuan Video Avatar

Browse Source
This commit is contained in:
DeepBeepMeep
2025-06-05 15:35:58 +02:00
parent 7670af9610
commit 1976868f6a
15 changed files with 2562 additions and 81 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
hyvideo/diffusion/pipelines/pipeline_hunyuan_video.py
View File

@@ -1142,7 +1142,7 @@ class HunyuanVideoPipeline(DiffusionPipeline):
target_dtype = PRECISION_TO_TYPE[precision]
autocast_enabled = target_dtype != torch.float32 and not disable_autocast
vae_dtype = PRECISION_TO_TYPE[vae_precision]
vae_dtype = self.vae._model_dtype # PRECISION_TO_TYPE[vae_precision]
vae_autocast_enabled = vae_dtype != torch.float32 and not disable_autocast
# 7. Denoising loop
@@ -1262,6 +1262,7 @@ class HunyuanVideoPipeline(DiffusionPipeline):
guidance=guidance_expand,
pipeline=self,
x_id=j,
step_no=i,
callback = callback,
)
if self._interrupt:
@@ -1290,6 +1291,7 @@ class HunyuanVideoPipeline(DiffusionPipeline):
freqs_sin=freqs_cis[1], # [seqlen, head_dim]
guidance=guidance_expand,
pipeline=self,
step_no=i,
callback = callback,
)
if self._interrupt:
@@ -1404,7 +1406,6 @@ class HunyuanVideoPipeline(DiffusionPipeline):
else:
image = latents
image = (image / 2 + 0.5).clamp(0, 1)
# we always cast to float32 as this does not cause significant overhead and is compatible with bfloa16
image = image.cpu().float()