Skip to content

Open-Sora: Democratizing Efficient Video Production for All

Python

fixed (#98)

Shen Chenhuib8d99d7a8b65
diff --git a/configs/opensora-v1-2/inference/sample.py b/configs/opensora-v1-2/inference/sample.py
index c46e4dd..65ccc1b 100644
--- a/configs/opensora-v1-2/inference/sample.py
+++ b/configs/opensora-v1-2/inference/sample.py
@@ -20,7 +20,9 @@ model = dict(
vae = dict(
type="VideoAutoencoderPipeline",
from_pretrained="pretrained_models/vae-v3",
- scale=2.5,
+ # scale=2.5,
+ shift=(-0.10, 0.34, 0.27, 0.98),
+ scale=(3.85, 2.32, 2.33, 3.06),
micro_frame_size=17,
vae_2d=dict(
type="VideoAutoencoderKL",
diff --git a/opensora/models/vae/vae.py b/opensora/models/vae/vae.py
index 8ac8729..2115e63 100644
--- a/opensora/models/vae/vae.py
+++ b/opensora/models/vae/vae.py
@@ -171,7 +171,7 @@ class VideoAutoencoderPipeline(nn.Module):
def decode(self, z, num_frames=None):
if not self.cal_loss:
- z = z * self.scale + self.shift
+ z = z * self.scale.to(z.dtype) + self.shift.to(z.dtype)
if self.micro_frame_size is None:
x_z = self.temporal_vae.decode(z, num_frames=num_frames)