Add temporal_downscale_ratio to LatentFormat (#13702)

Co-authored-by: ozbayb <17261091+ozbayb@users.noreply.github.com>
Co-authored-by: Alexis Rolland <alexisrolland@hotmail.com>
Co-authored-by: Jukka Seppänen <40791699+kijai@users.noreply.github.com>
Co-authored-by: Jedrzej Kosinski <kosinkadink1@gmail.com>
This commit is contained in:
drozbay 2026-05-05 17:30:00 -06:00 committed by GitHub
parent 89014792c9
commit 1655f8089a
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194

View File

@ -9,6 +9,7 @@ class LatentFormat:
latent_rgb_factors_reshape = None
taesd_decoder_name = None
spacial_downscale_ratio = 8
temporal_downscale_ratio = 1
def process_in(self, latent):
return latent * self.scale_factor
@ -235,6 +236,7 @@ class Flux2(LatentFormat):
class Mochi(LatentFormat):
latent_channels = 12
latent_dimensions = 3
temporal_downscale_ratio = 6
def __init__(self):
self.scale_factor = 1.0
@ -278,6 +280,7 @@ class LTXV(LatentFormat):
latent_channels = 128
latent_dimensions = 3
spacial_downscale_ratio = 32
temporal_downscale_ratio = 8
def __init__(self):
self.latent_rgb_factors = [
@ -421,6 +424,7 @@ class LTXAV(LTXV):
class HunyuanVideo(LatentFormat):
latent_channels = 16
latent_dimensions = 3
temporal_downscale_ratio = 4
scale_factor = 0.476986
latent_rgb_factors = [
[-0.0395, -0.0331, 0.0445],
@ -447,6 +451,7 @@ class HunyuanVideo(LatentFormat):
class Cosmos1CV8x8x8(LatentFormat):
latent_channels = 16
latent_dimensions = 3
temporal_downscale_ratio = 8
latent_rgb_factors = [
[ 0.1817, 0.2284, 0.2423],
@ -472,6 +477,7 @@ class Cosmos1CV8x8x8(LatentFormat):
class Wan21(LatentFormat):
latent_channels = 16
latent_dimensions = 3
temporal_downscale_ratio = 4
latent_rgb_factors = [
[-0.1299, -0.1692, 0.2932],
@ -734,6 +740,7 @@ class HunyuanVideo15(LatentFormat):
latent_channels = 32
latent_dimensions = 3
spacial_downscale_ratio = 16
temporal_downscale_ratio = 4
scale_factor = 1.03682
taesd_decoder_name = "lighttaehy1_5"
@ -788,6 +795,7 @@ class ZImagePixelSpace(ChromaRadiance):
class CogVideoX(LatentFormat):
latent_channels = 16
latent_dimensions = 3
temporal_downscale_ratio = 4
def __init__(self):
self.scale_factor = 1.15258426