Merge pull request kohya-ss#1187 from kohya-ss/fix-timeemb

fix sdxl timestep embedding

Author: kohya-ss

README.md

@@ -249,6 +249,16 @@ ControlNet-LLLite, a novel method for ControlNet with SDXL, is added. See [docum
 
 ## Change History
 
+### Mar 15, 2024 / 2024/3/15: v0.8.5
+
+- Fixed a bug that the value of timestep embedding during SDXL training was incorrect.
+  - The inference with the generation script is also fixed.
+  - The impact is unknown, but please update for SDXL training.
+
+- SDXL 学習時の timestep embedding の値が誤っていたのを修正しました。
+  - 生成スクリプトでの推論時についてもあわせて修正しました。
+  - 影響の度合いは不明ですが、SDXL の学習時にはアップデートをお願いいたします。
+
 ### Feb 24, 2024 / 2024/2/24: v0.8.4
 
 - The log output has been improved. PR [#905](https://github.com/kohya-ss/sd-scripts/pull/905) Thanks to shirayu!

library/sdxl_original_unet.py

@@ -31,8 +31,10 @@
 from torch.nn import functional as F
 from einops import rearrange
 from .utils import setup_logging
+
 setup_logging()
 import logging
+
 logger = logging.getLogger(__name__)
 
 IN_CHANNELS: int = 4
@@ -1074,7 +1076,7 @@ def forward(self, x, timesteps=None, context=None, y=None, **kwargs):
         timesteps = timesteps.expand(x.shape[0])
 
         hs = []
-        t_emb = get_timestep_embedding(timesteps, self.model_channels)  # , repeat_only=False)
+        t_emb = get_timestep_embedding(timesteps, self.model_channels, downscale_freq_shift=0)  # , repeat_only=False)
         t_emb = t_emb.to(x.dtype)
         emb = self.time_embed(t_emb)
 
@@ -1132,7 +1134,7 @@ def __init__(self, original_unet: SdxlUNet2DConditionModel, **kwargs):
     # call original model's methods
     def __getattr__(self, name):
         return getattr(self.delegate, name)
-    
+
     def __call__(self, *args, **kwargs):
         return self.delegate(*args, **kwargs)
 
@@ -1164,7 +1166,7 @@ def forward(self, x, timesteps=None, context=None, y=None, **kwargs):
         timesteps = timesteps.expand(x.shape[0])
 
         hs = []
-        t_emb = get_timestep_embedding(timesteps, _self.model_channels)  # , repeat_only=False)
+        t_emb = get_timestep_embedding(timesteps, _self.model_channels, downscale_freq_shift=0)  # , repeat_only=False)
         t_emb = t_emb.to(x.dtype)
         emb = _self.time_embed(t_emb)