Spaces:
Sleeping
Sleeping
Update ledits/pipeline_leditspp_stable_diffusion_xl.py
Browse files
ledits/pipeline_leditspp_stable_diffusion_xl.py
CHANGED
@@ -882,6 +882,8 @@ class LEditsPPPipelineStableDiffusionXL(
|
|
882 |
avg_diff_2 = None,
|
883 |
correlation_weight_factor = 0.7,
|
884 |
scale=2,
|
|
|
|
|
885 |
**kwargs,
|
886 |
):
|
887 |
r"""
|
@@ -1014,9 +1016,10 @@ class LEditsPPPipelineStableDiffusionXL(
|
|
1014 |
|
1015 |
eta = self.eta
|
1016 |
num_images_per_prompt = 1
|
1017 |
-
latents = self.init_latents
|
|
|
1018 |
|
1019 |
-
zs = self.zs
|
1020 |
self.scheduler.set_timesteps(len(self.scheduler.timesteps))
|
1021 |
|
1022 |
if use_intersect_mask:
|
@@ -1094,6 +1097,7 @@ class LEditsPPPipelineStableDiffusionXL(
|
|
1094 |
# self.scheduler.set_timesteps(num_inference_steps, device=device)
|
1095 |
|
1096 |
timesteps = self.inversion_steps
|
|
|
1097 |
t_to_idx = {int(v): k for k, v in enumerate(timesteps)}
|
1098 |
|
1099 |
if use_cross_attn_mask:
|
@@ -1698,7 +1702,8 @@ class LEditsPPPipelineStableDiffusionXL(
|
|
1698 |
if num_zero_noise_steps > 0:
|
1699 |
zs[-num_zero_noise_steps:] = torch.zeros_like(zs[-num_zero_noise_steps:])
|
1700 |
self.zs = zs
|
1701 |
-
return LEditsPPInversionPipelineOutput(images=resized, vae_reconstruction_images=image_rec)
|
|
|
1702 |
|
1703 |
|
1704 |
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.rescale_noise_cfg
|
|
|
882 |
avg_diff_2 = None,
|
883 |
correlation_weight_factor = 0.7,
|
884 |
scale=2,
|
885 |
+
init_latents: [torch.Tensor] = None,
|
886 |
+
zs: [torch.Tensor] = None,
|
887 |
**kwargs,
|
888 |
):
|
889 |
r"""
|
|
|
1016 |
|
1017 |
eta = self.eta
|
1018 |
num_images_per_prompt = 1
|
1019 |
+
#latents = self.init_latents
|
1020 |
+
latents = init_latents
|
1021 |
|
1022 |
+
#zs = self.zs
|
1023 |
self.scheduler.set_timesteps(len(self.scheduler.timesteps))
|
1024 |
|
1025 |
if use_intersect_mask:
|
|
|
1097 |
# self.scheduler.set_timesteps(num_inference_steps, device=device)
|
1098 |
|
1099 |
timesteps = self.inversion_steps
|
1100 |
+
timesteps = inversion_steps
|
1101 |
t_to_idx = {int(v): k for k, v in enumerate(timesteps)}
|
1102 |
|
1103 |
if use_cross_attn_mask:
|
|
|
1702 |
if num_zero_noise_steps > 0:
|
1703 |
zs[-num_zero_noise_steps:] = torch.zeros_like(zs[-num_zero_noise_steps:])
|
1704 |
self.zs = zs
|
1705 |
+
#return LEditsPPInversionPipelineOutput(images=resized, vae_reconstruction_images=image_rec)
|
1706 |
+
return xts[-1], zs
|
1707 |
|
1708 |
|
1709 |
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.rescale_noise_cfg
|