Release: v0.2.1

README.md

@@ -123,7 +123,7 @@ The class provides functionality to compute previous image according to alpha, b
 **With `pip`**
     
 ```bash
-pip install --upgrade diffusers  # should install diffusers 0.2.0
+pip install --upgrade diffusers  # should install diffusers 0.2.1
 ```
 
 **With `conda`**

setup.py

@@ -181,7 +181,7 @@ install_requires = [
 
 setup(
     name="diffusers",
-    version="0.2.0",
+    version="0.2.1",
     description="Diffusers",
     long_description=open("README.md", "r", encoding="utf-8").read(),
     long_description_content_type="text/markdown",

src/diffusers/__init__.py

@@ -4,7 +4,7 @@
 from .utils import is_inflect_available, is_scipy_available, is_transformers_available, is_unidecode_available
 
 
-__version__ = "0.2.0"
+__version__ = "0.2.1"
 
 from .modeling_utils import ModelMixin
 from .models import AutoencoderKL, UNet2DConditionModel, UNet2DModel, VQModel

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion.py

@@ -96,6 +96,10 @@ class StableDiffusionPipeline(DiffusionPipeline):
 
         self.scheduler.set_timesteps(num_inference_steps, **extra_set_kwargs)
 
+        # if we use LMSDiscreteScheduler, let's make sure latents are mulitplied by sigmas
+        if isinstance(self.scheduler, LMSDiscreteScheduler):
+            latents = latents * self.scheduler.sigmas[0]
+
         # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
         # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
         # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
@@ -105,10 +109,6 @@ class StableDiffusionPipeline(DiffusionPipeline):
         if accepts_eta:
             extra_step_kwargs["eta"] = eta
 
-        self.scheduler.set_timesteps(num_inference_steps)
-        if isinstance(self.scheduler, LMSDiscreteScheduler):
-            latents = latents * self.scheduler.sigmas[0]
-
         for i, t in tqdm(enumerate(self.scheduler.timesteps)):
             # expand the latents if we are doing classifier free guidance
             latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents

src/diffusers/schedulers/scheduling_lms_discrete.py

@@ -36,8 +36,8 @@ class LMSDiscreteScheduler(SchedulerMixin, ConfigMixin):
         tensor_format="pt",
     ):
         """
-        Linear Multistep Scheduler for discrete beta schedules.
-        Based on the original k-diffusion implementation by Katherine Crowson:
+        Linear Multistep Scheduler for discrete beta schedules. Based on the original k-diffusion implementation by
+        Katherine Crowson:
         https://github.com/crowsonkb/k-diffusion/blob/481677d114f6ea445aa009cf5bd7a9cdee909e47/k_diffusion/sampling.py#L181
         """