diff --git a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py
index 66f33a65e8da..8d1646e4d887 100644
--- a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py
+++ b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl.py
@@ -1193,7 +1193,16 @@ def __call__(
                     noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=self.guidance_rescale)
 
                 # compute the previous noisy sample x_t -> x_t-1
+                latents_dtype = latents.dtype
                 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+                if latents.dtype != latents_dtype:
+                    if torch.backends.mps.is_available():
+                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                        latents = latents.to(latents_dtype)
+                    else:
+                        raise ValueError(
+                            "For the given accelerator, there seems to be an unexpected problem in type-casting. Please file an issue on the PyTorch GitHub repository. See also: https://github.com/huggingface/diffusers/pull/7446/."
+                        )
 
                 if callback_on_step_end is not None:
                     callback_kwargs = {}
@@ -1228,6 +1237,14 @@ def __call__(
             if needs_upcasting:
                 self.upcast_vae()
                 latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+            elif latents.dtype != self.vae.dtype:
+                if torch.backends.mps.is_available():
+                    # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                    self.vae = self.vae.to(latents.dtype)
+                else:
+                    raise ValueError(
+                        "For the given accelerator, there seems to be an unexpected problem in type-casting. Please file an issue on the PyTorch GitHub repository. See also: https://github.com/huggingface/diffusers/pull/7446/."
+                    )
 
             # unscale/denormalize the latents
             # denormalize with the mean and std if available and not None
diff --git a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py
index bb9101d2384b..af9da5073e06 100644
--- a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py
+++ b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py
@@ -1370,7 +1370,16 @@ def denoising_value_valid(dnv):
                     noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=self.guidance_rescale)
 
                 # compute the previous noisy sample x_t -> x_t-1
+                latents_dtype = latents.dtype
                 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+                if latents.dtype != latents_dtype:
+                    if torch.backends.mps.is_available():
+                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                        latents = latents.to(latents_dtype)
+                    else:
+                        raise ValueError(
+                            "For the given accelerator, there seems to be an unexpected problem in type-casting. Please file an issue on the PyTorch GitHub repository. See also: https://github.com/huggingface/diffusers/pull/7446/."
+                        )
 
                 if callback_on_step_end is not None:
                     callback_kwargs = {}
@@ -1405,6 +1414,14 @@ def denoising_value_valid(dnv):
             if needs_upcasting:
                 self.upcast_vae()
                 latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+            elif latents.dtype != self.vae.dtype:
+                if torch.backends.mps.is_available():
+                    # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                    self.vae = self.vae.to(latents.dtype)
+                else:
+                    raise ValueError(
+                        "For the given accelerator, there seems to be an unexpected problem in type-casting. Please file an issue on the PyTorch GitHub repository. See also: https://github.com/huggingface/diffusers/pull/7446/."
+                    )
 
             # unscale/denormalize the latents
             # denormalize with the mean and std if available and not None
diff --git a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py
index 49388a4f1db7..c9a72ccda985 100644
--- a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py
+++ b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_inpaint.py
@@ -1720,7 +1720,16 @@ def denoising_value_valid(dnv):
                     noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=self.guidance_rescale)
 
                 # compute the previous noisy sample x_t -> x_t-1
+                latents_dtype = latents.dtype
                 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+                if latents.dtype != latents_dtype:
+                    if torch.backends.mps.is_available():
+                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                        latents = latents.to(latents_dtype)
+                    else:
+                        raise ValueError(
+                            "For the given accelerator, there seems to be an unexpected problem in type-casting. Please file an issue on the PyTorch GitHub repository. See also: https://github.com/huggingface/diffusers/pull/7446/."
+                        )
 
                 if num_channels_unet == 4:
                     init_latents_proper = image_latents
@@ -1772,6 +1781,14 @@ def denoising_value_valid(dnv):
             if needs_upcasting:
                 self.upcast_vae()
                 latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+            elif latents.dtype != self.vae.dtype:
+                if torch.backends.mps.is_available():
+                    # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                    self.vae = self.vae.to(latents.dtype)
+                else:
+                    raise ValueError(
+                        "For the given accelerator, there seems to be an unexpected problem in type-casting. Please file an issue on the PyTorch GitHub repository. See also: https://github.com/huggingface/diffusers/pull/7446/."
+                    )
 
             # unscale/denormalize the latents
             # denormalize with the mean and std if available and not None
diff --git a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py
index 026854d5c40f..9aedb8667587 100644
--- a/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py
+++ b/src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py
@@ -918,7 +918,16 @@ def __call__(
                     noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
 
                 # compute the previous noisy sample x_t -> x_t-1
+                latents_dtype = latents.dtype
                 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
+                if latents.dtype != latents_dtype:
+                    if torch.backends.mps.is_available():
+                        # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                        latents = latents.to(latents_dtype)
+                    else:
+                        raise ValueError(
+                            "For the given accelerator, there seems to be an unexpected problem in type-casting. Please file an issue on the PyTorch GitHub repository. See also: https://github.com/huggingface/diffusers/pull/7446/."
+                        )
 
                 # call the callback, if provided
                 if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0):
@@ -937,6 +946,14 @@ def __call__(
             if needs_upcasting:
                 self.upcast_vae()
                 latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+            elif latents.dtype != self.vae.dtype:
+                if torch.backends.mps.is_available():
+                    # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
+                    self.vae = self.vae.to(latents.dtype)
+                else:
+                    raise ValueError(
+                        "For the given accelerator, there seems to be an unexpected problem in type-casting. Please file an issue on the PyTorch GitHub repository. See also: https://github.com/huggingface/diffusers/pull/7446/."
+                    )
 
             # unscale/denormalize the latents
             # denormalize with the mean and std if available and not None