Improve pipeline_stable_diffusion_inpaint_legacy.py (huggingface#1585)

* update inpaint_legacy to allow the use of predicted noise to construct intermediate diffused images

* Update src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py

Co-authored-by: Patrick von Platen <[email protected]>

Co-authored-by: Patrick von Platen <[email protected]>

Author: cyber-meow

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint_legacy.py

@@ -430,6 +430,7 @@ def __call__(
         guidance_scale: Optional[float] = 7.5,
         negative_prompt: Optional[Union[str, List[str]]] = None,
         num_images_per_prompt: Optional[int] = 1,
+        add_predicted_noise: Optional[bool] = False,
         eta: Optional[float] = 0.0,
         generator: Optional[torch.Generator] = None,
         output_type: Optional[str] = "pil",
@@ -471,6 +472,9 @@ def __call__(
                 if `guidance_scale` is less than `1`).
             num_images_per_prompt (`int`, *optional*, defaults to 1):
                 The number of images to generate per prompt.
+            add_predicted_noise (`bool`, *optional*, defaults to True):
+                Use predicted noise instead of random noise when constructing noisy versions of the original image in
+                the reverse diffusion process
             eta (`float`, *optional*, defaults to 0.0):
                 Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
                 [`schedulers.DDIMScheduler`], will be ignored for others.
@@ -561,7 +565,12 @@ def __call__(
                 # compute the previous noisy sample x_t -> x_t-1
                 latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs).prev_sample
                 # masking
-                init_latents_proper = self.scheduler.add_noise(init_latents_orig, noise, torch.tensor([t]))
+                if add_predicted_noise:
+                    init_latents_proper = self.scheduler.add_noise(
+                        init_latents_orig, noise_pred_uncond, torch.tensor([t])
+                    )
+                else:
+                    init_latents_proper = self.scheduler.add_noise(init_latents_orig, noise, torch.tensor([t]))
 
                 latents = (init_latents_proper * mask) + (latents * (1 - mask))