Merge branch 'main' into introduce-cosxl-arg-ip2p

Co-authored-by: Yiyi Xu <yixu310@gmail.com>

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_instruct_pix2pix.py

@@ -169,6 +169,8 @@ class StableDiffusionXLInstructPix2PixPipeline(
             Whether to use the [invisible_watermark library](https://github.com/ShieldMnt/invisible-watermark/) to
             watermark output images. If not defined, it will default to True if the package is installed, otherwise no
             watermarker will be used.
+        is_cosxl_edit (`bool`, *optional*):
+            When set the image latents are scaled.
     """
 
     model_cpu_offload_seq = "text_encoder->text_encoder_2->unet->vae"
@@ -185,6 +187,7 @@ class StableDiffusionXLInstructPix2PixPipeline(
         scheduler: KarrasDiffusionSchedulers,
         force_zeros_for_empty_prompt: bool = True,
         add_watermarker: Optional[bool] = None,
+        is_cosxl_edit: Optional[bool] = False,
     ):
         super().__init__()
 
@@ -201,6 +204,7 @@ class StableDiffusionXLInstructPix2PixPipeline(
         self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
         self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
         self.default_sample_size = self.unet.config.sample_size
+        self.is_cosxl_edit = is_cosxl_edit
 
         add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
 
@@ -551,6 +555,9 @@ class StableDiffusionXLInstructPix2PixPipeline(
         if image_latents.dtype != self.vae.dtype:
             image_latents = image_latents.to(dtype=self.vae.dtype)
 
+        if self.is_cosxl_edit:
+            image_latents = image_latents * self.vae.config.scaling_factor
+
         return image_latents
 
     # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline._get_add_time_ids