Fix/Enable all schedulers for in-painting (#1331)

* inpaint fix k lms

* onnox as well

* up

src/diffusers/pipelines/stable_diffusion/pipeline_onnx_stable_diffusion_inpaint.py

@@ -408,8 +408,8 @@ class OnnxStableDiffusionInpaintPipeline(DiffusionPipeline):
             # expand the latents if we are doing classifier free guidance
             latent_model_input = np.concatenate([latents] * 2) if do_classifier_free_guidance else latents
             # concat latents, mask, masked_image_latnets in the channel dimension
-            latent_model_input = np.concatenate([latent_model_input, mask, masked_image_latents], axis=1)
             latent_model_input = self.scheduler.scale_model_input(torch.from_numpy(latent_model_input), t)
+            latent_model_input = np.concatenate([latent_model_input, mask, masked_image_latents], axis=1)
             latent_model_input = latent_model_input.cpu().numpy()
 
             # predict the noise residual

src/diffusers/pipelines/stable_diffusion/pipeline_stable_diffusion_inpaint.py

@@ -586,9 +586,8 @@ class StableDiffusionInpaintPipeline(DiffusionPipeline):
             latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
 
             # concat latents, mask, masked_image_latents in the channel dimension
-            latent_model_input = torch.cat([latent_model_input, mask, masked_image_latents], dim=1)
-
             latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+            latent_model_input = torch.cat([latent_model_input, mask, masked_image_latents], dim=1)
 
             # predict the noise residual
             noise_pred = self.unet(latent_model_input, t, encoder_hidden_states=text_embeddings).sample

tests/pipelines/stable_diffusion/test_stable_diffusion_inpaint.py

@@ -22,6 +22,7 @@ import torch
 
 from diffusers import (
     AutoencoderKL,
+    LMSDiscreteScheduler,
     PNDMScheduler,
     StableDiffusionInpaintPipeline,
     UNet2DConditionModel,
@@ -421,6 +422,45 @@ class StableDiffusionInpaintPipelineIntegrationTests(unittest.TestCase):
         assert image.shape == (512, 512, 3)
         assert np.abs(expected_image - image).max() < 1e-2
 
+    def test_stable_diffusion_inpaint_pipeline_k_lms(self):
+        init_image = load_image(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
+            "/in_paint/overture-creations-5sI6fQgYIuo.png"
+        )
+        mask_image = load_image(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"
+            "/in_paint/overture-creations-5sI6fQgYIuo_mask.png"
+        )
+        expected_image = load_numpy(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/in_paint"
+            "/yellow_cat_sitting_on_a_park_bench_k_lms.npy"
+        )
+
+        model_id = "runwayml/stable-diffusion-inpainting"
+        pipe = StableDiffusionInpaintPipeline.from_pretrained(model_id, safety_checker=None)
+        pipe.to(torch_device)
+
+        # switch to LMS
+        pipe.scheduler = LMSDiscreteScheduler.from_config(pipe.scheduler.config)
+
+        pipe.set_progress_bar_config(disable=None)
+        pipe.enable_attention_slicing()
+
+        prompt = "Face of a yellow cat, high resolution, sitting on a park bench"
+
+        generator = torch.Generator(device=torch_device).manual_seed(0)
+        output = pipe(
+            prompt=prompt,
+            image=init_image,
+            mask_image=mask_image,
+            generator=generator,
+            output_type="np",
+        )
+        image = output.images[0]
+
+        assert image.shape == (512, 512, 3)
+        assert np.abs(expected_image - image).max() < 1e-2
+
     @unittest.skipIf(torch_device == "cpu", "This test is supposed to run on GPU")
     def test_stable_diffusion_pipeline_with_sequential_cpu_offloading(self):
         torch.cuda.empty_cache()