Add img2img test

src/diffusers/pipelines/stable_diffusion_xl/pipeline_stable_diffusion_xl_img2img.py

@@ -786,7 +786,7 @@ class StableDiffusionXLImg2ImgPipeline(DiffusionPipeline):
 
         prompt_embeds = prompt_embeds.to(device)
         add_text_embeds = add_text_embeds.to(device)
-        add_time_ids = add_time_ids.to(device).repeat(num_images_per_prompt, 1)
+        add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1)
 
         # 9. Denoising loop
         num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order

tests/pipelines/stable_diffusion_xl/test_stable_diffusion_xl.py

@@ -154,7 +154,7 @@ class StableDiffusionXLPipelineFastTests(PipelineLatentTesterMixin, PipelineTest
 
 @slow
 @require_torch_gpu
-class StableDiffusion2PipelineSlowTests(unittest.TestCase):
+class StableDiffusionXLPipelineSlowTests(unittest.TestCase):
     def tearDown(self):
         super().tearDown()
         gc.collect()
@@ -175,7 +175,7 @@ class StableDiffusion2PipelineSlowTests(unittest.TestCase):
         return inputs
 
     def test_stable_diffusion_default_euler(self):
-        pipe = StableDiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-base")
+        pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-base")
         pipe.to(torch_device)
         pipe.set_progress_bar_config(disable=None)
 

tests/pipelines/stable_diffusion_xl/test_stable_diffusion_xl_img2img.py

@@ -0,0 +1,194 @@
+# coding=utf-8
+# Copyright 2023 HuggingFace Inc.
+#
+# Licensed under the Apache License, Version 2.0 (the "License");
+# you may not use this file except in compliance with the License.
+# You may obtain a copy of the License at
+#
+#     http://www.apache.org/licenses/LICENSE-2.0
+#
+# Unless required by applicable law or agreed to in writing, software
+# distributed under the License is distributed on an "AS IS" BASIS,
+# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+# See the License for the specific language governing permissions and
+# limitations under the License.
+
+import gc
+import unittest
+import random
+
+import numpy as np
+import torch
+from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokenizer, CLIPTextConfig
+
+from diffusers import (
+    AutoencoderKL,
+    EulerDiscreteScheduler,
+    StableDiffusionXLImg2ImgPipeline,
+    UNet2DConditionModel,
+)
+from diffusers.utils import floats_tensor, load_image, load_numpy, nightly, slow, torch_device
+from diffusers.utils.testing_utils import CaptureLogger, enable_full_determinism, require_torch_gpu
+
+from ..pipeline_params import (
+    IMAGE_TO_IMAGE_IMAGE_PARAMS,
+    TEXT_GUIDED_IMAGE_VARIATION_BATCH_PARAMS,
+    TEXT_GUIDED_IMAGE_VARIATION_PARAMS,
+)
+from ..test_pipelines_common import PipelineLatentTesterMixin, PipelineTesterMixin
+
+
+enable_full_determinism()
+
+
+class StableDiffusionXLImg2ImgPipelineFastTests(PipelineLatentTesterMixin, PipelineTesterMixin, unittest.TestCase):
+    pipeline_class = StableDiffusionXLImg2ImgPipeline
+    params = TEXT_GUIDED_IMAGE_VARIATION_PARAMS - {"height", "width"}
+    required_optional_params = PipelineTesterMixin.required_optional_params - {"latents"}
+    batch_params = TEXT_GUIDED_IMAGE_VARIATION_BATCH_PARAMS
+    image_params = IMAGE_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = IMAGE_TO_IMAGE_IMAGE_PARAMS
+
+    def get_dummy_components(self):
+        torch.manual_seed(0)
+        unet = UNet2DConditionModel(
+            block_out_channels=(32, 64),
+            layers_per_block=2,
+            sample_size=32,
+            in_channels=4,
+            out_channels=4,
+            down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
+            up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
+            # SD2-specific config below
+            attention_head_dim=(2, 4),
+            use_linear_projection=True,
+            addition_embed_type="text_time",
+            addition_time_embed_dim=8,
+            num_transformer_blocks=(1, 2),
+            projection_class_embeddings_input_dim=80, # 6 * 8 + 32
+            cross_attention_dim=64,
+        )
+        scheduler = EulerDiscreteScheduler(
+            beta_start=0.00085,
+            beta_end=0.012,
+            beta_schedule="scaled_linear",
+            timestep_spacing="leading",
+        )
+        torch.manual_seed(0)
+        vae = AutoencoderKL(
+            block_out_channels=[32, 64],
+            in_channels=3,
+            out_channels=3,
+            down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
+            up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
+            latent_channels=4,
+            sample_size=128,
+        )
+        torch.manual_seed(0)
+        text_encoder_config = CLIPTextConfig(
+            bos_token_id=0,
+            eos_token_id=2,
+            hidden_size=32,
+            intermediate_size=37,
+            layer_norm_eps=1e-05,
+            num_attention_heads=4,
+            num_hidden_layers=5,
+            pad_token_id=1,
+            vocab_size=1000,
+            # SD2-specific config below
+            hidden_act="gelu",
+            projection_dim=32,
+        )
+        text_encoder = CLIPTextModel(text_encoder_config)
+        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip", local_files_only=True)
+
+        text_encoder_2 = CLIPTextModelWithProjection(text_encoder_config)
+        tokenizer_2 = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip", local_files_only=True)
+
+        components = {
+            "unet": unet,
+            "scheduler": scheduler,
+            "vae": vae,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+            "text_encoder_2": text_encoder_2,
+            "tokenizer_2": tokenizer_2,
+            # "safety_checker": None,
+            # "feature_extractor": None,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        image = floats_tensor((1, 3, 32, 32), rng=random.Random(seed)).to(device)
+        image = image / 2 + 0.5
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+        inputs = {
+            "prompt": "A painting of a squirrel eating a burger",
+            "image": image,
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 5.0,
+            "output_type": "numpy",
+        }
+        return inputs
+
+    def test_stable_diffusion_xl_img2img_euler(self):
+        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
+        components = self.get_dummy_components()
+        sd_pipe = StableDiffusionXLImg2ImgPipeline(**components)
+        sd_pipe = sd_pipe.to(device)
+        sd_pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        image = sd_pipe(**inputs).images
+        image_slice = image[0, -3:, -3:, -1]
+
+        assert image.shape == (1, 64, 64, 3)
+        expected_slice = np.array([0.5753, 0.6113, 0.5005, 0.5036, 0.5464, 0.4725, 0.4982, 0.4865, 0.4861])
+
+        assert np.abs(image_slice.flatten() - expected_slice).max() < 1e-2
+
+    def test_attention_slicing_forward_pass(self):
+        super().test_attention_slicing_forward_pass(expected_max_diff=3e-3)
+
+    def test_inference_batch_single_identical(self):
+        super().test_inference_batch_single_identical(expected_max_diff=3e-3)
+
+
+@slow
+@require_torch_gpu
+class StableDiffusionXLImg2ImgPipelineSlowTests(unittest.TestCase):
+    def tearDown(self):
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def get_inputs(self, device, generator_device="cpu", dtype=torch.float32, seed=0):
+        generator = torch.Generator(device=generator_device).manual_seed(seed)
+        latents = np.random.RandomState(seed).standard_normal((1, 4, 64, 64))
+        latents = torch.from_numpy(latents).to(device=device, dtype=dtype)
+        inputs = {
+            "prompt": "a photograph of an astronaut riding a horse",
+            "latents": latents,
+            "generator": generator,
+            "num_inference_steps": 3,
+            "guidance_scale": 7.5,
+            "output_type": "numpy",
+        }
+        return inputs
+
+    def test_stable_diffusion_default_euler(self):
+        pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-base")
+        pipe.to(torch_device)
+        pipe.set_progress_bar_config(disable=None)
+
+        inputs = self.get_inputs(torch_device)
+        image = pipe(**inputs).images
+        image_slice = image[0, -3:, -3:, -1].flatten()
+
+        assert image.shape == (1, 512, 512, 3)
+        expected_slice = np.array([0.49493, 0.47896, 0.40798, 0.54214, 0.53212, 0.48202, 0.47656, 0.46329, 0.48506])
+        assert np.abs(image_slice - expected_slice).max() < 7e-3