add PAG support sd15 controlnet (#8820)

* add pag support sd15 controlnet * fix quality import * remove unecessary import * remove if state * fix tests * remove useless function * add sd1.5 controlnet pag docs --------- Co-authored-by: anhnct8 <anhnct8@fpt.com>
2026-01-27 17:22:53 +03:00 · 2024-07-12 17:12:56 +07:00
parent 9f963e7349
commit d704b3bf8c
8 changed files with 1605 additions and 0 deletions
--- a/docs/source/en/api/pipelines/pag.md
+++ b/docs/source/en/api/pipelines/pag.md
@@ -25,6 +25,11 @@ The abstract from the paper is:
 	- all
 	- __call__

+## StableDiffusionControlNetPAGPipeline
+[[autodoc]] StableDiffusionControlNetPAGPipeline
+	- all
+	- __call__
+
 ## StableDiffusionXLPAGPipeline
 [[autodoc]] StableDiffusionXLPAGPipeline
 	- all
--- a/src/diffusers/init.py
+++ b/src/diffusers/init.py
@@ -302,6 +302,7 @@ else:
            "StableDiffusionAttendAndExcitePipeline",
            "StableDiffusionControlNetImg2ImgPipeline",
            "StableDiffusionControlNetInpaintPipeline",
+            "StableDiffusionControlNetPAGPipeline",
            "StableDiffusionControlNetPipeline",
            "StableDiffusionControlNetXSPipeline",
            "StableDiffusionDepth2ImgPipeline",
@@ -713,6 +714,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
            StableDiffusionAttendAndExcitePipeline,
            StableDiffusionControlNetImg2ImgPipeline,
            StableDiffusionControlNetInpaintPipeline,
+            StableDiffusionControlNetPAGPipeline,
            StableDiffusionControlNetPipeline,
            StableDiffusionControlNetXSPipeline,
            StableDiffusionDepth2ImgPipeline,
--- a/src/diffusers/pipelines/init.py
+++ b/src/diffusers/pipelines/init.py
@@ -142,6 +142,7 @@ else:
    _import_structure["pag"].extend(
        [
            "StableDiffusionPAGPipeline",
+            "StableDiffusionControlNetPAGPipeline",
            "StableDiffusionXLPAGPipeline",
            "StableDiffusionXLPAGInpaintPipeline",
            "StableDiffusionXLControlNetPAGPipeline",
@@ -514,6 +515,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
        )
        from .musicldm import MusicLDMPipeline
        from .pag import (
+            StableDiffusionControlNetPAGPipeline,
            StableDiffusionPAGPipeline,
            StableDiffusionXLControlNetPAGPipeline,
            StableDiffusionXLPAGImg2ImgPipeline,
--- a/src/diffusers/pipelines/auto_pipeline.py
+++ b/src/diffusers/pipelines/auto_pipeline.py
@@ -47,6 +47,7 @@ from .kandinsky2_2 import (
 from .kandinsky3 import Kandinsky3Img2ImgPipeline, Kandinsky3Pipeline
 from .latent_consistency_models import LatentConsistencyModelImg2ImgPipeline, LatentConsistencyModelPipeline
 from .pag import (
+    StableDiffusionControlNetPAGPipeline,
    StableDiffusionPAGPipeline,
    StableDiffusionXLControlNetPAGPipeline,
    StableDiffusionXLPAGImg2ImgPipeline,
@@ -90,6 +91,7 @@ AUTO_TEXT2IMAGE_PIPELINES_MAPPING = OrderedDict(
        ("pixart-alpha", PixArtAlphaPipeline),
        ("pixart-sigma", PixArtSigmaPipeline),
        ("stable-diffusion-pag", StableDiffusionPAGPipeline),
+        ("stable-diffusion-controlnet-pag", StableDiffusionControlNetPAGPipeline),
        ("stable-diffusion-xl-pag", StableDiffusionXLPAGPipeline),
        ("stable-diffusion-xl-controlnet-pag", StableDiffusionXLControlNetPAGPipeline),
    ]
--- a/src/diffusers/pipelines/pag/init.py
+++ b/src/diffusers/pipelines/pag/init.py
@@ -22,6 +22,7 @@ except OptionalDependencyNotAvailable:

    _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
 else:
+    _import_structure["pipeline_pag_controlnet_sd"] = ["StableDiffusionControlNetPAGPipeline"]
    _import_structure["pipeline_pag_controlnet_sd_xl"] = ["StableDiffusionXLControlNetPAGPipeline"]
    _import_structure["pipeline_pag_sd"] = ["StableDiffusionPAGPipeline"]
    _import_structure["pipeline_pag_sd_xl"] = ["StableDiffusionXLPAGPipeline"]
@@ -36,6 +37,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
    except OptionalDependencyNotAvailable:
        from ...utils.dummy_torch_and_transformers_objects import *
    else:
+        from .pipeline_pag_controlnet_sd import StableDiffusionControlNetPAGPipeline
        from .pipeline_pag_controlnet_sd_xl import StableDiffusionXLControlNetPAGPipeline
        from .pipeline_pag_sd import StableDiffusionPAGPipeline
        from .pipeline_pag_sd_xl import StableDiffusionXLPAGPipeline
--- a/src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd.py
+++ b/src/diffusers/pipelines/pag/pipeline_pag_controlnet_sd.py
--- a/src/diffusers/utils/dummy_torch_and_transformers_objects.py
+++ b/src/diffusers/utils/dummy_torch_and_transformers_objects.py
@@ -1142,6 +1142,21 @@ class StableDiffusionControlNetInpaintPipeline(metaclass=DummyObject):
        requires_backends(cls, ["torch", "transformers"])


+class StableDiffusionControlNetPAGPipeline(metaclass=DummyObject):
+    _backends = ["torch", "transformers"]
+
+    def __init__(self, *args, **kwargs):
+        requires_backends(self, ["torch", "transformers"])
+
+    @classmethod
+    def from_config(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+    @classmethod
+    def from_pretrained(cls, *args, **kwargs):
+        requires_backends(cls, ["torch", "transformers"])
+
+
 class StableDiffusionControlNetPipeline(metaclass=DummyObject):
    _backends = ["torch", "transformers"]

--- a/tests/pipelines/pag/test_pag_controlnet_sd.py
+++ b/tests/pipelines/pag/test_pag_controlnet_sd.py
@@ -0,0 +1,248 @@
+# coding=utf-8
+# Copyright 2024 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 inspect
+import unittest
+
+import numpy as np
+import torch
+from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
+
+from diffusers import (
+    AutoencoderKL,
+    ControlNetModel,
+    DDIMScheduler,
+    StableDiffusionControlNetPAGPipeline,
+    StableDiffusionControlNetPipeline,
+    UNet2DConditionModel,
+)
+from diffusers.utils.testing_utils import (
+    enable_full_determinism,
+)
+from diffusers.utils.torch_utils import randn_tensor
+
+from ..pipeline_params import (
+    TEXT_TO_IMAGE_BATCH_PARAMS,
+    TEXT_TO_IMAGE_CALLBACK_CFG_PARAMS,
+    TEXT_TO_IMAGE_IMAGE_PARAMS,
+    TEXT_TO_IMAGE_PARAMS,
+)
+from ..test_pipelines_common import (
+    IPAdapterTesterMixin,
+    PipelineFromPipeTesterMixin,
+    PipelineLatentTesterMixin,
+    PipelineTesterMixin,
+)
+
+
+enable_full_determinism()
+
+
+class StableDiffusionControlNetPAGPipelineFastTests(
+    PipelineTesterMixin,
+    IPAdapterTesterMixin,
+    PipelineLatentTesterMixin,
+    PipelineFromPipeTesterMixin,
+    unittest.TestCase,
+):
+    pipeline_class = StableDiffusionControlNetPAGPipeline
+    params = TEXT_TO_IMAGE_PARAMS.union({"pag_scale", "pag_adaptive_scale"})
+    batch_params = TEXT_TO_IMAGE_BATCH_PARAMS
+    image_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    image_latents_params = TEXT_TO_IMAGE_IMAGE_PARAMS
+    callback_cfg_params = TEXT_TO_IMAGE_CALLBACK_CFG_PARAMS.union({"add_text_embeds", "add_time_ids"})
+
+    def get_dummy_components(self, time_cond_proj_dim=None):
+        # Copied from tests.pipelines.controlnet.test_controlnet_sdxl.StableDiffusionXLControlNetPipelineFastTests.get_dummy_components
+        torch.manual_seed(0)
+        unet = UNet2DConditionModel(
+            block_out_channels=(4, 8),
+            layers_per_block=2,
+            sample_size=32,
+            in_channels=4,
+            out_channels=4,
+            down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
+            up_block_types=("CrossAttnUpBlock2D", "UpBlock2D"),
+            cross_attention_dim=8,
+            time_cond_proj_dim=time_cond_proj_dim,
+            norm_num_groups=2,
+        )
+        torch.manual_seed(0)
+        controlnet = ControlNetModel(
+            block_out_channels=(4, 8),
+            layers_per_block=2,
+            in_channels=4,
+            down_block_types=("DownBlock2D", "CrossAttnDownBlock2D"),
+            conditioning_embedding_out_channels=(2, 4),
+            cross_attention_dim=8,
+            norm_num_groups=2,
+        )
+        torch.manual_seed(0)
+        scheduler = DDIMScheduler(
+            beta_start=0.00085,
+            beta_end=0.012,
+            beta_schedule="scaled_linear",
+            clip_sample=False,
+            set_alpha_to_one=False,
+        )
+        torch.manual_seed(0)
+        vae = AutoencoderKL(
+            block_out_channels=[4, 8],
+            in_channels=3,
+            out_channels=3,
+            down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
+            up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
+            latent_channels=4,
+            norm_num_groups=2,
+        )
+        torch.manual_seed(0)
+        text_encoder_config = CLIPTextConfig(
+            bos_token_id=0,
+            eos_token_id=2,
+            hidden_size=8,
+            intermediate_size=16,
+            layer_norm_eps=1e-05,
+            num_attention_heads=2,
+            num_hidden_layers=2,
+            pad_token_id=1,
+            vocab_size=1000,
+        )
+        text_encoder = CLIPTextModel(text_encoder_config)
+        tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip")
+
+        components = {
+            "unet": unet,
+            "controlnet": controlnet,
+            "scheduler": scheduler,
+            "vae": vae,
+            "text_encoder": text_encoder,
+            "tokenizer": tokenizer,
+            "safety_checker": None,
+            "feature_extractor": None,
+            "image_encoder": None,
+        }
+        return components
+
+    def get_dummy_inputs(self, device, seed=0):
+        if str(device).startswith("mps"):
+            generator = torch.manual_seed(seed)
+        else:
+            generator = torch.Generator(device=device).manual_seed(seed)
+
+        controlnet_embedder_scale_factor = 2
+        image = randn_tensor(
+            (1, 3, 32 * controlnet_embedder_scale_factor, 32 * controlnet_embedder_scale_factor),
+            generator=generator,
+            device=torch.device(device),
+        )
+
+        inputs = {
+            "prompt": "A painting of a squirrel eating a burger",
+            "generator": generator,
+            "num_inference_steps": 2,
+            "guidance_scale": 6.0,
+            "pag_scale": 3.0,
+            "output_type": "np",
+            "image": image,
+        }
+
+        return inputs
+
+    def test_pag_disable_enable(self):
+        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
+        components = self.get_dummy_components()
+
+        # base  pipeline (expect same output when pag is disabled)
+        pipe_sd = StableDiffusionControlNetPipeline(**components)
+        pipe_sd = pipe_sd.to(device)
+        pipe_sd.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        del inputs["pag_scale"]
+        assert (
+            "pag_scale" not in inspect.signature(pipe_sd.__call__).parameters
+        ), f"`pag_scale` should not be a call parameter of the base pipeline {pipe_sd.__calss__.__name__}."
+        out = pipe_sd(**inputs).images[0, -3:, -3:, -1]
+
+        # pag disabled with pag_scale=0.0
+        pipe_pag = self.pipeline_class(**components)
+        pipe_pag = pipe_pag.to(device)
+        pipe_pag.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        inputs["pag_scale"] = 0.0
+        out_pag_disabled = pipe_pag(**inputs).images[0, -3:, -3:, -1]
+
+        # pag enabled
+        pipe_pag = self.pipeline_class(**components, pag_applied_layers=["mid", "up", "down"])
+        pipe_pag = pipe_pag.to(device)
+        pipe_pag.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        out_pag_enabled = pipe_pag(**inputs).images[0, -3:, -3:, -1]
+
+        assert np.abs(out.flatten() - out_pag_disabled.flatten()).max() < 1e-3
+        assert np.abs(out.flatten() - out_pag_enabled.flatten()).max() > 1e-3
+
+    def test_pag_cfg(self):
+        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
+        components = self.get_dummy_components()
+
+        pipe_pag = self.pipeline_class(**components, pag_applied_layers=["mid", "up", "down"])
+        pipe_pag = pipe_pag.to(device)
+        pipe_pag.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        image = pipe_pag(**inputs).images
+        image_slice = image[0, -3:, -3:, -1]
+
+        assert image.shape == (
+            1,
+            64,
+            64,
+            3,
+        ), f"the shape of the output image should be (1, 64, 64, 3) but got {image.shape}"
+        expected_slice = np.array(
+            [0.45505235, 0.2785938, 0.16334778, 0.79689944, 0.53095645, 0.40135607, 0.7052706, 0.69065094, 0.41548574]
+        )
+
+        max_diff = np.abs(image_slice.flatten() - expected_slice).max()
+        assert max_diff < 1e-3, f"output is different from expected, {image_slice.flatten()}"
+
+    def test_pag_uncond(self):
+        device = "cpu"  # ensure determinism for the device-dependent torch.Generator
+        components = self.get_dummy_components()
+
+        pipe_pag = self.pipeline_class(**components, pag_applied_layers=["mid", "up", "down"])
+        pipe_pag = pipe_pag.to(device)
+        pipe_pag.set_progress_bar_config(disable=None)
+
+        inputs = self.get_dummy_inputs(device)
+        inputs["guidance_scale"] = 0.0
+        image = pipe_pag(**inputs).images
+        image_slice = image[0, -3:, -3:, -1]
+
+        assert image.shape == (
+            1,
+            64,
+            64,
+            3,
+        ), f"the shape of the output image should be (1, 64, 64, 3) but got {image.shape}"
+        expected_slice = np.array(
+            [0.45127502, 0.2797252, 0.15970308, 0.7993157, 0.5414344, 0.40160775, 0.7114598, 0.69803864, 0.4217583]
+        )
+
+        max_diff = np.abs(image_slice.flatten() - expected_slice).max()
+        assert max_diff < 1e-3, f"output is different from expected, {image_slice.flatten()}"