# coding=utf-8
# Copyright 2026 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 random

import numpy as np
import PIL
import torch

from diffusers.image_processor import VaeImageProcessor
from diffusers.modular_pipelines import (
    FluxAutoBlocks,
    FluxKontextAutoBlocks,
    FluxKontextModularPipeline,
    FluxModularPipeline,
    ModularPipeline,
)

from ...testing_utils import floats_tensor, torch_device
from ..testing_utils import (
    BaseModularPipelineTesterConfig,
    ModularAutoOffloadTesterMixin,
    ModularLoadingTesterMixin,
    ModularMemoryTesterMixin,
    ModularPipelineTesterMixin,
    ModularWorkflowTesterMixin,
)


FLUX_TEXT2IMAGE_WORKFLOWS = {
    "text2image": [
        ("text_encoder", "FluxTextEncoderStep"),
        ("denoise.input", "FluxTextInputStep"),
        ("denoise.before_denoise.prepare_latents", "FluxPrepareLatentsStep"),
        ("denoise.before_denoise.set_timesteps", "FluxSetTimestepsStep"),
        ("denoise.before_denoise.prepare_rope_inputs", "FluxRoPEInputsStep"),
        ("denoise.denoise", "FluxDenoiseStep"),
        ("decode", "FluxDecodeStep"),
    ]
}


class FluxModularPipelineTesterConfig(BaseModularPipelineTesterConfig):
    pipeline_class = FluxModularPipeline
    pipeline_blocks_class = FluxAutoBlocks
    pretrained_model_name_or_path = "hf-internal-testing/tiny-flux-modular"

    params = frozenset(["prompt", "height", "width", "guidance_scale"])
    batch_params = frozenset(["prompt"])
    expected_workflow_blocks = FLUX_TEXT2IMAGE_WORKFLOWS

    def get_dummy_inputs(self, seed=0):
        generator = self.get_generator(seed)
        inputs = {
            "prompt": "A painting of a squirrel eating a burger",
            "generator": generator,
            "num_inference_steps": 2,
            "guidance_scale": 5.0,
            "height": 8,
            "width": 8,
            "max_sequence_length": 48,
            "output_type": "pt",
        }
        return inputs


class TestFluxModularPipelineFast(FluxModularPipelineTesterConfig, ModularPipelineTesterMixin):
    def test_float16_inference(self):
        super().test_float16_inference(9e-2)


class TestFluxModularPipelineLoading(FluxModularPipelineTesterConfig, ModularLoadingTesterMixin):
    pass


class TestFluxModularPipelineWorkflow(FluxModularPipelineTesterConfig, ModularWorkflowTesterMixin):
    pass


class TestFluxModularPipelineMemory(FluxModularPipelineTesterConfig, ModularMemoryTesterMixin):
    pass


class TestFluxModularPipelineAutoOffload(FluxModularPipelineTesterConfig, ModularAutoOffloadTesterMixin):
    pass


FLUX_IMAGE2IMAGE_WORKFLOWS = {
    "image2image": [
        ("text_encoder", "FluxTextEncoderStep"),
        ("vae_encoder.preprocess", "FluxProcessImagesInputStep"),
        ("vae_encoder.encode", "FluxVaeEncoderStep"),
        ("denoise.input.text_inputs", "FluxTextInputStep"),
        ("denoise.input.additional_inputs", "FluxAdditionalInputsStep"),
        ("denoise.before_denoise.prepare_latents", "FluxPrepareLatentsStep"),
        ("denoise.before_denoise.set_timesteps", "FluxImg2ImgSetTimestepsStep"),
        ("denoise.before_denoise.prepare_img2img_latents", "FluxImg2ImgPrepareLatentsStep"),
        ("denoise.before_denoise.prepare_rope_inputs", "FluxRoPEInputsStep"),
        ("denoise.denoise", "FluxDenoiseStep"),
        ("decode", "FluxDecodeStep"),
    ]
}


class FluxImg2ImgModularPipelineTesterConfig(BaseModularPipelineTesterConfig):
    pipeline_class = FluxModularPipeline
    pipeline_blocks_class = FluxAutoBlocks
    pretrained_model_name_or_path = "hf-internal-testing/tiny-flux-modular"

    params = frozenset(["prompt", "height", "width", "guidance_scale", "image"])
    batch_params = frozenset(["prompt", "image"])
    expected_workflow_blocks = FLUX_IMAGE2IMAGE_WORKFLOWS

    def get_pipeline(self, components_manager=None, dtype=torch.float32):
        pipeline = super().get_pipeline(components_manager, dtype)

        # Override `vae_scale_factor` here as currently, `image_processor` is initialized with
        # fixed constants instead of
        # https://github.com/huggingface/diffusers/blob/d54622c2679d700b425ad61abce9b80fc36212c0/src/diffusers/pipelines/flux/pipeline_flux_img2img.py#L230C9-L232C10
        pipeline.image_processor = VaeImageProcessor(vae_scale_factor=2)
        return pipeline

    def get_dummy_inputs(self, seed=0):
        generator = self.get_generator(seed)
        inputs = {
            "prompt": "A painting of a squirrel eating a burger",
            "generator": generator,
            "num_inference_steps": 4,
            "guidance_scale": 5.0,
            "height": 8,
            "width": 8,
            "max_sequence_length": 48,
            "output_type": "pt",
        }
        image = floats_tensor((1, 3, 32, 32), rng=random.Random(seed)).to(torch_device)
        image = image.cpu().permute(0, 2, 3, 1)[0]
        init_image = PIL.Image.fromarray(np.uint8(image)).convert("RGB")

        inputs["image"] = init_image
        inputs["strength"] = 0.5

        return inputs


class TestFluxImg2ImgModularPipelineFast(FluxImg2ImgModularPipelineTesterConfig, ModularPipelineTesterMixin):
    def test_float16_inference(self):
        super().test_float16_inference(8e-2)


class TestFluxImg2ImgModularPipelineLoading(FluxImg2ImgModularPipelineTesterConfig, ModularLoadingTesterMixin):
    def test_save_from_pretrained(self, tmp_path, base_pipe_output):
        base_pipe = self.get_pipeline().to(torch_device)
        base_pipe.save_pretrained(str(tmp_path))

        pipe = ModularPipeline.from_pretrained(tmp_path)
        pipe.load_components(dtype=torch.float32)
        pipe.to(torch_device)
        # Re-apply the `vae_scale_factor` override `get_pipeline` makes (see the comment there).
        pipe.image_processor = VaeImageProcessor(vae_scale_factor=2)

        image = pipe(**self.get_dummy_inputs(), output=self.output_name)

        expected_slice = base_pipe_output[0, -3:, -3:, -1].flatten()
        image_slice = image[0, -3:, -3:, -1].flatten()
        assert torch.abs(expected_slice - image_slice).max() < 1e-3


class TestFluxImg2ImgModularPipelineWorkflow(FluxImg2ImgModularPipelineTesterConfig, ModularWorkflowTesterMixin):
    pass


class TestFluxImg2ImgModularPipelineMemory(FluxImg2ImgModularPipelineTesterConfig, ModularMemoryTesterMixin):
    pass


FLUX_KONTEXT_WORKFLOWS = {
    "text2image": [
        ("text_encoder", "FluxTextEncoderStep"),
        ("denoise.input", "FluxTextInputStep"),
        ("denoise.before_denoise.prepare_latents", "FluxPrepareLatentsStep"),
        ("denoise.before_denoise.set_timesteps", "FluxSetTimestepsStep"),
        ("denoise.before_denoise.prepare_rope_inputs", "FluxRoPEInputsStep"),
        ("denoise.denoise", "FluxKontextDenoiseStep"),
        ("decode", "FluxDecodeStep"),
    ],
    "image_conditioned": [
        ("text_encoder", "FluxTextEncoderStep"),
        ("vae_encoder.preprocess", "FluxKontextProcessImagesInputStep"),
        ("vae_encoder.encode", "FluxVaeEncoderStep"),
        ("denoise.input.set_resolution", "FluxKontextSetResolutionStep"),
        ("denoise.input.text_inputs", "FluxTextInputStep"),
        ("denoise.input.additional_inputs", "FluxKontextAdditionalInputsStep"),
        ("denoise.before_denoise.prepare_latents", "FluxPrepareLatentsStep"),
        ("denoise.before_denoise.set_timesteps", "FluxSetTimestepsStep"),
        ("denoise.before_denoise.prepare_rope_inputs", "FluxKontextRoPEInputsStep"),
        ("denoise.denoise", "FluxKontextDenoiseStep"),
        ("decode", "FluxDecodeStep"),
    ],
}


class FluxKontextModularPipelineTesterConfig(BaseModularPipelineTesterConfig):
    pipeline_class = FluxKontextModularPipeline
    pipeline_blocks_class = FluxKontextAutoBlocks
    pretrained_model_name_or_path = "hf-internal-testing/tiny-flux-kontext-pipe"

    params = frozenset(["prompt", "height", "width", "guidance_scale", "image"])
    batch_params = frozenset(["prompt", "image"])
    expected_workflow_blocks = FLUX_KONTEXT_WORKFLOWS

    def get_dummy_inputs(self, seed=0):
        generator = self.get_generator(seed)
        inputs = {
            "prompt": "A painting of a squirrel eating a burger",
            "generator": generator,
            "num_inference_steps": 2,
            "guidance_scale": 5.0,
            "height": 8,
            "width": 8,
            "max_sequence_length": 48,
            "output_type": "pt",
        }
        image = PIL.Image.new("RGB", (32, 32), 0)

        inputs["image"] = image
        inputs["max_area"] = inputs["height"] * inputs["width"]
        inputs["_auto_resize"] = False

        return inputs


class TestFluxKontextModularPipelineFast(FluxKontextModularPipelineTesterConfig, ModularPipelineTesterMixin):
    def test_float16_inference(self):
        super().test_float16_inference(9e-2)


class TestFluxKontextModularPipelineLoading(FluxKontextModularPipelineTesterConfig, ModularLoadingTesterMixin):
    def test_save_from_pretrained(self, tmp_path, base_pipe_output):
        base_pipe = self.get_pipeline().to(torch_device)
        base_pipe.save_pretrained(str(tmp_path))

        pipe = ModularPipeline.from_pretrained(tmp_path)
        pipe.load_components(dtype=torch.float32)
        pipe.to(torch_device)
        pipe.image_processor = VaeImageProcessor(vae_scale_factor=2)

        image = pipe(**self.get_dummy_inputs(), output=self.output_name)

        expected_slice = base_pipe_output[0, -3:, -3:, -1].flatten()
        image_slice = image[0, -3:, -3:, -1].flatten()
        assert torch.abs(expected_slice - image_slice).max() < 1e-3


class TestFluxKontextModularPipelineWorkflow(FluxKontextModularPipelineTesterConfig, ModularWorkflowTesterMixin):
    pass


class TestFluxKontextModularPipelineMemory(FluxKontextModularPipelineTesterConfig, ModularMemoryTesterMixin):
    pass