denoise.py

# Copyright 2026 Krea AI and The HuggingFace Team. All rights reserved.
#
# 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 torch

from diffusers.configuration_utils import FrozenDict
from diffusers.guiders import ClassifierFreeGuidance
from .transformer_krea2 import Krea2Transformer2DModel
from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
from diffusers.utils import logging
from diffusers.modular_pipelines.modular_pipeline import BlockState, LoopSequentialPipelineBlocks, ModularPipelineBlocks, PipelineState
from diffusers.modular_pipelines.modular_pipeline_utils import ComponentSpec, InputParam, OutputParam
from .modular_pipeline import Krea2ModularPipeline


logger = logging.get_logger(__name__)

# ====================
# 1. LOOP STEPS (run at each denoising step)
# ====================


# loop step:before denoiser
class Krea2LoopBeforeDenoiser(ModularPipelineBlocks):
    model_name = "krea2"

    @property
    def description(self) -> str:
        return (
            "step within the denoising loop that prepares the latent input for the denoiser. "
            "This block should be used to compose the `sub_blocks` attribute of a `LoopSequentialPipelineBlocks` "
            "object (e.g. `Krea2DenoiseLoopWrapper`)"
        )

    @property
    def inputs(self) -> list[InputParam]:
        return [
            InputParam(
                name="latents",
                required=True,
                type_hint=torch.Tensor,
                description="The initial latents to use for the denoising process. Can be generated in prepare_latent step.",
            ),
        ]

    @torch.no_grad()
    def __call__(self, components: Krea2ModularPipeline, block_state: BlockState, i: int, t: torch.Tensor):
        # one timestep
        block_state.timestep = t.expand(block_state.latents.shape[0]).to(block_state.latents.dtype)
        block_state.latent_model_input = block_state.latents
        return components, block_state


# loop step:before denoiser (edit) -- appends the clean reference tokens to the denoiser input each step
class Krea2EditLoopBeforeDenoiser(ModularPipelineBlocks):
    model_name = "krea2"

    @property
    def description(self) -> str:
        return (
            "step within the denoising loop that prepares the latent input for the edit denoiser: it appends the "
            "packed clean reference tokens after the noisy image tokens. This block should be used to compose the "
            "`sub_blocks` attribute of a `LoopSequentialPipelineBlocks` object (e.g. `Krea2EditDenoiseStep`)."
        )

    @property
    def inputs(self) -> list[InputParam]:
        return [
            InputParam(
                name="latents",
                required=True,
                type_hint=torch.Tensor,
                description="The initial latents to use for the denoising process. Can be generated in prepare_latent step.",
            ),
            InputParam(
                name="reference_latents",
                required=True,
                type_hint=torch.Tensor,
                description="Packed clean reference tokens to append to the denoiser sequence. Can be generated in the reference latents step.",
            ),
        ]

    @torch.no_grad()
    def __call__(self, components: Krea2ModularPipeline, block_state: BlockState, i: int, t: torch.Tensor):
        block_state.timestep = t.expand(block_state.latents.shape[0]).to(block_state.latents.dtype)
        # Reference tokens are shared across the batch; expand and append them after the noisy image tokens.
        reference_latents = block_state.reference_latents.expand(block_state.latents.shape[0], -1, -1)
        block_state.latent_model_input = torch.cat(
            [block_state.latents, reference_latents.to(block_state.latents.dtype)], dim=1
        )
        return components, block_state


# loop step:denoiser
class Krea2LoopDenoiser(ModularPipelineBlocks):
    model_name = "krea2"

    @property
    def description(self) -> str:
        return (
            "step within the denoising loop that denoise the latent input for the denoiser. "
            "This block should be used to compose the `sub_blocks` attribute of a `LoopSequentialPipelineBlocks` "
            "object (e.g. `Krea2DenoiseLoopWrapper`)"
        )

    @property
    def expected_components(self) -> list[ComponentSpec]:
        return [
            ComponentSpec(
                "guider",
                ClassifierFreeGuidance,
                config=FrozenDict({"guidance_scale": 4.5, "use_original_formulation": True}),
                default_creation_method="from_config",
            ),
            ComponentSpec("transformer", Krea2Transformer2DModel),
        ]

    @property
    def inputs(self) -> list[InputParam]:
        return [
            InputParam.template("denoiser_input_fields"),
            InputParam(
                "position_ids",
                required=True,
                type_hint=torch.Tensor,
                description="The rotary coordinates for the combined text-image sequence. Can be generated in prepare_rope_inputs step.",
            ),
        ]

    @torch.no_grad()
    def __call__(self, components: Krea2ModularPipeline, block_state: BlockState, i: int, t: torch.Tensor):
        guider_inputs = {
            "encoder_hidden_states": (
                getattr(block_state, "prompt_embeds", None),
                getattr(block_state, "negative_prompt_embeds", None),
            ),
            "encoder_attention_mask": (
                getattr(block_state, "prompt_embeds_mask", None),
                getattr(block_state, "negative_prompt_embeds_mask", None),
            ),
        }

        transformer_args = set(inspect.signature(components.transformer.forward).parameters.keys())
        additional_cond_kwargs = {}
        for field_name, field_value in block_state.denoiser_input_fields.items():
            if field_name in transformer_args and field_name not in guider_inputs:
                additional_cond_kwargs[field_name] = field_value
        block_state.additional_cond_kwargs.update(additional_cond_kwargs)

        components.guider.set_state(step=i, num_inference_steps=block_state.num_inference_steps, timestep=t)
        guider_state = components.guider.prepare_inputs(guider_inputs)

        for guider_state_batch in guider_state:
            components.guider.prepare_models(components.transformer)
            cond_kwargs = {input_name: getattr(guider_state_batch, input_name) for input_name in guider_inputs.keys()}

            guider_state_batch.noise_pred = components.transformer(
                hidden_states=block_state.latent_model_input,
                timestep=block_state.timestep / 1000,
                return_dict=False,
                **cond_kwargs,
                **block_state.additional_cond_kwargs,
            )[0]

            components.guider.cleanup_models(components.transformer)

        guider_output = components.guider(guider_state)
        block_state.noise_pred = guider_output.pred

        return components, block_state


# loop step:after denoiser
class Krea2LoopAfterDenoiser(ModularPipelineBlocks):
    model_name = "krea2"

    @property
    def description(self) -> str:
        return (
            "step within the denoising loop that updates the latents. "
            "This block should be used to compose the `sub_blocks` attribute of a `LoopSequentialPipelineBlocks` "
            "object (e.g. `Krea2DenoiseLoopWrapper`)"
        )

    @property
    def expected_components(self) -> list[ComponentSpec]:
        return [
            ComponentSpec("scheduler", FlowMatchEulerDiscreteScheduler),
        ]

    @property
    def intermediate_outputs(self) -> list[OutputParam]:
        return [
            OutputParam.template("latents"),
        ]

    @torch.no_grad()
    def __call__(self, components: Krea2ModularPipeline, block_state: BlockState, i: int, t: torch.Tensor):
        latents_dtype = block_state.latents.dtype
        block_state.latents = components.scheduler.step(
            block_state.noise_pred,
            t,
            block_state.latents,
            return_dict=False,
        )[0]

        if block_state.latents.dtype != latents_dtype:
            if torch.backends.mps.is_available():
                # some platforms (eg. apple mps) misbehave due to a pytorch bug: https://github.com/pytorch/pytorch/pull/99272
                block_state.latents = block_state.latents.to(latents_dtype)

        return components, block_state


class Krea2LoopAfterDenoiserInpaint(ModularPipelineBlocks):
    model_name = "krea2"

    @property
    def description(self) -> str:
        return (
            "step within the denoising loop that updates the latents using mask and image_latents for inpainting. "
            "This block should be used to compose the `sub_blocks` attribute of a `LoopSequentialPipelineBlocks` "
            "object (e.g. `Krea2DenoiseLoopWrapper`)"
        )

    @property
    def inputs(self) -> list[InputParam]:
        return [
            InputParam(
                "mask",
                required=True,
                type_hint=torch.Tensor,
                description="The mask to use for the inpainting process. Can be generated in inpaint prepare latents step.",
            ),
            InputParam.template("image_latents"),
            InputParam(
                "initial_noise",
                required=True,
                type_hint=torch.Tensor,
                description="The initial noise to use for the inpainting process. Can be generated in inpaint prepare latents step.",
            ),
        ]

    @property
    def intermediate_outputs(self) -> list[OutputParam]:
        return [
            OutputParam.template("latents"),
        ]

    @torch.no_grad()
    def __call__(self, components: Krea2ModularPipeline, block_state: BlockState, i: int, t: torch.Tensor):
        block_state.init_latents_proper = block_state.image_latents
        if i < len(block_state.timesteps) - 1:
            block_state.noise_timestep = block_state.timesteps[i + 1]
            block_state.init_latents_proper = components.scheduler.scale_noise(
                block_state.init_latents_proper, torch.tensor([block_state.noise_timestep]), block_state.initial_noise
            )

        block_state.latents = (
            1 - block_state.mask
        ) * block_state.init_latents_proper + block_state.mask * block_state.latents

        return components, block_state


# ====================
# 2. DENOISE LOOP WRAPPER: define the denoising loop logic
# ====================
class Krea2DenoiseLoopWrapper(LoopSequentialPipelineBlocks):
    model_name = "krea2"

    @property
    def description(self) -> str:
        return (
            "Pipeline block that iteratively denoise the latents over `timesteps`. "
            "The specific steps with each iteration can be customized with `sub_blocks` attributes"
        )

    @property
    def loop_expected_components(self) -> list[ComponentSpec]:
        return [
            ComponentSpec("scheduler", FlowMatchEulerDiscreteScheduler),
        ]

    @property
    def loop_inputs(self) -> list[InputParam]:
        return [
            InputParam(
                name="timesteps",
                required=True,
                type_hint=torch.Tensor,
                description="The timesteps to use for the denoising process. Can be generated in set_timesteps step.",
            ),
            InputParam.template("num_inference_steps", required=True),
        ]

    @torch.no_grad()
    def __call__(self, components: Krea2ModularPipeline, state: PipelineState) -> PipelineState:
        block_state = self.get_block_state(state)

        block_state.num_warmup_steps = max(
            len(block_state.timesteps) - block_state.num_inference_steps * components.scheduler.order, 0
        )

        block_state.additional_cond_kwargs = {}

        with self.progress_bar(total=block_state.num_inference_steps) as progress_bar:
            for i, t in enumerate(block_state.timesteps):
                components, block_state = self.loop_step(components, block_state, i=i, t=t)
                if i == len(block_state.timesteps) - 1 or (
                    (i + 1) > block_state.num_warmup_steps and (i + 1) % components.scheduler.order == 0
                ):
                    progress_bar.update()

        self.set_block_state(state, block_state)

        return components, state


# ====================
# 3. DENOISE STEPS: compose the denoising loop with loop wrapper + loop steps
# ====================


# Krea 2 (text2image, image2image)
class Krea2DenoiseStep(Krea2DenoiseLoopWrapper):
    model_name = "krea2"

    block_classes = [
        Krea2LoopBeforeDenoiser,
        Krea2LoopDenoiser,
        Krea2LoopAfterDenoiser,
    ]
    block_names = ["before_denoiser", "denoiser", "after_denoiser"]

    @property
    def description(self) -> str:
        return (
            "Denoise step that iteratively denoise the latents.\n"
            "Its loop logic is defined in `Krea2DenoiseLoopWrapper.__call__` method\n"
            "At each iteration, it runs blocks defined in `sub_blocks` sequencially:\n"
            " - `Krea2LoopBeforeDenoiser`\n"
            " - `Krea2LoopDenoiser`\n"
            " - `Krea2LoopAfterDenoiser`\n"
            "This block supports text2image and image2image tasks for Krea 2."
        )


# Krea 2 (inpainting)
class Krea2InpaintDenoiseStep(Krea2DenoiseLoopWrapper):
    model_name = "krea2"
    block_classes = [
        Krea2LoopBeforeDenoiser,
        Krea2LoopDenoiser,
        Krea2LoopAfterDenoiser,
        Krea2LoopAfterDenoiserInpaint,
    ]
    block_names = ["before_denoiser", "denoiser", "after_denoiser", "after_denoiser_inpaint"]

    @property
    def description(self) -> str:
        return (
            "Denoise step that iteratively denoise the latents. \n"
            "Its loop logic is defined in `Krea2DenoiseLoopWrapper.__call__` method \n"
            "At each iteration, it runs blocks defined in `sub_blocks` sequencially:\n"
            " - `Krea2LoopBeforeDenoiser`\n"
            " - `Krea2LoopDenoiser`\n"
            " - `Krea2LoopAfterDenoiser`\n"
            " - `Krea2LoopAfterDenoiserInpaint`\n"
            "This block supports inpainting tasks for Krea 2."
        )


# Krea 2 (reference-image edit)
class Krea2EditDenoiseStep(Krea2DenoiseLoopWrapper):
    model_name = "krea2"
    block_classes = [
        Krea2EditLoopBeforeDenoiser,
        Krea2LoopDenoiser,
        Krea2LoopAfterDenoiser,
    ]
    block_names = ["before_denoiser", "denoiser", "after_denoiser"]

    @property
    def description(self) -> str:
        return (
            "Denoise step that iteratively denoise the latents for the reference-image edit task.\n"
            "Its loop logic is defined in `Krea2DenoiseLoopWrapper.__call__` method\n"
            "At each iteration, it runs blocks defined in `sub_blocks` sequencially:\n"
            " - `Krea2EditLoopBeforeDenoiser` (appends the clean reference tokens)\n"
            " - `Krea2LoopDenoiser`\n"
            " - `Krea2LoopAfterDenoiser`\n"
            "This block supports reference-image (edit) generation for Krea 2."
        )