diff --git a/src/diffusers/__init__.py b/src/diffusers/__init__.py index d205b93c1f..3092b2ddc0 100644 --- a/src/diffusers/__init__.py +++ b/src/diffusers/__init__.py @@ -88,6 +88,7 @@ else: "ControlNetModel", "ControlNetXSAdapter", "DiTTransformer2DModel", + "FluxControlNetModel", "FluxTransformer2DModel", "HunyuanDiT2DControlNetModel", "HunyuanDiT2DModel", @@ -254,6 +255,7 @@ else: "CLIPImageProjection", "CogVideoXPipeline", "CycleDiffusionPipeline", + "FluxControlNetPipeline", "FluxPipeline", "HunyuanDiTControlNetPipeline", "HunyuanDiTPAGPipeline", @@ -550,6 +552,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT: ControlNetModel, ControlNetXSAdapter, DiTTransformer2DModel, + FluxControlNetModel, FluxTransformer2DModel, HunyuanDiT2DControlNetModel, HunyuanDiT2DModel, @@ -694,6 +697,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT: CLIPImageProjection, CogVideoXPipeline, CycleDiffusionPipeline, + FluxControlNetPipeline, FluxPipeline, HunyuanDiTControlNetPipeline, HunyuanDiTPAGPipeline, diff --git a/src/diffusers/models/__init__.py b/src/diffusers/models/__init__.py index fe57b64666..4230c1a488 100644 --- a/src/diffusers/models/__init__.py +++ b/src/diffusers/models/__init__.py @@ -35,6 +35,7 @@ if is_torch_available(): _import_structure["autoencoders.consistency_decoder_vae"] = ["ConsistencyDecoderVAE"] _import_structure["autoencoders.vq_model"] = ["VQModel"] _import_structure["controlnet"] = ["ControlNetModel"] + _import_structure["controlnet_flux"] = ["FluxControlNetModel"] _import_structure["controlnet_hunyuan"] = ["HunyuanDiT2DControlNetModel", "HunyuanDiT2DMultiControlNetModel"] _import_structure["controlnet_sd3"] = ["SD3ControlNetModel", "SD3MultiControlNetModel"] _import_structure["controlnet_sparsectrl"] = ["SparseControlNetModel"] @@ -87,6 +88,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT: VQModel, ) from .controlnet import ControlNetModel + from .controlnet_flux import FluxControlNetModel from .controlnet_hunyuan import HunyuanDiT2DControlNetModel, HunyuanDiT2DMultiControlNetModel from .controlnet_sd3 import SD3ControlNetModel, SD3MultiControlNetModel from .controlnet_sparsectrl import SparseControlNetModel diff --git a/src/diffusers/models/controlnet_flux.py b/src/diffusers/models/controlnet_flux.py new file mode 100644 index 0000000000..ba4933dcad --- /dev/null +++ b/src/diffusers/models/controlnet_flux.py @@ -0,0 +1,374 @@ +# Copyright 2024 Black Forest Labs, The HuggingFace Team and The InstantX 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. + +from dataclasses import dataclass +from typing import Any, Dict, List, Optional, Tuple, Union + +import torch +import torch.nn as nn + +from ..configuration_utils import ConfigMixin, register_to_config +from ..loaders import PeftAdapterMixin +from ..models.attention_processor import AttentionProcessor +from ..models.modeling_utils import ModelMixin +from ..utils import USE_PEFT_BACKEND, is_torch_version, logging, scale_lora_layers, unscale_lora_layers +from .controlnet import BaseOutput, zero_module +from .embeddings import CombinedTimestepGuidanceTextProjEmbeddings, CombinedTimestepTextProjEmbeddings +from .modeling_outputs import Transformer2DModelOutput +from .transformers.transformer_flux import EmbedND, FluxSingleTransformerBlock, FluxTransformerBlock + + +logger = logging.get_logger(__name__) # pylint: disable=invalid-name + + +@dataclass +class FluxControlNetOutput(BaseOutput): + controlnet_block_samples: Tuple[torch.Tensor] + controlnet_single_block_samples: Tuple[torch.Tensor] + + +class FluxControlNetModel(ModelMixin, ConfigMixin, PeftAdapterMixin): + _supports_gradient_checkpointing = True + + @register_to_config + def __init__( + self, + patch_size: int = 1, + in_channels: int = 64, + num_layers: int = 19, + num_single_layers: int = 38, + attention_head_dim: int = 128, + num_attention_heads: int = 24, + joint_attention_dim: int = 4096, + pooled_projection_dim: int = 768, + guidance_embeds: bool = False, + axes_dims_rope: List[int] = [16, 56, 56], + ): + super().__init__() + self.out_channels = in_channels + self.inner_dim = num_attention_heads * attention_head_dim + + self.pos_embed = EmbedND(dim=self.inner_dim, theta=10000, axes_dim=axes_dims_rope) + text_time_guidance_cls = ( + CombinedTimestepGuidanceTextProjEmbeddings if guidance_embeds else CombinedTimestepTextProjEmbeddings + ) + self.time_text_embed = text_time_guidance_cls( + embedding_dim=self.inner_dim, pooled_projection_dim=pooled_projection_dim + ) + + self.context_embedder = nn.Linear(joint_attention_dim, self.inner_dim) + self.x_embedder = torch.nn.Linear(in_channels, self.inner_dim) + + self.transformer_blocks = nn.ModuleList( + [ + FluxTransformerBlock( + dim=self.inner_dim, + num_attention_heads=num_attention_heads, + attention_head_dim=attention_head_dim, + ) + for i in range(num_layers) + ] + ) + + self.single_transformer_blocks = nn.ModuleList( + [ + FluxSingleTransformerBlock( + dim=self.inner_dim, + num_attention_heads=num_attention_heads, + attention_head_dim=attention_head_dim, + ) + for i in range(num_single_layers) + ] + ) + + # controlnet_blocks + self.controlnet_blocks = nn.ModuleList([]) + for _ in range(len(self.transformer_blocks)): + self.controlnet_blocks.append(zero_module(nn.Linear(self.inner_dim, self.inner_dim))) + + self.controlnet_single_blocks = nn.ModuleList([]) + for _ in range(len(self.single_transformer_blocks)): + self.controlnet_single_blocks.append(zero_module(nn.Linear(self.inner_dim, self.inner_dim))) + + self.controlnet_x_embedder = zero_module(torch.nn.Linear(in_channels, self.inner_dim)) + + self.gradient_checkpointing = False + + @property + # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.attn_processors + def attn_processors(self): + r""" + Returns: + `dict` of attention processors: A dictionary containing all attention processors used in the model with + indexed by its weight name. + """ + # set recursively + processors = {} + + def fn_recursive_add_processors(name: str, module: torch.nn.Module, processors: Dict[str, AttentionProcessor]): + if hasattr(module, "get_processor"): + processors[f"{name}.processor"] = module.get_processor() + + for sub_name, child in module.named_children(): + fn_recursive_add_processors(f"{name}.{sub_name}", child, processors) + + return processors + + for name, module in self.named_children(): + fn_recursive_add_processors(name, module, processors) + + return processors + + # Copied from diffusers.models.unets.unet_2d_condition.UNet2DConditionModel.set_attn_processor + def set_attn_processor(self, processor): + r""" + Sets the attention processor to use to compute attention. + + Parameters: + processor (`dict` of `AttentionProcessor` or only `AttentionProcessor`): + The instantiated processor class or a dictionary of processor classes that will be set as the processor + for **all** `Attention` layers. + + If `processor` is a dict, the key needs to define the path to the corresponding cross attention + processor. This is strongly recommended when setting trainable attention processors. + + """ + count = len(self.attn_processors.keys()) + + if isinstance(processor, dict) and len(processor) != count: + raise ValueError( + f"A dict of processors was passed, but the number of processors {len(processor)} does not match the" + f" number of attention layers: {count}. Please make sure to pass {count} processor classes." + ) + + def fn_recursive_attn_processor(name: str, module: torch.nn.Module, processor): + if hasattr(module, "set_processor"): + if not isinstance(processor, dict): + module.set_processor(processor) + else: + module.set_processor(processor.pop(f"{name}.processor")) + + for sub_name, child in module.named_children(): + fn_recursive_attn_processor(f"{name}.{sub_name}", child, processor) + + for name, module in self.named_children(): + fn_recursive_attn_processor(name, module, processor) + + def _set_gradient_checkpointing(self, module, value=False): + if hasattr(module, "gradient_checkpointing"): + module.gradient_checkpointing = value + + @classmethod + def from_transformer( + cls, + transformer, + num_layers=4, + num_single_layers=10, + attention_head_dim: int = 128, + num_attention_heads: int = 24, + load_weights_from_transformer=True, + ): + config = transformer.config + config["num_layers"] = num_layers + config["num_single_layers"] = num_single_layers + config["attention_head_dim"] = attention_head_dim + config["num_attention_heads"] = num_attention_heads + + controlnet = cls(**config) + + if load_weights_from_transformer: + controlnet.pos_embed.load_state_dict(transformer.pos_embed.state_dict()) + controlnet.time_text_embed.load_state_dict(transformer.time_text_embed.state_dict()) + controlnet.context_embedder.load_state_dict(transformer.context_embedder.state_dict()) + controlnet.x_embedder.load_state_dict(transformer.x_embedder.state_dict()) + controlnet.transformer_blocks.load_state_dict(transformer.transformer_blocks.state_dict(), strict=False) + controlnet.single_transformer_blocks.load_state_dict( + transformer.single_transformer_blocks.state_dict(), strict=False + ) + + controlnet.controlnet_x_embedder = zero_module(controlnet.controlnet_x_embedder) + + return controlnet + + def forward( + self, + hidden_states: torch.Tensor, + controlnet_cond: torch.Tensor, + conditioning_scale: float = 1.0, + encoder_hidden_states: torch.Tensor = None, + pooled_projections: torch.Tensor = None, + timestep: torch.LongTensor = None, + img_ids: torch.Tensor = None, + txt_ids: torch.Tensor = None, + guidance: torch.Tensor = None, + joint_attention_kwargs: Optional[Dict[str, Any]] = None, + return_dict: bool = True, + ) -> Union[torch.FloatTensor, Transformer2DModelOutput]: + """ + The [`FluxTransformer2DModel`] forward method. + + Args: + hidden_states (`torch.FloatTensor` of shape `(batch size, channel, height, width)`): + Input `hidden_states`. + encoder_hidden_states (`torch.FloatTensor` of shape `(batch size, sequence_len, embed_dims)`): + Conditional embeddings (embeddings computed from the input conditions such as prompts) to use. + pooled_projections (`torch.FloatTensor` of shape `(batch_size, projection_dim)`): Embeddings projected + from the embeddings of input conditions. + timestep ( `torch.LongTensor`): + Used to indicate denoising step. + block_controlnet_hidden_states: (`list` of `torch.Tensor`): + A list of tensors that if specified are added to the residuals of transformer blocks. + joint_attention_kwargs (`dict`, *optional*): + A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under + `self.processor` in + [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). + return_dict (`bool`, *optional*, defaults to `True`): + Whether or not to return a [`~models.transformer_2d.Transformer2DModelOutput`] instead of a plain + tuple. + + Returns: + If `return_dict` is True, an [`~models.transformer_2d.Transformer2DModelOutput`] is returned, otherwise a + `tuple` where the first element is the sample tensor. + """ + if joint_attention_kwargs is not None: + joint_attention_kwargs = joint_attention_kwargs.copy() + lora_scale = joint_attention_kwargs.pop("scale", 1.0) + else: + lora_scale = 1.0 + + if USE_PEFT_BACKEND: + # weight the lora layers by setting `lora_scale` for each PEFT layer + scale_lora_layers(self, lora_scale) + else: + if joint_attention_kwargs is not None and joint_attention_kwargs.get("scale", None) is not None: + logger.warning( + "Passing `scale` via `joint_attention_kwargs` when not using the PEFT backend is ineffective." + ) + hidden_states = self.x_embedder(hidden_states) + + # add + hidden_states = hidden_states + self.controlnet_x_embedder(controlnet_cond) + + timestep = timestep.to(hidden_states.dtype) * 1000 + if guidance is not None: + guidance = guidance.to(hidden_states.dtype) * 1000 + else: + guidance = None + temb = ( + self.time_text_embed(timestep, pooled_projections) + if guidance is None + else self.time_text_embed(timestep, guidance, pooled_projections) + ) + encoder_hidden_states = self.context_embedder(encoder_hidden_states) + + txt_ids = txt_ids.expand(img_ids.size(0), -1, -1) + ids = torch.cat((txt_ids, img_ids), dim=1) + image_rotary_emb = self.pos_embed(ids) + + block_samples = () + for index_block, block in enumerate(self.transformer_blocks): + if self.training and self.gradient_checkpointing: + + def create_custom_forward(module, return_dict=None): + def custom_forward(*inputs): + if return_dict is not None: + return module(*inputs, return_dict=return_dict) + else: + return module(*inputs) + + return custom_forward + + ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {} + encoder_hidden_states, hidden_states = torch.utils.checkpoint.checkpoint( + create_custom_forward(block), + hidden_states, + encoder_hidden_states, + temb, + image_rotary_emb, + **ckpt_kwargs, + ) + + else: + encoder_hidden_states, hidden_states = block( + hidden_states=hidden_states, + encoder_hidden_states=encoder_hidden_states, + temb=temb, + image_rotary_emb=image_rotary_emb, + ) + block_samples = block_samples + (hidden_states,) + + hidden_states = torch.cat([encoder_hidden_states, hidden_states], dim=1) + + single_block_samples = () + for index_block, block in enumerate(self.single_transformer_blocks): + if self.training and self.gradient_checkpointing: + + def create_custom_forward(module, return_dict=None): + def custom_forward(*inputs): + if return_dict is not None: + return module(*inputs, return_dict=return_dict) + else: + return module(*inputs) + + return custom_forward + + ckpt_kwargs: Dict[str, Any] = {"use_reentrant": False} if is_torch_version(">=", "1.11.0") else {} + hidden_states = torch.utils.checkpoint.checkpoint( + create_custom_forward(block), + hidden_states, + temb, + image_rotary_emb, + **ckpt_kwargs, + ) + + else: + hidden_states = block( + hidden_states=hidden_states, + temb=temb, + image_rotary_emb=image_rotary_emb, + ) + single_block_samples = single_block_samples + (hidden_states[:, encoder_hidden_states.shape[1] :],) + + # controlnet block + controlnet_block_samples = () + for block_sample, controlnet_block in zip(block_samples, self.controlnet_blocks): + block_sample = controlnet_block(block_sample) + controlnet_block_samples = controlnet_block_samples + (block_sample,) + + controlnet_single_block_samples = () + for single_block_sample, controlnet_block in zip(single_block_samples, self.controlnet_single_blocks): + single_block_sample = controlnet_block(single_block_sample) + controlnet_single_block_samples = controlnet_single_block_samples + (single_block_sample,) + + # scaling + controlnet_block_samples = [sample * conditioning_scale for sample in controlnet_block_samples] + controlnet_single_block_samples = [sample * conditioning_scale for sample in controlnet_single_block_samples] + + # + controlnet_block_samples = None if len(controlnet_block_samples) == 0 else controlnet_block_samples + controlnet_single_block_samples = ( + None if len(controlnet_single_block_samples) == 0 else controlnet_single_block_samples + ) + + if USE_PEFT_BACKEND: + # remove `lora_scale` from each PEFT layer + unscale_lora_layers(self, lora_scale) + + if not return_dict: + return (controlnet_block_samples, controlnet_single_block_samples) + + return FluxControlNetOutput( + controlnet_block_samples=controlnet_block_samples, + controlnet_single_block_samples=controlnet_single_block_samples, + ) diff --git a/src/diffusers/models/transformers/transformer_flux.py b/src/diffusers/models/transformers/transformer_flux.py index 3983606e46..3168fd9a62 100644 --- a/src/diffusers/models/transformers/transformer_flux.py +++ b/src/diffusers/models/transformers/transformer_flux.py @@ -1,4 +1,4 @@ -# Copyright 2024 Black Forest Labs, The HuggingFace Team. All rights reserved. +# Copyright 2024 Black Forest Labs, The HuggingFace Team and The InstantX 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. @@ -15,6 +15,7 @@ from typing import Any, Dict, List, Optional, Union +import numpy as np import torch import torch.nn as nn import torch.nn.functional as F @@ -321,6 +322,8 @@ class FluxTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrig txt_ids: torch.Tensor = None, guidance: torch.Tensor = None, joint_attention_kwargs: Optional[Dict[str, Any]] = None, + controlnet_block_samples=None, + controlnet_single_block_samples=None, return_dict: bool = True, ) -> Union[torch.FloatTensor, Transformer2DModelOutput]: """ @@ -377,6 +380,7 @@ class FluxTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrig ) encoder_hidden_states = self.context_embedder(encoder_hidden_states) + txt_ids = txt_ids.expand(img_ids.size(0), -1, -1) ids = torch.cat((txt_ids, img_ids), dim=1) image_rotary_emb = self.pos_embed(ids) @@ -410,6 +414,12 @@ class FluxTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrig image_rotary_emb=image_rotary_emb, ) + # controlnet residual + if controlnet_block_samples is not None: + interval_control = len(self.transformer_blocks) / len(controlnet_block_samples) + interval_control = int(np.ceil(interval_control)) + hidden_states = hidden_states + controlnet_block_samples[index_block // interval_control] + hidden_states = torch.cat([encoder_hidden_states, hidden_states], dim=1) for index_block, block in enumerate(self.single_transformer_blocks): @@ -440,6 +450,15 @@ class FluxTransformer2DModel(ModelMixin, ConfigMixin, PeftAdapterMixin, FromOrig image_rotary_emb=image_rotary_emb, ) + # controlnet residual + if controlnet_single_block_samples is not None: + interval_control = len(self.single_transformer_blocks) / len(controlnet_single_block_samples) + interval_control = int(np.ceil(interval_control)) + hidden_states[:, encoder_hidden_states.shape[1] :, ...] = ( + hidden_states[:, encoder_hidden_states.shape[1] :, ...] + + controlnet_single_block_samples[index_block // interval_control] + ) + hidden_states = hidden_states[:, encoder_hidden_states.shape[1] :, ...] hidden_states = self.norm_out(hidden_states, temb) diff --git a/src/diffusers/pipelines/__init__.py b/src/diffusers/pipelines/__init__.py index 7e46ed7e49..84e9f0b2e0 100644 --- a/src/diffusers/pipelines/__init__.py +++ b/src/diffusers/pipelines/__init__.py @@ -124,7 +124,7 @@ else: "AnimateDiffSparseControlNetPipeline", "AnimateDiffVideoToVideoPipeline", ] - _import_structure["flux"] = ["FluxPipeline"] + _import_structure["flux"] = ["FluxPipeline", "FluxControlNetPipeline"] _import_structure["audioldm"] = ["AudioLDMPipeline"] _import_structure["audioldm2"] = [ "AudioLDM2Pipeline", @@ -493,7 +493,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT: VersatileDiffusionTextToImagePipeline, VQDiffusionPipeline, ) - from .flux import FluxPipeline + from .flux import FluxControlNetPipeline, FluxPipeline from .hunyuandit import HunyuanDiTPipeline from .i2vgen_xl import I2VGenXLPipeline from .kandinsky import ( diff --git a/src/diffusers/pipelines/flux/__init__.py b/src/diffusers/pipelines/flux/__init__.py index d8c3edf0ea..900189102c 100644 --- a/src/diffusers/pipelines/flux/__init__.py +++ b/src/diffusers/pipelines/flux/__init__.py @@ -23,6 +23,7 @@ except OptionalDependencyNotAvailable: _dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects)) else: _import_structure["pipeline_flux"] = ["FluxPipeline"] + _import_structure["pipeline_flux_controlnet"] = ["FluxControlNetPipeline"] if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT: try: if not (is_transformers_available() and is_torch_available()): @@ -31,6 +32,7 @@ if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT: from ...utils.dummy_torch_and_transformers_objects import * # noqa F403 else: from .pipeline_flux import FluxPipeline + from .pipeline_flux_controlnet import FluxControlNetPipeline else: import sys diff --git a/src/diffusers/pipelines/flux/pipeline_flux_controlnet.py b/src/diffusers/pipelines/flux/pipeline_flux_controlnet.py new file mode 100644 index 0000000000..84450374cb --- /dev/null +++ b/src/diffusers/pipelines/flux/pipeline_flux_controlnet.py @@ -0,0 +1,861 @@ +# Copyright 2024 Black Forest Labs, The HuggingFace Team and The InstantX 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 +from typing import Any, Callable, Dict, List, Optional, Union + +import numpy as np +import torch +from transformers import ( + CLIPTextModel, + CLIPTokenizer, + T5EncoderModel, + T5TokenizerFast, +) + +from ...image_processor import PipelineImageInput, VaeImageProcessor +from ...loaders import FluxLoraLoaderMixin +from ...models.autoencoders import AutoencoderKL +from ...models.controlnet_flux import FluxControlNetModel +from ...models.transformers import FluxTransformer2DModel +from ...schedulers import FlowMatchEulerDiscreteScheduler +from ...utils import ( + USE_PEFT_BACKEND, + is_torch_xla_available, + logging, + replace_example_docstring, + scale_lora_layers, + unscale_lora_layers, +) +from ...utils.torch_utils import randn_tensor +from ..pipeline_utils import DiffusionPipeline +from .pipeline_output import FluxPipelineOutput + + +if is_torch_xla_available(): + import torch_xla.core.xla_model as xm + + XLA_AVAILABLE = True +else: + XLA_AVAILABLE = False + + +logger = logging.get_logger(__name__) # pylint: disable=invalid-name + +EXAMPLE_DOC_STRING = """ + Examples: + ```py + >>> import torch + >>> from diffusers.utils import load_image + >>> from diffusers import FluxControlNetPipeline + >>> from diffusers import FluxControlNetModel + + >>> controlnet_model = "InstantX/FLUX.1-dev-controlnet-canny-alpha" + >>> controlnet = FluxControlNetModel.from_pretrained(controlnet_model, torch_dtype=torch.bfloat16) + >>> pipe = FluxControlNetPipeline.from_pretrained( + ... base_model, controlnet=controlnet, torch_dtype=torch.bfloat16 + ... ) + >>> pipe.to("cuda") + >>> control_image = load_image("https://huggingface.co/InstantX/SD3-Controlnet-Canny/resolve/main/canny.jpg") + >>> prompt = "A girl in city, 25 years old, cool, futuristic" + >>> image = pipe( + ... prompt, + ... control_image=control_image, + ... controlnet_conditioning_scale=0.6, + ... num_inference_steps=28, + ... guidance_scale=3.5, + ... ).images[0] + >>> image.save("flux.png") + ``` +""" + + +# Copied from diffusers.pipelines.flux.pipeline_flux.calculate_shift +def calculate_shift( + image_seq_len, + base_seq_len: int = 256, + max_seq_len: int = 4096, + base_shift: float = 0.5, + max_shift: float = 1.16, +): + m = (max_shift - base_shift) / (max_seq_len - base_seq_len) + b = base_shift - m * base_seq_len + mu = image_seq_len * m + b + return mu + + +# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.retrieve_timesteps +def retrieve_timesteps( + scheduler, + num_inference_steps: Optional[int] = None, + device: Optional[Union[str, torch.device]] = None, + timesteps: Optional[List[int]] = None, + sigmas: Optional[List[float]] = None, + **kwargs, +): + """ + Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles + custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. + + Args: + scheduler (`SchedulerMixin`): + The scheduler to get timesteps from. + num_inference_steps (`int`): + The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps` + must be `None`. + device (`str` or `torch.device`, *optional*): + The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. + timesteps (`List[int]`, *optional*): + Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed, + `num_inference_steps` and `sigmas` must be `None`. + sigmas (`List[float]`, *optional*): + Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed, + `num_inference_steps` and `timesteps` must be `None`. + + Returns: + `Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the + second element is the number of inference steps. + """ + if timesteps is not None and sigmas is not None: + raise ValueError("Only one of `timesteps` or `sigmas` can be passed. Please choose one to set custom values") + if timesteps is not None: + accepts_timesteps = "timesteps" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) + if not accepts_timesteps: + raise ValueError( + f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" + f" timestep schedules. Please check whether you are using the correct scheduler." + ) + scheduler.set_timesteps(timesteps=timesteps, device=device, **kwargs) + timesteps = scheduler.timesteps + num_inference_steps = len(timesteps) + elif sigmas is not None: + accept_sigmas = "sigmas" in set(inspect.signature(scheduler.set_timesteps).parameters.keys()) + if not accept_sigmas: + raise ValueError( + f"The current scheduler class {scheduler.__class__}'s `set_timesteps` does not support custom" + f" sigmas schedules. Please check whether you are using the correct scheduler." + ) + scheduler.set_timesteps(sigmas=sigmas, device=device, **kwargs) + timesteps = scheduler.timesteps + num_inference_steps = len(timesteps) + else: + scheduler.set_timesteps(num_inference_steps, device=device, **kwargs) + timesteps = scheduler.timesteps + return timesteps, num_inference_steps + + +class FluxControlNetPipeline(DiffusionPipeline, FluxLoraLoaderMixin): + r""" + The Flux pipeline for text-to-image generation. + + Reference: https://blackforestlabs.ai/announcing-black-forest-labs/ + + Args: + transformer ([`FluxTransformer2DModel`]): + Conditional Transformer (MMDiT) architecture to denoise the encoded image latents. + scheduler ([`FlowMatchEulerDiscreteScheduler`]): + A scheduler to be used in combination with `transformer` to denoise the encoded image latents. + vae ([`AutoencoderKL`]): + Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations. + text_encoder ([`CLIPTextModel`]): + [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically + the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant. + text_encoder_2 ([`T5EncoderModel`]): + [T5](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5EncoderModel), specifically + the [google/t5-v1_1-xxl](https://huggingface.co/google/t5-v1_1-xxl) variant. + tokenizer (`CLIPTokenizer`): + Tokenizer of class + [CLIPTokenizer](https://huggingface.co/docs/transformers/en/model_doc/clip#transformers.CLIPTokenizer). + tokenizer_2 (`T5TokenizerFast`): + Second Tokenizer of class + [T5TokenizerFast](https://huggingface.co/docs/transformers/en/model_doc/t5#transformers.T5TokenizerFast). + """ + + model_cpu_offload_seq = "text_encoder->text_encoder_2->transformer->vae" + _optional_components = [] + _callback_tensor_inputs = ["latents", "prompt_embeds"] + + def __init__( + self, + scheduler: FlowMatchEulerDiscreteScheduler, + vae: AutoencoderKL, + text_encoder: CLIPTextModel, + tokenizer: CLIPTokenizer, + text_encoder_2: T5EncoderModel, + tokenizer_2: T5TokenizerFast, + transformer: FluxTransformer2DModel, + controlnet: FluxControlNetModel, + ): + super().__init__() + + self.register_modules( + vae=vae, + text_encoder=text_encoder, + text_encoder_2=text_encoder_2, + tokenizer=tokenizer, + tokenizer_2=tokenizer_2, + transformer=transformer, + scheduler=scheduler, + controlnet=controlnet, + ) + self.vae_scale_factor = ( + 2 ** (len(self.vae.config.block_out_channels)) if hasattr(self, "vae") and self.vae is not None else 16 + ) + self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor) + self.tokenizer_max_length = ( + self.tokenizer.model_max_length if hasattr(self, "tokenizer") and self.tokenizer is not None else 77 + ) + self.default_sample_size = 64 + + def _get_t5_prompt_embeds( + self, + prompt: Union[str, List[str]] = None, + num_images_per_prompt: int = 1, + max_sequence_length: int = 512, + device: Optional[torch.device] = None, + dtype: Optional[torch.dtype] = None, + ): + device = device or self._execution_device + dtype = dtype or self.text_encoder.dtype + + prompt = [prompt] if isinstance(prompt, str) else prompt + batch_size = len(prompt) + + text_inputs = self.tokenizer_2( + prompt, + padding="max_length", + max_length=max_sequence_length, + truncation=True, + return_length=False, + return_overflowing_tokens=False, + return_tensors="pt", + ) + text_input_ids = text_inputs.input_ids + untruncated_ids = self.tokenizer_2(prompt, padding="longest", return_tensors="pt").input_ids + + if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids): + removed_text = self.tokenizer_2.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1]) + logger.warning( + "The following part of your input was truncated because `max_sequence_length` is set to " + f" {max_sequence_length} tokens: {removed_text}" + ) + + prompt_embeds = self.text_encoder_2(text_input_ids.to(device), output_hidden_states=False)[0] + + dtype = self.text_encoder_2.dtype + prompt_embeds = prompt_embeds.to(dtype=dtype, device=device) + + _, seq_len, _ = prompt_embeds.shape + + # duplicate text embeddings and attention mask for each generation per prompt, using mps friendly method + prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) + prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1) + + return prompt_embeds + + def _get_clip_prompt_embeds( + self, + prompt: Union[str, List[str]], + num_images_per_prompt: int = 1, + device: Optional[torch.device] = None, + ): + device = device or self._execution_device + + prompt = [prompt] if isinstance(prompt, str) else prompt + batch_size = len(prompt) + + text_inputs = self.tokenizer( + prompt, + padding="max_length", + max_length=self.tokenizer_max_length, + truncation=True, + return_overflowing_tokens=False, + return_length=False, + return_tensors="pt", + ) + + text_input_ids = text_inputs.input_ids + untruncated_ids = self.tokenizer(prompt, padding="longest", return_tensors="pt").input_ids + if untruncated_ids.shape[-1] >= text_input_ids.shape[-1] and not torch.equal(text_input_ids, untruncated_ids): + removed_text = self.tokenizer.batch_decode(untruncated_ids[:, self.tokenizer_max_length - 1 : -1]) + logger.warning( + "The following part of your input was truncated because CLIP can only handle sequences up to" + f" {self.tokenizer_max_length} tokens: {removed_text}" + ) + prompt_embeds = self.text_encoder(text_input_ids.to(device), output_hidden_states=False) + + # Use pooled output of CLIPTextModel + prompt_embeds = prompt_embeds.pooler_output + prompt_embeds = prompt_embeds.to(dtype=self.text_encoder.dtype, device=device) + + # duplicate text embeddings for each generation per prompt, using mps friendly method + prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1) + prompt_embeds = prompt_embeds.view(batch_size * num_images_per_prompt, -1) + + return prompt_embeds + + def encode_prompt( + self, + prompt: Union[str, List[str]], + prompt_2: Union[str, List[str]], + device: Optional[torch.device] = None, + num_images_per_prompt: int = 1, + prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + max_sequence_length: int = 512, + lora_scale: Optional[float] = None, + ): + r""" + + Args: + prompt (`str` or `List[str]`, *optional*): + prompt to be encoded + prompt_2 (`str` or `List[str]`, *optional*): + The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is + used in all text-encoders + device: (`torch.device`): + torch device + num_images_per_prompt (`int`): + number of images that should be generated per prompt + prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + clip_skip (`int`, *optional*): + Number of layers to be skipped from CLIP while computing the prompt embeddings. A value of 1 means that + the output of the pre-final layer will be used for computing the prompt embeddings. + lora_scale (`float`, *optional*): + A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded. + """ + device = device or self._execution_device + + # set lora scale so that monkey patched LoRA + # function of text encoder can correctly access it + if lora_scale is not None and isinstance(self, FluxLoraLoaderMixin): + self._lora_scale = lora_scale + + # dynamically adjust the LoRA scale + if self.text_encoder is not None and USE_PEFT_BACKEND: + scale_lora_layers(self.text_encoder, lora_scale) + if self.text_encoder_2 is not None and USE_PEFT_BACKEND: + scale_lora_layers(self.text_encoder_2, lora_scale) + + prompt = [prompt] if isinstance(prompt, str) else prompt + if prompt is not None: + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + if prompt_embeds is None: + prompt_2 = prompt_2 or prompt + prompt_2 = [prompt_2] if isinstance(prompt_2, str) else prompt_2 + + # We only use the pooled prompt output from the CLIPTextModel + pooled_prompt_embeds = self._get_clip_prompt_embeds( + prompt=prompt, + device=device, + num_images_per_prompt=num_images_per_prompt, + ) + prompt_embeds = self._get_t5_prompt_embeds( + prompt=prompt_2, + num_images_per_prompt=num_images_per_prompt, + max_sequence_length=max_sequence_length, + device=device, + ) + + if self.text_encoder is not None: + if isinstance(self, FluxLoraLoaderMixin) and USE_PEFT_BACKEND: + # Retrieve the original scale by scaling back the LoRA layers + unscale_lora_layers(self.text_encoder, lora_scale) + + if self.text_encoder_2 is not None: + if isinstance(self, FluxLoraLoaderMixin) and USE_PEFT_BACKEND: + # Retrieve the original scale by scaling back the LoRA layers + unscale_lora_layers(self.text_encoder_2, lora_scale) + + dtype = self.text_encoder.dtype if self.text_encoder is not None else self.transformer.dtype + text_ids = torch.zeros(batch_size, prompt_embeds.shape[1], 3).to(device=device, dtype=dtype) + text_ids = text_ids.repeat(num_images_per_prompt, 1, 1) + + return prompt_embeds, pooled_prompt_embeds, text_ids + + def check_inputs( + self, + prompt, + prompt_2, + height, + width, + prompt_embeds=None, + pooled_prompt_embeds=None, + callback_on_step_end_tensor_inputs=None, + max_sequence_length=None, + ): + if height % 8 != 0 or width % 8 != 0: + raise ValueError(f"`height` and `width` have to be divisible by 8 but are {height} and {width}.") + + if callback_on_step_end_tensor_inputs is not None and not all( + k in self._callback_tensor_inputs for k in callback_on_step_end_tensor_inputs + ): + raise ValueError( + f"`callback_on_step_end_tensor_inputs` has to be in {self._callback_tensor_inputs}, but found {[k for k in callback_on_step_end_tensor_inputs if k not in self._callback_tensor_inputs]}" + ) + + if prompt is not None and prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to" + " only forward one of the two." + ) + elif prompt_2 is not None and prompt_embeds is not None: + raise ValueError( + f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to" + " only forward one of the two." + ) + elif prompt is None and prompt_embeds is None: + raise ValueError( + "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined." + ) + elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)): + raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}") + elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)): + raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}") + + if prompt_embeds is not None and pooled_prompt_embeds is None: + raise ValueError( + "If `prompt_embeds` are provided, `pooled_prompt_embeds` also have to be passed. Make sure to generate `pooled_prompt_embeds` from the same text encoder that was used to generate `prompt_embeds`." + ) + + if max_sequence_length is not None and max_sequence_length > 512: + raise ValueError(f"`max_sequence_length` cannot be greater than 512 but is {max_sequence_length}") + + @staticmethod + # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._prepare_latent_image_ids + def _prepare_latent_image_ids(batch_size, height, width, device, dtype): + latent_image_ids = torch.zeros(height // 2, width // 2, 3) + latent_image_ids[..., 1] = latent_image_ids[..., 1] + torch.arange(height // 2)[:, None] + latent_image_ids[..., 2] = latent_image_ids[..., 2] + torch.arange(width // 2)[None, :] + + latent_image_id_height, latent_image_id_width, latent_image_id_channels = latent_image_ids.shape + + latent_image_ids = latent_image_ids[None, :].repeat(batch_size, 1, 1, 1) + latent_image_ids = latent_image_ids.reshape( + batch_size, latent_image_id_height * latent_image_id_width, latent_image_id_channels + ) + + return latent_image_ids.to(device=device, dtype=dtype) + + @staticmethod + # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._pack_latents + def _pack_latents(latents, batch_size, num_channels_latents, height, width): + latents = latents.view(batch_size, num_channels_latents, height // 2, 2, width // 2, 2) + latents = latents.permute(0, 2, 4, 1, 3, 5) + latents = latents.reshape(batch_size, (height // 2) * (width // 2), num_channels_latents * 4) + + return latents + + @staticmethod + # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline._unpack_latents + def _unpack_latents(latents, height, width, vae_scale_factor): + batch_size, num_patches, channels = latents.shape + + height = height // vae_scale_factor + width = width // vae_scale_factor + + latents = latents.view(batch_size, height, width, channels // 4, 2, 2) + latents = latents.permute(0, 3, 1, 4, 2, 5) + + latents = latents.reshape(batch_size, channels // (2 * 2), height * 2, width * 2) + + return latents + + # Copied from diffusers.pipelines.flux.pipeline_flux.FluxPipeline.prepare_latents + def prepare_latents( + self, + batch_size, + num_channels_latents, + height, + width, + dtype, + device, + generator, + latents=None, + ): + height = 2 * (int(height) // self.vae_scale_factor) + width = 2 * (int(width) // self.vae_scale_factor) + + shape = (batch_size, num_channels_latents, height, width) + + if latents is not None: + latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype) + return latents.to(device=device, dtype=dtype), latent_image_ids + + if isinstance(generator, list) and len(generator) != batch_size: + raise ValueError( + f"You have passed a list of generators of length {len(generator)}, but requested an effective batch" + f" size of {batch_size}. Make sure the batch size matches the length of the generators." + ) + + latents = randn_tensor(shape, generator=generator, device=device, dtype=dtype) + latents = self._pack_latents(latents, batch_size, num_channels_latents, height, width) + + latent_image_ids = self._prepare_latent_image_ids(batch_size, height, width, device, dtype) + + return latents, latent_image_ids + + # Copied from diffusers.pipelines.controlnet_sd3.pipeline_stable_diffusion_3_controlnet.StableDiffusion3ControlNetPipeline.prepare_image + def prepare_image( + self, + image, + width, + height, + batch_size, + num_images_per_prompt, + device, + dtype, + do_classifier_free_guidance=False, + guess_mode=False, + ): + if isinstance(image, torch.Tensor): + pass + else: + image = self.image_processor.preprocess(image, height=height, width=width) + + image_batch_size = image.shape[0] + + if image_batch_size == 1: + repeat_by = batch_size + else: + # image batch size is the same as prompt batch size + repeat_by = num_images_per_prompt + + image = image.repeat_interleave(repeat_by, dim=0) + + image = image.to(device=device, dtype=dtype) + + if do_classifier_free_guidance and not guess_mode: + image = torch.cat([image] * 2) + + return image + + @property + def guidance_scale(self): + return self._guidance_scale + + @property + def joint_attention_kwargs(self): + return self._joint_attention_kwargs + + @property + def num_timesteps(self): + return self._num_timesteps + + @property + def interrupt(self): + return self._interrupt + + @torch.no_grad() + @replace_example_docstring(EXAMPLE_DOC_STRING) + def __call__( + self, + prompt: Union[str, List[str]] = None, + prompt_2: Optional[Union[str, List[str]]] = None, + height: Optional[int] = None, + width: Optional[int] = None, + num_inference_steps: int = 28, + timesteps: List[int] = None, + guidance_scale: float = 7.0, + control_image: PipelineImageInput = None, + controlnet_conditioning_scale: Union[float, List[float]] = 1.0, + num_images_per_prompt: Optional[int] = 1, + generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None, + latents: Optional[torch.FloatTensor] = None, + prompt_embeds: Optional[torch.FloatTensor] = None, + pooled_prompt_embeds: Optional[torch.FloatTensor] = None, + output_type: Optional[str] = "pil", + return_dict: bool = True, + joint_attention_kwargs: Optional[Dict[str, Any]] = None, + callback_on_step_end: Optional[Callable[[int, int, Dict], None]] = None, + callback_on_step_end_tensor_inputs: List[str] = ["latents"], + max_sequence_length: int = 512, + ): + r""" + Function invoked when calling the pipeline for generation. + + Args: + prompt (`str` or `List[str]`, *optional*): + The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`. + instead. + prompt_2 (`str` or `List[str]`, *optional*): + The prompt or prompts to be sent to `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is + will be used instead + height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor): + The height in pixels of the generated image. This is set to 1024 by default for the best results. + width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor): + The width in pixels of the generated image. This is set to 1024 by default for the best results. + num_inference_steps (`int`, *optional*, defaults to 50): + The number of denoising steps. More denoising steps usually lead to a higher quality image at the + expense of slower inference. + timesteps (`List[int]`, *optional*): + Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument + in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is + passed will be used. Must be in descending order. + guidance_scale (`float`, *optional*, defaults to 7.0): + Guidance scale as defined in [Classifier-Free Diffusion Guidance](https://arxiv.org/abs/2207.12598). + `guidance_scale` is defined as `w` of equation 2. of [Imagen + Paper](https://arxiv.org/pdf/2205.11487.pdf). Guidance scale is enabled by setting `guidance_scale > + 1`. Higher guidance scale encourages to generate images that are closely linked to the text `prompt`, + usually at the expense of lower image quality. + num_images_per_prompt (`int`, *optional*, defaults to 1): + The number of images to generate per prompt. + generator (`torch.Generator` or `List[torch.Generator]`, *optional*): + One or a list of [torch generator(s)](https://pytorch.org/docs/stable/generated/torch.Generator.html) + to make generation deterministic. + latents (`torch.FloatTensor`, *optional*): + Pre-generated noisy latents, sampled from a Gaussian distribution, to be used as inputs for image + generation. Can be used to tweak the same generation with different prompts. If not provided, a latents + tensor will ge generated by sampling using the supplied random `generator`. + prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. If not + provided, text embeddings will be generated from `prompt` input argument. + pooled_prompt_embeds (`torch.FloatTensor`, *optional*): + Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting. + If not provided, pooled text embeddings will be generated from `prompt` input argument. + output_type (`str`, *optional*, defaults to `"pil"`): + The output format of the generate image. Choose between + [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`. + return_dict (`bool`, *optional*, defaults to `True`): + Whether or not to return a [`~pipelines.flux.FluxPipelineOutput`] instead of a plain tuple. + joint_attention_kwargs (`dict`, *optional*): + A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under + `self.processor` in + [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). + callback_on_step_end (`Callable`, *optional*): + A function that calls at the end of each denoising steps during the inference. The function is called + with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int, + callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by + `callback_on_step_end_tensor_inputs`. + callback_on_step_end_tensor_inputs (`List`, *optional*): + The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list + will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the + `._callback_tensor_inputs` attribute of your pipeline class. + max_sequence_length (`int` defaults to 512): Maximum sequence length to use with the `prompt`. + + Examples: + + Returns: + [`~pipelines.flux.FluxPipelineOutput`] or `tuple`: [`~pipelines.flux.FluxPipelineOutput`] if `return_dict` + is True, otherwise a `tuple`. When returning a tuple, the first element is a list with the generated + images. + """ + + height = height or self.default_sample_size * self.vae_scale_factor + width = width or self.default_sample_size * self.vae_scale_factor + + # 1. Check inputs. Raise error if not correct + self.check_inputs( + prompt, + prompt_2, + height, + width, + prompt_embeds=prompt_embeds, + pooled_prompt_embeds=pooled_prompt_embeds, + callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs, + max_sequence_length=max_sequence_length, + ) + + self._guidance_scale = guidance_scale + self._joint_attention_kwargs = joint_attention_kwargs + self._interrupt = False + + # 2. Define call parameters + if prompt is not None and isinstance(prompt, str): + batch_size = 1 + elif prompt is not None and isinstance(prompt, list): + batch_size = len(prompt) + else: + batch_size = prompt_embeds.shape[0] + + device = self._execution_device + dtype = self.transformer.dtype + + lora_scale = ( + self.joint_attention_kwargs.get("scale", None) if self.joint_attention_kwargs is not None else None + ) + ( + prompt_embeds, + pooled_prompt_embeds, + text_ids, + ) = self.encode_prompt( + prompt=prompt, + prompt_2=prompt_2, + prompt_embeds=prompt_embeds, + pooled_prompt_embeds=pooled_prompt_embeds, + device=device, + num_images_per_prompt=num_images_per_prompt, + max_sequence_length=max_sequence_length, + lora_scale=lora_scale, + ) + + # 3. Prepare control image + num_channels_latents = self.transformer.config.in_channels // 4 + if isinstance(self.controlnet, FluxControlNetModel): + control_image = self.prepare_image( + image=control_image, + width=width, + height=height, + batch_size=batch_size * num_images_per_prompt, + num_images_per_prompt=num_images_per_prompt, + device=device, + dtype=dtype, + ) + height, width = control_image.shape[-2:] + + # vae encode + control_image = self.vae.encode(control_image).latent_dist.sample() + control_image = (control_image - self.vae.config.shift_factor) * self.vae.config.scaling_factor + + # pack + height_control_image, width_control_image = control_image.shape[2:] + control_image = self._pack_latents( + control_image, + batch_size * num_images_per_prompt, + num_channels_latents, + height_control_image, + width_control_image, + ) + + # 4. Prepare latent variables + num_channels_latents = self.transformer.config.in_channels // 4 + latents, latent_image_ids = self.prepare_latents( + batch_size * num_images_per_prompt, + num_channels_latents, + height, + width, + prompt_embeds.dtype, + device, + generator, + latents, + ) + + # 5. Prepare timesteps + sigmas = np.linspace(1.0, 1 / num_inference_steps, num_inference_steps) + image_seq_len = latents.shape[1] + mu = calculate_shift( + image_seq_len, + self.scheduler.config.base_image_seq_len, + self.scheduler.config.max_image_seq_len, + self.scheduler.config.base_shift, + self.scheduler.config.max_shift, + ) + timesteps, num_inference_steps = retrieve_timesteps( + self.scheduler, + num_inference_steps, + device, + timesteps, + sigmas, + mu=mu, + ) + + num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0) + self._num_timesteps = len(timesteps) + + # 6. Denoising loop + with self.progress_bar(total=num_inference_steps) as progress_bar: + for i, t in enumerate(timesteps): + if self.interrupt: + continue + + # broadcast to batch dimension in a way that's compatible with ONNX/Core ML + timestep = t.expand(latents.shape[0]).to(latents.dtype) + + # handle guidance + if self.transformer.config.guidance_embeds: + guidance = torch.tensor([guidance_scale], device=device) + guidance = guidance.expand(latents.shape[0]) + else: + guidance = None + + # controlnet + controlnet_block_samples, controlnet_single_block_samples = self.controlnet( + hidden_states=latents, + controlnet_cond=control_image, + conditioning_scale=controlnet_conditioning_scale, + timestep=timestep / 1000, + guidance=guidance, + pooled_projections=pooled_prompt_embeds, + encoder_hidden_states=prompt_embeds, + txt_ids=text_ids, + img_ids=latent_image_ids, + joint_attention_kwargs=self.joint_attention_kwargs, + return_dict=False, + ) + + noise_pred = self.transformer( + hidden_states=latents, + # YiYi notes: divide it by 1000 for now because we scale it by 1000 in the transforme rmodel (we should not keep it but I want to keep the inputs same for the model for testing) + timestep=timestep / 1000, + guidance=guidance, + pooled_projections=pooled_prompt_embeds, + encoder_hidden_states=prompt_embeds, + controlnet_block_samples=controlnet_block_samples, + controlnet_single_block_samples=controlnet_single_block_samples, + txt_ids=text_ids, + img_ids=latent_image_ids, + joint_attention_kwargs=self.joint_attention_kwargs, + return_dict=False, + )[0] + + # compute the previous noisy sample x_t -> x_t-1 + latents_dtype = latents.dtype + latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0] + + if 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 + latents = latents.to(latents_dtype) + + if callback_on_step_end is not None: + callback_kwargs = {} + for k in callback_on_step_end_tensor_inputs: + callback_kwargs[k] = locals()[k] + callback_outputs = callback_on_step_end(self, i, t, callback_kwargs) + + latents = callback_outputs.pop("latents", latents) + prompt_embeds = callback_outputs.pop("prompt_embeds", prompt_embeds) + + # call the callback, if provided + if i == len(timesteps) - 1 or ((i + 1) > num_warmup_steps and (i + 1) % self.scheduler.order == 0): + progress_bar.update() + + if XLA_AVAILABLE: + xm.mark_step() + + if output_type == "latent": + image = latents + + else: + latents = self._unpack_latents(latents, height, width, self.vae_scale_factor) + latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor + + image = self.vae.decode(latents, return_dict=False)[0] + image = self.image_processor.postprocess(image, output_type=output_type) + + # Offload all models + self.maybe_free_model_hooks() + + if not return_dict: + return (image,) + + return FluxPipelineOutput(images=image) diff --git a/src/diffusers/utils/dummy_pt_objects.py b/src/diffusers/utils/dummy_pt_objects.py index 740b082491..0827dea44e 100644 --- a/src/diffusers/utils/dummy_pt_objects.py +++ b/src/diffusers/utils/dummy_pt_objects.py @@ -182,6 +182,21 @@ class DiTTransformer2DModel(metaclass=DummyObject): requires_backends(cls, ["torch"]) +class FluxControlNetModel(metaclass=DummyObject): + _backends = ["torch"] + + def __init__(self, *args, **kwargs): + requires_backends(self, ["torch"]) + + @classmethod + def from_config(cls, *args, **kwargs): + requires_backends(cls, ["torch"]) + + @classmethod + def from_pretrained(cls, *args, **kwargs): + requires_backends(cls, ["torch"]) + + class FluxTransformer2DModel(metaclass=DummyObject): _backends = ["torch"] diff --git a/src/diffusers/utils/dummy_torch_and_transformers_objects.py b/src/diffusers/utils/dummy_torch_and_transformers_objects.py index fb722553ed..e3d18121a1 100644 --- a/src/diffusers/utils/dummy_torch_and_transformers_objects.py +++ b/src/diffusers/utils/dummy_torch_and_transformers_objects.py @@ -287,6 +287,21 @@ class CycleDiffusionPipeline(metaclass=DummyObject): requires_backends(cls, ["torch", "transformers"]) +class FluxControlNetPipeline(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 FluxPipeline(metaclass=DummyObject): _backends = ["torch", "transformers"] diff --git a/tests/pipelines/controlnet_flux/__init__.py b/tests/pipelines/controlnet_flux/__init__.py new file mode 100644 index 0000000000..e69de29bb2 diff --git a/tests/pipelines/controlnet_flux/test_controlnet_flux.py b/tests/pipelines/controlnet_flux/test_controlnet_flux.py new file mode 100644 index 0000000000..d2db28bdda --- /dev/null +++ b/tests/pipelines/controlnet_flux/test_controlnet_flux.py @@ -0,0 +1,233 @@ +# coding=utf-8 +# Copyright 2024 HuggingFace Inc and The InstantX Team. +# +# 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 numpy as np +import torch +from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer, T5EncoderModel, T5TokenizerFast + +from diffusers import ( + AutoencoderKL, + FlowMatchEulerDiscreteScheduler, + FluxControlNetPipeline, + FluxTransformer2DModel, +) +from diffusers.models import FluxControlNetModel +from diffusers.utils import load_image +from diffusers.utils.testing_utils import ( + enable_full_determinism, + require_torch_gpu, + slow, + torch_device, +) +from diffusers.utils.torch_utils import randn_tensor + +from ..test_pipelines_common import PipelineTesterMixin + + +enable_full_determinism() + + +class FluxControlNetPipelineFastTests(unittest.TestCase, PipelineTesterMixin): + pipeline_class = FluxControlNetPipeline + + params = frozenset(["prompt", "height", "width", "guidance_scale", "prompt_embeds", "pooled_prompt_embeds"]) + batch_params = frozenset(["prompt"]) + + def get_dummy_components(self): + torch.manual_seed(0) + transformer = FluxTransformer2DModel( + patch_size=1, + in_channels=16, + num_layers=1, + num_single_layers=1, + attention_head_dim=16, + num_attention_heads=2, + joint_attention_dim=32, + pooled_projection_dim=32, + axes_dims_rope=[4, 4, 8], + ) + + torch.manual_seed(0) + controlnet = FluxControlNetModel( + patch_size=1, + in_channels=16, + num_layers=1, + num_single_layers=1, + attention_head_dim=16, + num_attention_heads=2, + joint_attention_dim=32, + pooled_projection_dim=32, + axes_dims_rope=[4, 4, 8], + ) + + clip_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, + hidden_act="gelu", + projection_dim=32, + ) + torch.manual_seed(0) + text_encoder = CLIPTextModel(clip_text_encoder_config) + + torch.manual_seed(0) + text_encoder_2 = T5EncoderModel.from_pretrained("hf-internal-testing/tiny-random-t5") + + tokenizer = CLIPTokenizer.from_pretrained("hf-internal-testing/tiny-random-clip") + tokenizer_2 = T5TokenizerFast.from_pretrained("hf-internal-testing/tiny-random-t5") + + torch.manual_seed(0) + vae = AutoencoderKL( + sample_size=32, + in_channels=3, + out_channels=3, + block_out_channels=(4,), + layers_per_block=1, + latent_channels=4, + norm_num_groups=1, + use_quant_conv=False, + use_post_quant_conv=False, + shift_factor=0.0609, + scaling_factor=1.5035, + ) + + scheduler = FlowMatchEulerDiscreteScheduler() + + return { + "scheduler": scheduler, + "text_encoder": text_encoder, + "text_encoder_2": text_encoder_2, + "tokenizer": tokenizer, + "tokenizer_2": tokenizer_2, + "transformer": transformer, + "vae": vae, + "controlnet": controlnet, + } + + def get_dummy_inputs(self, device, seed=0): + if str(device).startswith("mps"): + generator = torch.manual_seed(seed) + else: + generator = torch.Generator(device="cpu").manual_seed(seed) + + control_image = randn_tensor( + (1, 3, 32, 32), + generator=generator, + device=torch.device(device), + dtype=torch.float16, + ) + + controlnet_conditioning_scale = 0.5 + + inputs = { + "prompt": "A painting of a squirrel eating a burger", + "generator": generator, + "num_inference_steps": 2, + "guidance_scale": 3.5, + "output_type": "np", + "control_image": control_image, + "controlnet_conditioning_scale": controlnet_conditioning_scale, + } + + return inputs + + def test_controlnet_flux(self): + components = self.get_dummy_components() + flux_pipe = FluxControlNetPipeline(**components) + flux_pipe = flux_pipe.to(torch_device, dtype=torch.float16) + flux_pipe.set_progress_bar_config(disable=None) + + inputs = self.get_dummy_inputs(torch_device) + output = flux_pipe(**inputs) + image = output.images + + image_slice = image[0, -3:, -3:, -1] + + assert image.shape == (1, 32, 32, 3) + + expected_slice = np.array( + [0.7348633, 0.41333008, 0.6621094, 0.5444336, 0.47607422, 0.5859375, 0.44677734, 0.4506836, 0.40454102] + ) + + assert ( + np.abs(image_slice.flatten() - expected_slice).max() < 1e-2 + ), f"Expected: {expected_slice}, got: {image_slice.flatten()}" + + @unittest.skip("xFormersAttnProcessor does not work with SD3 Joint Attention") + def test_xformers_attention_forwardGenerator_pass(self): + pass + + +@slow +@require_torch_gpu +class FluxControlNetPipelineSlowTests(unittest.TestCase): + pipeline_class = FluxControlNetPipeline + + def setUp(self): + super().setUp() + gc.collect() + torch.cuda.empty_cache() + + def tearDown(self): + super().tearDown() + gc.collect() + torch.cuda.empty_cache() + + def test_canny(self): + controlnet = FluxControlNetModel.from_pretrained( + "InstantX/FLUX.1-dev-Controlnet-Canny-alpha", torch_dtype=torch.bfloat16 + ) + pipe = FluxControlNetPipeline.from_pretrained( + "black-forest-labs/FLUX.1-dev", controlnet=controlnet, torch_dtype=torch.bfloat16 + ) + pipe.enable_model_cpu_offload() + pipe.set_progress_bar_config(disable=None) + + generator = torch.Generator(device="cpu").manual_seed(0) + prompt = "A girl in city, 25 years old, cool, futuristic" + control_image = load_image( + "https://huggingface.co/InstantX/FLUX.1-dev-Controlnet-Canny-alpha/resolve/main/canny.jpg" + ) + + output = pipe( + prompt, + control_image=control_image, + controlnet_conditioning_scale=0.6, + num_inference_steps=2, + guidance_scale=3.5, + output_type="np", + generator=generator, + ) + + image = output.images[0] + + assert image.shape == (1024, 1024, 3) + + original_image = image[-3:, -3:, -1].flatten() + + expected_image = np.array( + [0.33007812, 0.33984375, 0.33984375, 0.328125, 0.34179688, 0.33984375, 0.30859375, 0.3203125, 0.3203125] + ) + + assert np.abs(original_image.flatten() - expected_image).max() < 1e-2