From 6fb6f2bbcdff25e3abd83bb0371e29cae849c376 Mon Sep 17 00:00:00 2001
From: AstraliteHeart <81396681+AstraliteHeart@users.noreply.github.com>
Date: Wed, 16 Apr 2025 04:04:51 -0700
Subject: [PATCH 1/5] Add basic implementation for AuraFlowImg2ImgPipeline
---
src/diffusers/__init__.py | 1 +
src/diffusers/pipelines/__init__.py | 4 +-
src/diffusers/pipelines/aura_flow/__init__.py | 2 +
.../aura_flow/pipeline_aura_flow_img2img.py | 703 ++++++++++++++++++
src/diffusers/pipelines/auto_pipeline.py | 3 +-
.../test_pipeline_aura_flow_img2img.py | 117 +++
6 files changed, 827 insertions(+), 3 deletions(-)
create mode 100644 src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
create mode 100644 tests/pipelines/aura_flow/test_pipeline_aura_flow_img2img.py
diff --git a/src/diffusers/__init__.py b/src/diffusers/__init__.py
index f51a4ef2b3f6..fe2ff12dc4c8 100644
--- a/src/diffusers/__init__.py
+++ b/src/diffusers/__init__.py
@@ -346,6 +346,7 @@
"AudioLDM2UNet2DConditionModel",
"AudioLDMPipeline",
"AuraFlowPipeline",
+ "AuraFlowImg2ImgPipeline",
"BlipDiffusionControlNetPipeline",
"BlipDiffusionPipeline",
"CLIPImageProjection",
diff --git a/src/diffusers/pipelines/__init__.py b/src/diffusers/pipelines/__init__.py
index 011f23ed371c..a42b0b34bb83 100644
--- a/src/diffusers/pipelines/__init__.py
+++ b/src/diffusers/pipelines/__init__.py
@@ -309,7 +309,7 @@
"StableDiffusionLDM3DPipeline",
]
)
- _import_structure["aura_flow"] = ["AuraFlowPipeline"]
+ _import_structure["aura_flow"] = ["AuraFlowPipeline", "AuraFlowImg2ImgPipeline"]
_import_structure["stable_diffusion_3"] = [
"StableDiffusion3Pipeline",
"StableDiffusion3Img2ImgPipeline",
@@ -515,7 +515,7 @@
AudioLDM2ProjectionModel,
AudioLDM2UNet2DConditionModel,
)
- from .aura_flow import AuraFlowPipeline
+ from .aura_flow import AuraFlowPipeline, AuraFlowImg2ImgPipeline
from .blip_diffusion import BlipDiffusionPipeline
from .cogvideo import (
CogVideoXFunControlPipeline,
diff --git a/src/diffusers/pipelines/aura_flow/__init__.py b/src/diffusers/pipelines/aura_flow/__init__.py
index e1917baa61e2..ad4974bfeae3 100644
--- a/src/diffusers/pipelines/aura_flow/__init__.py
+++ b/src/diffusers/pipelines/aura_flow/__init__.py
@@ -23,6 +23,7 @@
_dummy_objects.update(get_objects_from_module(dummy_torch_and_transformers_objects))
else:
_import_structure["pipeline_aura_flow"] = ["AuraFlowPipeline"]
+ _import_structure["pipeline_aura_flow_img2img"] = ["AuraFlowImg2ImgPipeline"]
if TYPE_CHECKING or DIFFUSERS_SLOW_IMPORT:
try:
@@ -33,6 +34,7 @@
from ...utils.dummy_torch_and_transformers_objects import *
else:
from .pipeline_aura_flow import AuraFlowPipeline
+ from .pipeline_aura_flow_img2img import AuraFlowImg2ImgPipeline
else:
import sys
diff --git a/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
new file mode 100644
index 000000000000..1e91af7bc708
--- /dev/null
+++ b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
@@ -0,0 +1,703 @@
+# Copyright 2025 AuraFlow Authors 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.
+from typing import Callable, Dict, List, Optional, Tuple, Union
+
+import PIL
+import torch
+from transformers import T5Tokenizer, UMT5EncoderModel
+
+from diffusers.callbacks import MultiPipelineCallbacks, PipelineCallback
+from diffusers.image_processor import VaeImageProcessor
+from diffusers.models import AuraFlowTransformer2DModel, AutoencoderKL
+from diffusers.models.attention_processor import AttnProcessor2_0, FusedAttnProcessor2_0, XFormersAttnProcessor
+from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
+from diffusers.utils import is_torch_xla_available, logging, replace_example_docstring
+from diffusers.utils.torch_utils import randn_tensor
+from diffusers.pipelines.pipeline_utils import DiffusionPipeline, ImagePipelineOutput
+
+
+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 import AuraFlowImg2ImgPipeline
+ >>> import requests
+ >>> from PIL import Image
+ >>> from io import BytesIO
+
+ >>> # download an initial image
+ >>> url = "https://raw.githubusercontent.com/CompVis/stable-diffusion/main/assets/stable-samples/img2img/sketch-mountains-input.jpg"
+ >>> response = requests.get(url)
+ >>> init_image = Image.open(BytesIO(response.content)).convert("RGB")
+ >>> init_image = init_image.resize((768, 512))
+
+ >>> pipe = AuraFlowImg2ImgPipeline.from_pretrained("fal/AuraFlow", torch_dtype=torch.float16)
+ >>> pipe = pipe.to("cuda")
+ >>> prompt = "A fantasy landscape, trending on artstation"
+ >>> image = pipe(prompt=prompt, image=init_image, strength=0.75, num_inference_steps=50).images[0]
+ >>> image.save("aura_flow_img2img.png")
+ ```
+"""
+
+
+class AuraFlowImg2ImgPipeline(DiffusionPipeline):
+ r"""
+ Args:
+ tokenizer (`T5TokenizerFast`):
+ Tokenizer of class
+ [T5Tokenizer](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5Tokenizer).
+ text_encoder ([`T5EncoderModel`]):
+ Frozen text-encoder. AuraFlow uses
+ [T5](https://huggingface.co/docs/transformers/model_doc/t5#transformers.T5EncoderModel), specifically the
+ [EleutherAI/pile-t5-xl](https://huggingface.co/EleutherAI/pile-t5-xl) variant.
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+ transformer ([`AuraFlowTransformer2DModel`]):
+ Conditional Transformer (MMDiT and DiT) architecture to denoise the encoded image latents.
+ scheduler ([`FlowMatchEulerDiscreteScheduler`]):
+ A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
+ """
+
+ _optional_components = []
+ model_cpu_offload_seq = "text_encoder->transformer->vae"
+ _callback_tensor_inputs = [
+ "latents",
+ "prompt_embeds",
+ ]
+
+ def __init__(
+ self,
+ tokenizer: T5Tokenizer,
+ text_encoder: UMT5EncoderModel,
+ vae: AutoencoderKL,
+ transformer: AuraFlowTransformer2DModel,
+ scheduler: FlowMatchEulerDiscreteScheduler,
+ ):
+ super().__init__()
+
+ self.register_modules(
+ tokenizer=tokenizer, text_encoder=text_encoder, vae=vae, transformer=transformer, scheduler=scheduler
+ )
+
+ self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
+ self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+
+ def check_inputs(
+ self,
+ prompt,
+ height,
+ width,
+ strength,
+ image,
+ negative_prompt,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ prompt_attention_mask=None,
+ negative_prompt_attention_mask=None,
+ callback_on_step_end_tensor_inputs=None,
+ ):
+ if strength < 0 or strength > 1:
+ raise ValueError(f"The value of strength should be in [0.0, 1.0] but is {strength}")
+
+ patch_size = 2 # AuraFlow uses patch size of 2
+ required_divisor = self.vae_scale_factor * patch_size
+ if height % required_divisor != 0 or width % required_divisor != 0:
+ raise ValueError(
+ f"\`height\` and \`width\` have to be divisible by the VAE scale factor ({self.vae_scale_factor}) times the transformer patch size ({patch_size}), which is {required_divisor}. "
+ f"Your dimensions are ({height}, {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 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)}")
+
+ if prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if negative_prompt is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt`: {negative_prompt} and `negative_prompt_embeds`:"
+ f" {negative_prompt_embeds}. Please make sure to only forward one of the two."
+ )
+
+ if prompt_embeds is not None and prompt_attention_mask is None:
+ raise ValueError("Must provide `prompt_attention_mask` when specifying `prompt_embeds`.")
+
+ if negative_prompt_embeds is not None and negative_prompt_attention_mask is None:
+ raise ValueError("Must provide `negative_prompt_attention_mask` when specifying `negative_prompt_embeds`.")
+
+ if prompt_embeds is not None and negative_prompt_embeds is not None:
+ if prompt_embeds.shape != negative_prompt_embeds.shape:
+ raise ValueError(
+ "`prompt_embeds` and `negative_prompt_embeds` must have the same shape when passed directly, but"
+ f" got: `prompt_embeds` {prompt_embeds.shape} != `negative_prompt_embeds`"
+ f" {negative_prompt_embeds.shape}."
+ )
+ if prompt_attention_mask.shape != negative_prompt_attention_mask.shape:
+ raise ValueError(
+ "`prompt_attention_mask` and `negative_prompt_attention_mask` must have the same shape when passed directly, but"
+ f" got: `prompt_attention_mask` {prompt_attention_mask.shape} != `negative_prompt_attention_mask`"
+ f" {negative_prompt_attention_mask.shape}."
+ )
+
+ def encode_prompt(
+ self,
+ prompt: Union[str, List[str]],
+ negative_prompt: Union[str, List[str]] = None,
+ do_classifier_free_guidance: bool = True,
+ num_images_per_prompt: int = 1,
+ device: Optional[torch.device] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ prompt_attention_mask: Optional[torch.Tensor] = None,
+ negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+ max_sequence_length: int = 256,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt not to guide the image generation. If not defined, one has to pass `negative_prompt_embeds`
+ instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is less than `1`).
+ do_classifier_free_guidance (`bool`, *optional*, defaults to `True`):
+ whether to use classifier free guidance or not
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ number of images that should be generated per prompt
+ device: (`torch.device`, *optional*):
+ torch device to place the resulting embeddings on
+ prompt_embeds (`torch.Tensor`, *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.
+ prompt_attention_mask (`torch.Tensor`, *optional*):
+ Pre-generated attention mask for text embeddings.
+ negative_prompt_embeds (`torch.Tensor`, *optional*):
+ Pre-generated negative text embeddings.
+ negative_prompt_attention_mask (`torch.Tensor`, *optional*):
+ Pre-generated attention mask for negative text embeddings.
+ max_sequence_length (`int`, defaults to 256): Maximum sequence length to use for the prompt.
+ """
+ if device is None:
+ device = self._execution_device
+
+ 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]
+
+ max_length = max_sequence_length
+ if prompt_embeds is None:
+ text_inputs = self.tokenizer(
+ prompt,
+ truncation=True,
+ max_length=max_length,
+ padding="max_length",
+ 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[:, max_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because T5 can only handle sequences up to"
+ f" {max_length} tokens: {removed_text}"
+ )
+
+ text_inputs = {k: v.to(device) for k, v in text_inputs.items()}
+ prompt_embeds = self.text_encoder(**text_inputs)[0]
+ prompt_attention_mask = text_inputs["attention_mask"].unsqueeze(-1).expand(prompt_embeds.shape)
+ prompt_embeds = prompt_embeds * prompt_attention_mask
+
+ if self.text_encoder is not None:
+ dtype = self.text_encoder.dtype
+ elif self.transformer is not None:
+ dtype = self.transformer.dtype
+ else:
+ dtype = None
+
+ prompt_embeds = prompt_embeds.to(dtype=dtype, device=device)
+
+ bs_embed, 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(bs_embed * num_images_per_prompt, seq_len, -1)
+ prompt_attention_mask = prompt_attention_mask.reshape(bs_embed, -1)
+ prompt_attention_mask = prompt_attention_mask.repeat(num_images_per_prompt, 1)
+
+ # get unconditional embeddings for classifier free guidance
+ if do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ uncond_tokens = [negative_prompt] * batch_size if isinstance(negative_prompt, str) else negative_prompt
+ max_length = prompt_embeds.shape[1]
+ uncond_input = self.tokenizer(
+ uncond_tokens,
+ truncation=True,
+ max_length=max_length,
+ padding="max_length",
+ return_tensors="pt",
+ )
+ uncond_input = {k: v.to(device) for k, v in uncond_input.items()}
+ negative_prompt_embeds = self.text_encoder(**uncond_input)[0]
+ negative_prompt_attention_mask = (
+ uncond_input["attention_mask"].unsqueeze(-1).expand(negative_prompt_embeds.shape)
+ )
+ negative_prompt_embeds = negative_prompt_embeds * negative_prompt_attention_mask
+
+ if do_classifier_free_guidance:
+ # duplicate unconditional embeddings for each generation per prompt, using mps friendly method
+ seq_len = negative_prompt_embeds.shape[1]
+
+ negative_prompt_embeds = negative_prompt_embeds.to(dtype=dtype, device=device)
+
+ negative_prompt_embeds = negative_prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ negative_prompt_embeds = negative_prompt_embeds.view(batch_size * num_images_per_prompt, seq_len, -1)
+
+ negative_prompt_attention_mask = negative_prompt_attention_mask.reshape(bs_embed, -1)
+ negative_prompt_attention_mask = negative_prompt_attention_mask.repeat(num_images_per_prompt, 1)
+ else:
+ negative_prompt_embeds = None
+ negative_prompt_attention_mask = None
+
+ return prompt_embeds, prompt_attention_mask, negative_prompt_embeds, negative_prompt_attention_mask
+
+ # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3.StableDiffusion3Pipeline.prepare_latents
+ def prepare_latents(
+ self,
+ batch_size,
+ num_channels_latents,
+ height,
+ width,
+ dtype,
+ device,
+ generator,
+ latents=None,
+ ):
+ if latents is not None:
+ return latents.to(device=device, dtype=dtype)
+
+ shape = (
+ batch_size,
+ num_channels_latents,
+ int(height) // self.vae_scale_factor,
+ int(width) // self.vae_scale_factor,
+ )
+
+ 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)
+
+ return latents
+
+ def get_timesteps(self, num_inference_steps, strength, device):
+ # 1. Call set_timesteps with num_inference_steps
+ self.scheduler.set_timesteps(num_inference_steps, device=device) # Ensure scheduler uses the correct number of steps
+
+ # 2. Calculate strength-based number of steps and offset
+ init_timestep_count = min(int(num_inference_steps * strength), num_inference_steps)
+ t_start = max(num_inference_steps - init_timestep_count, 0)
+
+ # 3. Get the timesteps *after* set_timesteps has been called (now has length num_inference_steps)
+ timesteps = self.scheduler.timesteps[t_start:]
+
+ # 4. Return the correct slice and the number of actual steps
+ num_actual_inference_steps = len(timesteps)
+ return timesteps, num_actual_inference_steps
+
+ def prepare_img2img_latents(
+ self,
+ image,
+ timestep,
+ batch_size,
+ num_images_per_prompt,
+ dtype,
+ device,
+ generator=None
+ ):
+ if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
+ raise ValueError(
+ f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
+ )
+
+ image = image.to(device=device, dtype=dtype)
+
+ batch_size = batch_size * num_images_per_prompt
+
+ if image.shape[1] == 4:
+ latents = image
+ else:
+ 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."
+ )
+
+ if image.shape[0] == 1:
+ image = image.repeat(batch_size, 1, 1, 1)
+
+ # encode the init image into latents and scale the latents
+ latents = self.vae.encode(image).latent_dist.sample(generator=generator)
+ latents = latents * self.vae.config.scaling_factor
+
+ # get the original timestep using init_timestep
+ init_timestep = timestep
+
+ # add noise to latents using the timesteps
+ noise = torch.randn(latents.shape, generator=generator, device=device, dtype=dtype)
+
+ # Ensure timestep tensor is on the same device
+ t = init_timestep.to(latents.device)
+
+ # Normalize timestep to [0, 1] range (using scheduler's config)
+ t = t / self.scheduler.config.num_train_timesteps
+
+ # Reshape t to match the dimensions needed for broadcasting
+ required_dims = len(latents.shape)
+ current_dims = len(t.shape)
+ for _ in range(required_dims - current_dims):
+ t = t.unsqueeze(-1)
+
+ # Interpolation: x_t = t * x_1 + (1 - t) * x_0
+ latents = t * noise + (1 - t) * latents
+
+ return latents
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
+ def upcast_vae(self):
+ dtype = self.vae.dtype
+ self.vae.to(dtype=torch.float32)
+ use_torch_2_0_or_xformers = isinstance(
+ self.vae.decoder.mid_block.attentions[0].processor,
+ (
+ AttnProcessor2_0,
+ XFormersAttnProcessor,
+ FusedAttnProcessor2_0,
+ ),
+ )
+ # if xformers or torch_2_0 is used attention block does not need
+ # to be in float32 which can save lots of memory
+ if use_torch_2_0_or_xformers:
+ self.vae.post_quant_conv.to(dtype)
+ self.vae.decoder.conv_in.to(dtype)
+ self.vae.decoder.mid_block.to(dtype)
+
+ @property
+ def guidance_scale(self):
+ return self._guidance_scale
+
+ @property
+ def do_classifier_free_guidance(self):
+ return self._guidance_scale > 1.0
+
+ @property
+ def num_timesteps(self):
+ return self._num_timesteps
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ image: Optional[Union[torch.Tensor, PIL.Image.Image]] = None,
+ strength: float = 0.8,
+ negative_prompt: Union[str, List[str]] = None,
+ num_inference_steps: int = 50,
+ sigmas: List[float] = None,
+ guidance_scale: float = 3.5,
+ num_images_per_prompt: Optional[int] = 1,
+ height: Optional[int] = None,
+ width: Optional[int] = None,
+ generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
+ latents: Optional[torch.Tensor] = None,
+ prompt_embeds: Optional[torch.Tensor] = None,
+ prompt_attention_mask: Optional[torch.Tensor] = None,
+ negative_prompt_embeds: Optional[torch.Tensor] = None,
+ negative_prompt_attention_mask: Optional[torch.Tensor] = None,
+ max_sequence_length: int = 256,
+ output_type: Optional[str] = "pil",
+ return_dict: bool = True,
+ callback_on_step_end: Optional[
+ Union[Callable[[int, int, Dict], None], PipelineCallback, MultiPipelineCallbacks]
+ ] = None,
+ callback_on_step_end_tensor_inputs: List[str] = ["latents"],
+ ) -> Union[ImagePipelineOutput, Tuple]:
+ 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.
+ image (`torch.Tensor` or `PIL.Image.Image`):
+ `Image`, or tensor representing an image batch, that will be used as the starting point for the
+ process. This is the image whose style you want to transfer.
+ strength (`float`, *optional*, defaults to 0.8):
+ Conceptually, indicates how much to transform the reference `image`. Must be between 0 and 1. `image`
+ will be used as a starting point, adding more noise to it the larger the `strength`. The number of
+ denoising steps depends on the amount of noise initially added. When `strength` is 1, added noise will
+ be maximum and the denoising process will run for the full number of iterations specified in
+ `num_inference_steps`. A value of 1, therefore, essentially ignores `image`.
+ negative_prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation. If not defined, one has to pass
+ `negative_prompt_embeds` instead. Ignored when not using guidance (i.e., ignored if `guidance_scale` is
+ less than `1`).
+ height (`int`, *optional*, defaults to self.transformer.config.sample_size * self.vae_scale_factor):
+ The height in pixels of the generated image. This is set to 1024 by default for best results.
+ width (`int`, *optional*, defaults to self.transformer.config.sample_size * self.vae_scale_factor):
+ The width in pixels of the generated image. This is set to 1024 by default for 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.
+ 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`.
+ guidance_scale (`float`, *optional*, defaults to 5.0):
+ 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.
+ prompt_attention_mask (`torch.Tensor`, *optional*):
+ Pre-generated attention mask for text embeddings.
+ negative_prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated negative text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, negative_prompt_embeds will be generated from `negative_prompt` input
+ argument.
+ negative_prompt_attention_mask (`torch.Tensor`, *optional*):
+ Pre-generated attention mask for negative text embeddings.
+ 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.stable_diffusion_xl.StableDiffusionXLPipelineOutput`] instead
+ of a plain tuple.
+ 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 256): Maximum sequence length to use with the `prompt`.
+
+ Examples:
+
+ Returns: [`~pipelines.ImagePipelineOutput`] or `tuple`:
+ If `return_dict` is `True`, [`~pipelines.ImagePipelineOutput`] is returned, otherwise a `tuple` is returned
+ where the first element is a list with the generated images.
+ """
+ # 0. Default height and width to transformer config
+ height = height or self.transformer.config.sample_size * self.vae_scale_factor
+ width = width or self.transformer.config.sample_size * self.vae_scale_factor
+
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ height,
+ width,
+ strength,
+ image,
+ negative_prompt,
+ prompt_embeds,
+ negative_prompt_embeds,
+ prompt_attention_mask,
+ negative_prompt_attention_mask,
+ callback_on_step_end_tensor_inputs=callback_on_step_end_tensor_inputs,
+ )
+
+ self._guidance_scale = guidance_scale
+ self._num_inference_steps = num_inference_steps
+
+ # 2. Preprocess image
+ image = self.image_processor.preprocess(image)
+
+ # 3. Determine batch size.
+ 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
+
+ # here `guidance_scale` is defined analog to the guidance weight `w` of equation (2)
+ # of the Imagen paper: https://arxiv.org/pdf/2205.11487.pdf . `guidance_scale = 1`
+ # corresponds to doing no classifier free guidance.
+ do_classifier_free_guidance = guidance_scale > 1.0
+
+ # 4. Encode input prompt
+ (
+ prompt_embeds,
+ prompt_attention_mask,
+ negative_prompt_embeds,
+ negative_prompt_attention_mask,
+ ) = self.encode_prompt(
+ prompt=prompt,
+ negative_prompt=negative_prompt,
+ do_classifier_free_guidance=do_classifier_free_guidance,
+ num_images_per_prompt=num_images_per_prompt,
+ device=device,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ prompt_attention_mask=prompt_attention_mask,
+ negative_prompt_attention_mask=negative_prompt_attention_mask,
+ max_sequence_length=max_sequence_length,
+ )
+ if do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+
+ # 5. Prepare timesteps
+ timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
+ latent_timestep = timesteps[:1]
+
+ # 6. Prepare latent variables
+ latents = self.prepare_img2img_latents(
+ image,
+ latent_timestep,
+ batch_size,
+ num_images_per_prompt,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ )
+
+ # 7. Denoising loop
+ num_warmup_steps = len(timesteps) - num_inference_steps * self.scheduler.order
+ self._num_timesteps = len(timesteps)
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+
+ # aura use timestep value between 0 and 1, with t=1 as noise and t=0 as the image
+ # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
+ timestep = torch.tensor([t / 1000]).expand(latent_model_input.shape[0])
+ timestep = timestep.to(latents.device, dtype=latents.dtype)
+
+ # Make sure latent_model_input has the same dtype as the transformer
+ transformer_dtype = self.transformer.dtype
+ if latent_model_input.dtype != transformer_dtype:
+ latent_model_input = latent_model_input.to(dtype=transformer_dtype)
+
+ # predict noise model_output
+ noise_pred = self.transformer(
+ latent_model_input,
+ encoder_hidden_states=prompt_embeds,
+ timestep=timestep,
+ return_dict=False,
+ )[0]
+
+ # perform guidance
+ if do_classifier_free_guidance:
+ noise_pred_uncond, noise_pred_text = noise_pred.chunk(2)
+ noise_pred = noise_pred_uncond + guidance_scale * (noise_pred_text - noise_pred_uncond)
+
+ # compute the previous noisy sample x_t -> x_t-1
+ latents = self.scheduler.step(noise_pred, t, latents, return_dict=False)[0]
+
+ 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:
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast
+ if needs_upcasting:
+ self.upcast_vae()
+ latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+
+ # Apply proper scaling factor and shift factor if available
+ if hasattr(self.vae.config, "scaling_factor") and hasattr(self.vae.config, "shift_factor") and getattr(self.vae.config, "shift_factor", None) is not None:
+ # Handle both scaling and shifting
+ latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor
+ else:
+ # Just scale using standard approach
+ latents = latents / self.vae.config.scaling_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 ImagePipelineOutput(images=image)
\ No newline at end of file
diff --git a/src/diffusers/pipelines/auto_pipeline.py b/src/diffusers/pipelines/auto_pipeline.py
index 6a5f6098b6fb..56d25053123d 100644
--- a/src/diffusers/pipelines/auto_pipeline.py
+++ b/src/diffusers/pipelines/auto_pipeline.py
@@ -20,7 +20,7 @@
from ..configuration_utils import ConfigMixin
from ..models.controlnets import ControlNetUnionModel
from ..utils import is_sentencepiece_available
-from .aura_flow import AuraFlowPipeline
+from .aura_flow import AuraFlowPipeline, AuraFlowImg2ImgPipeline
from .cogview3 import CogView3PlusPipeline
from .cogview4 import CogView4ControlPipeline, CogView4Pipeline
from .controlnet import (
@@ -165,6 +165,7 @@
("stable-diffusion-xl-controlnet-union", StableDiffusionXLControlNetUnionImg2ImgPipeline),
("stable-diffusion-xl-pag", StableDiffusionXLPAGImg2ImgPipeline),
("stable-diffusion-xl-controlnet-pag", StableDiffusionXLControlNetPAGImg2ImgPipeline),
+ ("auraflow", AuraFlowImg2ImgPipeline),
("lcm", LatentConsistencyModelImg2ImgPipeline),
("flux", FluxImg2ImgPipeline),
("flux-controlnet", FluxControlNetImg2ImgPipeline),
diff --git a/tests/pipelines/aura_flow/test_pipeline_aura_flow_img2img.py b/tests/pipelines/aura_flow/test_pipeline_aura_flow_img2img.py
new file mode 100644
index 000000000000..734885f0eaca
--- /dev/null
+++ b/tests/pipelines/aura_flow/test_pipeline_aura_flow_img2img.py
@@ -0,0 +1,117 @@
+import unittest
+
+import numpy as np
+import PIL.Image
+import torch
+from diffusers.utils.testing_utils import require_torch_gpu, torch_device
+from transformers import AutoTokenizer, UMT5EncoderModel, AuraFlowPipelineFastTests
+
+from diffusers import (
+ AuraFlowImg2ImgPipeline, # Added for Img2Img
+ AuraFlowPipeline,
+ AuraFlowTransformer2DModel,
+ AutoencoderKL,
+ FlowMatchEulerDiscreteScheduler,
+)
+
+from ..test_pipelines_common import (
+ PipelineTesterMixin,
+ check_qkv_fusion_matches_attn_procs_length,
+ check_qkv_fusion_processors_exist,
+)
+
+class AuraFlowImg2ImgPipelineFastTests(AuraFlowPipelineFastTests):
+ pipeline_class = AuraFlowImg2ImgPipeline
+ params = frozenset(
+ [
+ "prompt",
+ "image",
+ "strength",
+ "guidance_scale",
+ "negative_prompt",
+ "prompt_embeds",
+ "negative_prompt_embeds",
+ ]
+ )
+ batch_params = frozenset(["prompt", "negative_prompt", "image"])
+ test_layerwise_casting = False # T5 uses multiple devices
+ test_group_offloading = False # T5 uses multiple devices
+
+ # Redefine get_dummy_inputs for Img2Img
+ def get_dummy_inputs(self, device, seed=0):
+ # Ensure image dimensions are divisible by VAE scale factor * transformer patch size
+ # vae_scale_factor = 8, patch_size = 2 => divisible by 16
+ image = PIL.Image.new("RGB", (64, 64))
+ if str(device).startswith("mps"):
+ generator = torch.manual_seed(seed)
+ else:
+ generator = torch.Generator(device="cpu").manual_seed(seed)
+
+ inputs = {
+ "prompt": "A painting of a squirrel eating a burger",
+ "image": image,
+ "strength": 0.75,
+ "generator": generator,
+ "num_inference_steps": 2,
+ "guidance_scale": 5.0,
+ "output_type": "np",
+ # height/width are inferred from image in img2img
+ }
+ return inputs
+
+ # Override T2I test that requires height/width
+ def test_fused_qkv_projections(self):
+ # Inherited test expects height/width, skip for img2img dummy inputs
+ # Call the parent T2I test method directly if needed for coverage,
+ # but adapt inputs or skip if incompatible.
+ # For now, simply reimplement with img2img inputs
+ device = "cpu" # ensure determinism for the device-dependent torch.Generator
+ components = self.get_dummy_components()
+ pipe = self.pipeline_class(**components)
+ pipe = pipe.to(device)
+ pipe.set_progress_bar_config(disable=None)
+
+ inputs = self.get_dummy_inputs(device)
+ image = pipe(**inputs).images
+ original_image_slice = image[0, -3:, -3:, -1]
+
+ pipe.transformer.fuse_qkv_projections()
+ assert check_qkv_fusion_processors_exist(pipe.transformer)
+ assert check_qkv_fusion_matches_attn_procs_length(
+ pipe.transformer, pipe.transformer.original_attn_processors
+ )
+
+ inputs = self.get_dummy_inputs(device)
+ image = pipe(**inputs).images
+ image_slice_fused = image[0, -3:, -3:, -1]
+
+ pipe.transformer.unfuse_qkv_projections()
+ inputs = self.get_dummy_inputs(device)
+ image = pipe(**inputs).images
+ image_slice_disabled = image[0, -3:, -3:, -1]
+
+ assert np.allclose(original_image_slice, image_slice_fused, atol=1e-3, rtol=1e-3)
+ assert np.allclose(image_slice_fused, image_slice_disabled, atol=1e-3, rtol=1e-3)
+ assert np.allclose(original_image_slice, image_slice_disabled, atol=1e-2, rtol=1e-2)
+
+
+ def test_aura_flow_img2img_output_shape(self):
+ components = self.get_dummy_components()
+ pipe = self.pipeline_class(**components).to(torch_device)
+
+ # Use dimensions divisible by vae_scale_factor * patch_size (8*2=16)
+ height_width_pairs = [(64, 64), (128, 48)] # 48 is divisible by 16
+
+ for height, width in height_width_pairs:
+ inputs = self.get_dummy_inputs(torch_device)
+ # Override dummy image size
+ inputs["image"] = PIL.Image.new("RGB", (width, height))
+ # Pass height/width explicitly to test pipeline handles them (though inferred by default)
+ inputs["height"] = height
+ inputs["width"] = width
+
+ output = pipe(**inputs)
+ image = output.images[0]
+
+ # Expected shape is (height, width, 3) for np output
+ self.assertEqual(image.shape, (height, width, 3))
\ No newline at end of file
From 6ff1af8c20683175fadef09a77cb2f79127d53e7 Mon Sep 17 00:00:00 2001
From: AstraliteHeart <astralite.heart@gmail.com>
Date: Thu, 17 Apr 2025 19:30:50 +0000
Subject: [PATCH 2/5] Update i2i tests, fix style
---
src/diffusers/__init__.py | 2 +-
src/diffusers/pipelines/__init__.py | 2 +-
.../aura_flow/pipeline_aura_flow_img2img.py | 116 +++++++++++++-----
src/diffusers/pipelines/auto_pipeline.py | 4 +-
.../dummy_torch_and_transformers_objects.py | 15 +++
.../aura_flow/test_pipeline_aura_flow.py | 3 +
.../test_pipeline_aura_flow_img2img.py | 109 +++++++++++++---
7 files changed, 196 insertions(+), 55 deletions(-)
diff --git a/src/diffusers/__init__.py b/src/diffusers/__init__.py
index fe2ff12dc4c8..4bebc3404220 100644
--- a/src/diffusers/__init__.py
+++ b/src/diffusers/__init__.py
@@ -345,8 +345,8 @@
"AudioLDM2ProjectionModel",
"AudioLDM2UNet2DConditionModel",
"AudioLDMPipeline",
- "AuraFlowPipeline",
"AuraFlowImg2ImgPipeline",
+ "AuraFlowPipeline",
"BlipDiffusionControlNetPipeline",
"BlipDiffusionPipeline",
"CLIPImageProjection",
diff --git a/src/diffusers/pipelines/__init__.py b/src/diffusers/pipelines/__init__.py
index a42b0b34bb83..352bf804d827 100644
--- a/src/diffusers/pipelines/__init__.py
+++ b/src/diffusers/pipelines/__init__.py
@@ -515,7 +515,7 @@
AudioLDM2ProjectionModel,
AudioLDM2UNet2DConditionModel,
)
- from .aura_flow import AuraFlowPipeline, AuraFlowImg2ImgPipeline
+ from .aura_flow import AuraFlowImg2ImgPipeline, AuraFlowPipeline
from .blip_diffusion import BlipDiffusionPipeline
from .cogvideo import (
CogVideoXFunControlPipeline,
diff --git a/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
index 1e91af7bc708..2a96833f4617 100644
--- a/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
+++ b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
@@ -21,10 +21,10 @@
from diffusers.image_processor import VaeImageProcessor
from diffusers.models import AuraFlowTransformer2DModel, AutoencoderKL
from diffusers.models.attention_processor import AttnProcessor2_0, FusedAttnProcessor2_0, XFormersAttnProcessor
+from diffusers.pipelines.pipeline_utils import DiffusionPipeline, ImagePipelineOutput
from diffusers.schedulers import FlowMatchEulerDiscreteScheduler
from diffusers.utils import is_torch_xla_available, logging, replace_example_docstring
from diffusers.utils.torch_utils import randn_tensor
-from diffusers.pipelines.pipeline_utils import DiffusionPipeline, ImagePipelineOutput
if is_torch_xla_available():
@@ -119,12 +119,12 @@ def check_inputs(
):
if strength < 0 or strength > 1:
raise ValueError(f"The value of strength should be in [0.0, 1.0] but is {strength}")
-
+
patch_size = 2 # AuraFlow uses patch size of 2
required_divisor = self.vae_scale_factor * patch_size
if height % required_divisor != 0 or width % required_divisor != 0:
raise ValueError(
- f"\`height\` and \`width\` have to be divisible by the VAE scale factor ({self.vae_scale_factor}) times the transformer patch size ({patch_size}), which is {required_divisor}. "
+ rf"\`height\` and \`width\` have to be divisible by the VAE scale factor ({self.vae_scale_factor}) times the transformer patch size ({patch_size}), which is {required_divisor}. "
f"Your dimensions are ({height}, {width})."
)
@@ -339,7 +339,7 @@ def prepare_latents(
def get_timesteps(self, num_inference_steps, strength, device):
# 1. Call set_timesteps with num_inference_steps
- self.scheduler.set_timesteps(num_inference_steps, device=device) # Ensure scheduler uses the correct number of steps
+ self.scheduler.set_timesteps(num_inference_steps, device=device)
# 2. Calculate strength-based number of steps and offset
init_timestep_count = min(int(num_inference_steps * strength), num_inference_steps)
@@ -353,14 +353,7 @@ def get_timesteps(self, num_inference_steps, strength, device):
return timesteps, num_actual_inference_steps
def prepare_img2img_latents(
- self,
- image,
- timestep,
- batch_size,
- num_images_per_prompt,
- dtype,
- device,
- generator=None
+ self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None
):
if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
raise ValueError(
@@ -380,34 +373,87 @@ def prepare_img2img_latents(
f" size of {batch_size}. Make sure the batch size matches the length of the generators."
)
- if image.shape[0] == 1:
- image = image.repeat(batch_size, 1, 1, 1)
+ # Handle different batch size scenarios
+ if image.shape[0] < batch_size:
+ if batch_size % image.shape[0] == 0:
+ # Duplicate the image to match the batch size
+ additional_image_per_prompt = batch_size // image.shape[0]
+ image = torch.cat([image] * additional_image_per_prompt, dim=0)
+ else:
+ raise ValueError(
+ f"Cannot duplicate `image` of batch size {image.shape[0]} to {batch_size} text prompts."
+ f" Batch size must be divisible by the image batch size."
+ )
# encode the init image into latents and scale the latents
- latents = self.vae.encode(image).latent_dist.sample(generator=generator)
+ # 1. Get VAE distribution parameters (on device)
+ latent_dist = self.vae.encode(image).latent_dist
+ mean, std = latent_dist.mean, latent_dist.std # Already on device
+
+ # 2. Sample noise for each batch element individually if using multiple generators
+ if isinstance(generator, list):
+ sample = torch.cat(
+ [
+ torch.randn(
+ (1, *mean.shape[1:]),
+ generator=generator[i],
+ device=generator[i].device if hasattr(generator[i], "device") else "cpu",
+ dtype=mean.dtype,
+ ).to(mean.device)
+ for i in range(batch_size)
+ ]
+ )
+ else:
+ # Single generator - use its device if it has one
+ generator_device = getattr(generator, "device", "cpu") if generator is not None else "cpu"
+ noise = torch.randn(mean.shape, generator=generator, device=generator_device, dtype=mean.dtype)
+ sample = noise.to(mean.device)
+
+ # Compute latents
+ latents = mean + std * sample
+
+ # Scale latents
latents = latents * self.vae.config.scaling_factor
# get the original timestep using init_timestep
init_timestep = timestep
# add noise to latents using the timesteps
- noise = torch.randn(latents.shape, generator=generator, device=device, dtype=dtype)
-
+ # Handle noise generation with multiple generators if provided
+ if isinstance(generator, list):
+ noise = torch.cat(
+ [
+ torch.randn(
+ (1, *latents.shape[1:]),
+ generator=generator[i],
+ device=generator[i].device if hasattr(generator[i], "device") else "cpu",
+ dtype=latents.dtype,
+ ).to(latents.device)
+ for i in range(batch_size)
+ ]
+ )
+ else:
+ # Single generator - use its device if it has one
+ generator_device = getattr(generator, "device", "cpu") if generator is not None else "cpu"
+ noise = torch.randn(
+ latents.shape, generator=generator, device=generator_device, dtype=latents.dtype
+ ).to(latents.device)
+
# Ensure timestep tensor is on the same device
t = init_timestep.to(latents.device)
-
+
# Normalize timestep to [0, 1] range (using scheduler's config)
t = t / self.scheduler.config.num_train_timesteps
-
+
# Reshape t to match the dimensions needed for broadcasting
required_dims = len(latents.shape)
current_dims = len(t.shape)
for _ in range(required_dims - current_dims):
t = t.unsqueeze(-1)
-
+
# Interpolation: x_t = t * x_1 + (1 - t) * x_0
latents = t * noise + (1 - t) * latents
-
+
return latents
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
@@ -606,13 +652,14 @@ def __call__(
negative_prompt_attention_mask=negative_prompt_attention_mask,
max_sequence_length=max_sequence_length,
)
+
if do_classifier_free_guidance:
prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
# 5. Prepare timesteps
timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
latent_timestep = timesteps[:1]
-
+
# 6. Prepare latent variables
latents = self.prepare_img2img_latents(
image,
@@ -632,10 +679,13 @@ def __call__(
# expand the latents if we are doing classifier free guidance
latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
- # aura use timestep value between 0 and 1, with t=1 as noise and t=0 as the image
- # broadcast to batch dimension in a way that's compatible with ONNX/Core ML
- timestep = torch.tensor([t / 1000]).expand(latent_model_input.shape[0])
- timestep = timestep.to(latents.device, dtype=latents.dtype)
+ # AureFlow use timestep value between 0 and 1, with t=1 as noise and t=0 as the image
+ # create a timestep tensor with the correct batch size
+ # ensure it matches the batch size of the model input
+ t_float = t / 1000
+ timestep_tensor = torch.full(
+ (latent_model_input.shape[0],), t_float, device=latents.device, dtype=latents.dtype
+ )
# Make sure latent_model_input has the same dtype as the transformer
transformer_dtype = self.transformer.dtype
@@ -646,7 +696,7 @@ def __call__(
noise_pred = self.transformer(
latent_model_input,
encoder_hidden_states=prompt_embeds,
- timestep=timestep,
+ timestep=timestep_tensor,
return_dict=False,
)[0]
@@ -682,15 +732,19 @@ def __call__(
if needs_upcasting:
self.upcast_vae()
latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
-
+
# Apply proper scaling factor and shift factor if available
- if hasattr(self.vae.config, "scaling_factor") and hasattr(self.vae.config, "shift_factor") and getattr(self.vae.config, "shift_factor", None) is not None:
+ if (
+ hasattr(self.vae.config, "scaling_factor")
+ and hasattr(self.vae.config, "shift_factor")
+ and getattr(self.vae.config, "shift_factor", None) is not None
+ ):
# Handle both scaling and shifting
latents = (latents / self.vae.config.scaling_factor) + self.vae.config.shift_factor
else:
# Just scale using standard approach
latents = latents / self.vae.config.scaling_factor
-
+
image = self.vae.decode(latents, return_dict=False)[0]
image = self.image_processor.postprocess(image, output_type=output_type)
@@ -700,4 +754,4 @@ def __call__(
if not return_dict:
return (image,)
- return ImagePipelineOutput(images=image)
\ No newline at end of file
+ return ImagePipelineOutput(images=image)
diff --git a/src/diffusers/pipelines/auto_pipeline.py b/src/diffusers/pipelines/auto_pipeline.py
index 56d25053123d..ccc3413d8e55 100644
--- a/src/diffusers/pipelines/auto_pipeline.py
+++ b/src/diffusers/pipelines/auto_pipeline.py
@@ -20,7 +20,7 @@
from ..configuration_utils import ConfigMixin
from ..models.controlnets import ControlNetUnionModel
from ..utils import is_sentencepiece_available
-from .aura_flow import AuraFlowPipeline, AuraFlowImg2ImgPipeline
+from .aura_flow import AuraFlowImg2ImgPipeline, AuraFlowPipeline
from .cogview3 import CogView3PlusPipeline
from .cogview4 import CogView4ControlPipeline, CogView4Pipeline
from .controlnet import (
@@ -165,7 +165,7 @@
("stable-diffusion-xl-controlnet-union", StableDiffusionXLControlNetUnionImg2ImgPipeline),
("stable-diffusion-xl-pag", StableDiffusionXLPAGImg2ImgPipeline),
("stable-diffusion-xl-controlnet-pag", StableDiffusionXLControlNetPAGImg2ImgPipeline),
- ("auraflow", AuraFlowImg2ImgPipeline),
+ ("auraflow", AuraFlowImg2ImgPipeline),
("lcm", LatentConsistencyModelImg2ImgPipeline),
("flux", FluxImg2ImgPipeline),
("flux-controlnet", FluxControlNetImg2ImgPipeline),
diff --git a/src/diffusers/utils/dummy_torch_and_transformers_objects.py b/src/diffusers/utils/dummy_torch_and_transformers_objects.py
index b3c6efb8cdcf..05801cd3a935 100644
--- a/src/diffusers/utils/dummy_torch_and_transformers_objects.py
+++ b/src/diffusers/utils/dummy_torch_and_transformers_objects.py
@@ -257,6 +257,21 @@ def from_pretrained(cls, *args, **kwargs):
requires_backends(cls, ["torch", "transformers"])
+class AuraFlowImg2ImgPipeline(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 AuraFlowPipeline(metaclass=DummyObject):
_backends = ["torch", "transformers"]
diff --git a/tests/pipelines/aura_flow/test_pipeline_aura_flow.py b/tests/pipelines/aura_flow/test_pipeline_aura_flow.py
index 1eb9d1035c33..aeaefb527327 100644
--- a/tests/pipelines/aura_flow/test_pipeline_aura_flow.py
+++ b/tests/pipelines/aura_flow/test_pipeline_aura_flow.py
@@ -135,3 +135,6 @@ def test_fused_qkv_projections(self):
@unittest.skip("xformers attention processor does not exist for AuraFlow")
def test_xformers_attention_forwardGenerator_pass(self):
pass
+
+ def test_inference_batch_single_identical(self, batch_size=3, expected_max_diff=0.0004):
+ self._test_inference_batch_single_identical(batch_size=batch_size, expected_max_diff=expected_max_diff)
diff --git a/tests/pipelines/aura_flow/test_pipeline_aura_flow_img2img.py b/tests/pipelines/aura_flow/test_pipeline_aura_flow_img2img.py
index 734885f0eaca..3983dfb07a19 100644
--- a/tests/pipelines/aura_flow/test_pipeline_aura_flow_img2img.py
+++ b/tests/pipelines/aura_flow/test_pipeline_aura_flow_img2img.py
@@ -3,16 +3,15 @@
import numpy as np
import PIL.Image
import torch
-from diffusers.utils.testing_utils import require_torch_gpu, torch_device
-from transformers import AutoTokenizer, UMT5EncoderModel, AuraFlowPipelineFastTests
+from transformers import AutoTokenizer, UMT5EncoderModel
from diffusers import (
- AuraFlowImg2ImgPipeline, # Added for Img2Img
- AuraFlowPipeline,
+ AuraFlowImg2ImgPipeline,
AuraFlowTransformer2DModel,
AutoencoderKL,
FlowMatchEulerDiscreteScheduler,
)
+from diffusers.utils.testing_utils import torch_device
from ..test_pipelines_common import (
PipelineTesterMixin,
@@ -20,7 +19,8 @@
check_qkv_fusion_processors_exist,
)
-class AuraFlowImg2ImgPipelineFastTests(AuraFlowPipelineFastTests):
+
+class AuraFlowImg2ImgPipelineFastTests(unittest.TestCase, PipelineTesterMixin):
pipeline_class = AuraFlowImg2ImgPipeline
params = frozenset(
[
@@ -33,11 +33,50 @@ class AuraFlowImg2ImgPipelineFastTests(AuraFlowPipelineFastTests):
"negative_prompt_embeds",
]
)
- batch_params = frozenset(["prompt", "negative_prompt", "image"])
- test_layerwise_casting = False # T5 uses multiple devices
- test_group_offloading = False # T5 uses multiple devices
+ batch_params = frozenset(["prompt", "image", "negative_prompt"])
+ test_layerwise_casting = False # T5 uses multiple devices
+ test_group_offloading = False # T5 uses multiple devices
+
+ def get_dummy_components(self):
+ torch.manual_seed(0)
+ transformer = AuraFlowTransformer2DModel(
+ sample_size=32,
+ patch_size=2,
+ in_channels=4,
+ num_mmdit_layers=1,
+ num_single_dit_layers=1,
+ attention_head_dim=8,
+ num_attention_heads=4,
+ caption_projection_dim=32,
+ joint_attention_dim=32,
+ out_channels=4,
+ pos_embed_max_size=256,
+ )
+
+ text_encoder = UMT5EncoderModel.from_pretrained("hf-internal-testing/tiny-random-umt5")
+ tokenizer = AutoTokenizer.from_pretrained("hf-internal-testing/tiny-random-t5")
+
+ torch.manual_seed(0)
+ vae = AutoencoderKL(
+ block_out_channels=[32, 64],
+ in_channels=3,
+ out_channels=3,
+ down_block_types=["DownEncoderBlock2D", "DownEncoderBlock2D"],
+ up_block_types=["UpDecoderBlock2D", "UpDecoderBlock2D"],
+ latent_channels=4,
+ sample_size=32,
+ )
+
+ scheduler = FlowMatchEulerDiscreteScheduler()
+
+ return {
+ "scheduler": scheduler,
+ "text_encoder": text_encoder,
+ "tokenizer": tokenizer,
+ "transformer": transformer,
+ "vae": vae,
+ }
- # Redefine get_dummy_inputs for Img2Img
def get_dummy_inputs(self, device, seed=0):
# Ensure image dimensions are divisible by VAE scale factor * transformer patch size
# vae_scale_factor = 8, patch_size = 2 => divisible by 16
@@ -59,12 +98,12 @@ def get_dummy_inputs(self, device, seed=0):
}
return inputs
- # Override T2I test that requires height/width
+ def test_attention_slicing_forward_pass(self):
+ # Attention slicing needs to implemented differently for this because how single DiT and MMDiT
+ # blocks interfere with each other.
+ return
+
def test_fused_qkv_projections(self):
- # Inherited test expects height/width, skip for img2img dummy inputs
- # Call the parent T2I test method directly if needed for coverage,
- # but adapt inputs or skip if incompatible.
- # For now, simply reimplement with img2img inputs
device = "cpu" # ensure determinism for the device-dependent torch.Generator
components = self.get_dummy_components()
pipe = self.pipeline_class(**components)
@@ -77,9 +116,7 @@ def test_fused_qkv_projections(self):
pipe.transformer.fuse_qkv_projections()
assert check_qkv_fusion_processors_exist(pipe.transformer)
- assert check_qkv_fusion_matches_attn_procs_length(
- pipe.transformer, pipe.transformer.original_attn_processors
- )
+ assert check_qkv_fusion_matches_attn_procs_length(pipe.transformer, pipe.transformer.original_attn_processors)
inputs = self.get_dummy_inputs(device)
image = pipe(**inputs).images
@@ -94,13 +131,18 @@ def test_fused_qkv_projections(self):
assert np.allclose(image_slice_fused, image_slice_disabled, atol=1e-3, rtol=1e-3)
assert np.allclose(original_image_slice, image_slice_disabled, atol=1e-2, rtol=1e-2)
+ @unittest.skip("xformers attention processor does not exist for AuraFlow")
+ def test_xformers_attention_forwardGenerator_pass(self):
+ pass
def test_aura_flow_img2img_output_shape(self):
components = self.get_dummy_components()
pipe = self.pipeline_class(**components).to(torch_device)
- # Use dimensions divisible by vae_scale_factor * patch_size (8*2=16)
- height_width_pairs = [(64, 64), (128, 48)] # 48 is divisible by 16
+ # The positional embedding has a max size of 256
+ # Each position is a (height/vae_scale_factor/patch_size) × (width/vae_scale_factor/patch_size) grid
+ # To stay within limits: (height/8/2) * (width/8/2) < 256
+ height_width_pairs = [(32, 32), (64, 32)] # creates 4 and 16 positions respectively
for height, width in height_width_pairs:
inputs = self.get_dummy_inputs(torch_device)
@@ -114,4 +156,31 @@ def test_aura_flow_img2img_output_shape(self):
image = output.images[0]
# Expected shape is (height, width, 3) for np output
- self.assertEqual(image.shape, (height, width, 3))
\ No newline at end of file
+ self.assertEqual(image.shape, (height, width, 3))
+
+ def test_inference_batch_single_identical(self, batch_size=3, expected_max_diff=0.001):
+ self._test_inference_batch_single_identical(batch_size=batch_size, expected_max_diff=expected_max_diff)
+
+ def test_num_images_per_prompt(self):
+ components = self.get_dummy_components()
+ pipe = self.pipeline_class(**components)
+ pipe = pipe.to(torch_device)
+ pipe.set_progress_bar_config(disable=None)
+
+ batch_sizes = [1]
+ num_images_per_prompts = [1, 2]
+
+ for batch_size in batch_sizes:
+ for num_images_per_prompt in num_images_per_prompts:
+ inputs = self.get_dummy_inputs(torch_device)
+ inputs["num_inference_steps"] = 2
+
+ inputs["image"] = PIL.Image.new("RGB", (32, 32))
+
+ for key in inputs.keys():
+ if key in self.batch_params:
+ inputs[key] = batch_size * [inputs[key]]
+
+ images = pipe(**inputs, num_images_per_prompt=num_images_per_prompt).images
+
+ assert len(images) == batch_size * num_images_per_prompt
From 6ac5cbb76fc3ab3d478a704c6ab747b5cde7d29b Mon Sep 17 00:00:00 2001
From: AstraliteHeart <astralite.heart@gmail.com>
Date: Thu, 17 Apr 2025 22:52:20 +0000
Subject: [PATCH 3/5] Use scale_noise directly and fix VAE decoding
---
.../aura_flow/pipeline_aura_flow_img2img.py | 68 ++++++++++---------
1 file changed, 36 insertions(+), 32 deletions(-)
diff --git a/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
index 2a96833f4617..423ce01a270d 100644
--- a/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
+++ b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
@@ -52,7 +52,7 @@
>>> init_image = Image.open(BytesIO(response.content)).convert("RGB")
>>> init_image = init_image.resize((768, 512))
- >>> pipe = AuraFlowImg2ImgPipeline.from_pretrained("fal/AuraFlow", torch_dtype=torch.float16)
+ >>> pipe = AuraFlowImg2ImgPipeline.from_pretrained("fal/AuraFlow-v0.3", torch_dtype=torch.float16)
>>> pipe = pipe.to("cuda")
>>> prompt = "A fantasy landscape, trending on artstation"
>>> image = pipe(prompt=prompt, image=init_image, strength=0.75, num_inference_steps=50).images[0]
@@ -338,19 +338,20 @@ def prepare_latents(
return latents
def get_timesteps(self, num_inference_steps, strength, device):
- # 1. Call set_timesteps with num_inference_steps
+ # Set timesteps using the full range initially
self.scheduler.set_timesteps(num_inference_steps, device=device)
+ timesteps = self.scheduler.timesteps.to(device=device)
- # 2. Calculate strength-based number of steps and offset
- init_timestep_count = min(int(num_inference_steps * strength), num_inference_steps)
- t_start = max(num_inference_steps - init_timestep_count, 0)
+ if len(timesteps) != num_inference_steps:
+ num_inference_steps = len(timesteps) # Adjust if scheduler changed num_steps
- # 3. Get the timesteps *after* set_timesteps has been called (now has length num_inference_steps)
+ # Get the original timestep using init_timestep
+ init_timestep = min(num_inference_steps * strength, num_inference_steps)
+
+ t_start = int(max(num_inference_steps - init_timestep, 0))
timesteps = self.scheduler.timesteps[t_start:]
- # 4. Return the correct slice and the number of actual steps
- num_actual_inference_steps = len(timesteps)
- return timesteps, num_actual_inference_steps
+ return timesteps, num_inference_steps - t_start
def prepare_img2img_latents(
self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None
@@ -385,11 +386,20 @@ def prepare_img2img_latents(
f" Batch size must be divisible by the image batch size."
)
+ # Temporarily move VAE to float32 for encoding
+ vae_dtype = self.vae.dtype
+ if vae_dtype != torch.float32:
+ self.vae.to(dtype=torch.float32)
+
# encode the init image into latents and scale the latents
# 1. Get VAE distribution parameters (on device)
- latent_dist = self.vae.encode(image).latent_dist
+ latent_dist = self.vae.encode(image.to(dtype=torch.float32)).latent_dist
mean, std = latent_dist.mean, latent_dist.std # Already on device
+ # Restore VAE dtype
+ if vae_dtype != torch.float32:
+ self.vae.to(dtype=vae_dtype)
+
# 2. Sample noise for each batch element individually if using multiple generators
if isinstance(generator, list):
sample = torch.cat(
@@ -416,7 +426,7 @@ def prepare_img2img_latents(
latents = latents * self.vae.config.scaling_factor
# get the original timestep using init_timestep
- init_timestep = timestep
+ init_timestep = timestep # Use the passed timestep directly
# add noise to latents using the timesteps
# Handle noise generation with multiple generators if provided
@@ -439,20 +449,7 @@ def prepare_img2img_latents(
latents.shape, generator=generator, device=generator_device, dtype=latents.dtype
).to(latents.device)
- # Ensure timestep tensor is on the same device
- t = init_timestep.to(latents.device)
-
- # Normalize timestep to [0, 1] range (using scheduler's config)
- t = t / self.scheduler.config.num_train_timesteps
-
- # Reshape t to match the dimensions needed for broadcasting
- required_dims = len(latents.shape)
- current_dims = len(t.shape)
- for _ in range(required_dims - current_dims):
- t = t.unsqueeze(-1)
-
- # Interpolation: x_t = t * x_1 + (1 - t) * x_0
- latents = t * noise + (1 - t) * latents
+ latents = self.scheduler.scale_noise(latents, init_timestep, noise)
return latents
@@ -657,8 +654,10 @@ def __call__(
prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
# 5. Prepare timesteps
- timesteps, num_inference_steps = self.get_timesteps(num_inference_steps, strength, device)
- latent_timestep = timesteps[:1]
+ timesteps, num_inference_steps = self.get_timesteps(
+ num_inference_steps, strength, device
+ )
+ latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt) # Get the first timestep(s) for initial noise
# 6. Prepare latent variables
latents = self.prepare_img2img_latents(
@@ -727,11 +726,11 @@ def __call__(
if output_type == "latent":
image = latents
else:
- # make sure the VAE is in float32 mode, as it overflows in float16
- needs_upcasting = self.vae.dtype == torch.float16 and self.vae.config.force_upcast
- if needs_upcasting:
- self.upcast_vae()
- latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+ # Always upcast VAE to float32 for decoding
+ vae_dtype = self.vae.dtype
+ if vae_dtype != torch.float32:
+ self.vae.to(dtype=torch.float32)
+ latents = latents.to(dtype=torch.float32)
# Apply proper scaling factor and shift factor if available
if (
@@ -746,6 +745,11 @@ def __call__(
latents = latents / self.vae.config.scaling_factor
image = self.vae.decode(latents, return_dict=False)[0]
+
+ # Restore VAE dtype
+ if vae_dtype != torch.float32:
+ self.vae.to(dtype=vae_dtype)
+
image = self.image_processor.postprocess(image, output_type=output_type)
# Offload all models
From 1b7fb36294d87e7c35e3e66c334c93b83f032f5d Mon Sep 17 00:00:00 2001
From: AstraliteHeart <astralite.heart@gmail.com>
Date: Fri, 18 Apr 2025 10:29:30 +0000
Subject: [PATCH 4/5] Review updates
---
.../aura_flow/pipeline_aura_flow_img2img.py | 124 +++++++++++-------
1 file changed, 73 insertions(+), 51 deletions(-)
diff --git a/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
index 423ce01a270d..dae182c921f3 100644
--- a/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
+++ b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
@@ -103,6 +103,31 @@ def __init__(
self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1) if getattr(self, "vae", None) else 8
self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
+ @staticmethod
+ def calculate_shift(
+ image_seq_len,
+ base_seq_len: int = 256,
+ max_seq_len: int = 4096,
+ base_shift: float = 0.5,
+ max_shift: float = 1.15,
+ ):
+ """Calculate shift parameter based on image dimensions.
+
+ Args:
+ image_seq_len: Length of the image sequence (height/vae_factor/2 * width/vae_factor/2)
+ base_seq_len: Base sequence length for interpolation
+ max_seq_len: Maximum sequence length for interpolation
+ base_shift: Base shift value
+ max_shift: Maximum shift value
+
+ Returns:
+ Calculated shift parameter (mu)
+ """
+ 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
+
def check_inputs(
self,
prompt,
@@ -305,41 +330,8 @@ def encode_prompt(
return prompt_embeds, prompt_attention_mask, negative_prompt_embeds, negative_prompt_attention_mask
- # Copied from diffusers.pipelines.stable_diffusion_3.pipeline_stable_diffusion_3.StableDiffusion3Pipeline.prepare_latents
- def prepare_latents(
- self,
- batch_size,
- num_channels_latents,
- height,
- width,
- dtype,
- device,
- generator,
- latents=None,
- ):
- if latents is not None:
- return latents.to(device=device, dtype=dtype)
-
- shape = (
- batch_size,
- num_channels_latents,
- int(height) // self.vae_scale_factor,
- int(width) // self.vae_scale_factor,
- )
-
- 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)
-
- return latents
-
def get_timesteps(self, num_inference_steps, strength, device):
# Set timesteps using the full range initially
- self.scheduler.set_timesteps(num_inference_steps, device=device)
timesteps = self.scheduler.timesteps.to(device=device)
if len(timesteps) != num_inference_steps:
@@ -349,11 +341,15 @@ def get_timesteps(self, num_inference_steps, strength, device):
init_timestep = min(num_inference_steps * strength, num_inference_steps)
t_start = int(max(num_inference_steps - init_timestep, 0))
- timesteps = self.scheduler.timesteps[t_start:]
+ timesteps = self.scheduler.timesteps[t_start * self.scheduler.order:]
+
+ # Set begin index if scheduler supports it
+ if hasattr(self.scheduler, "set_begin_index"):
+ self.scheduler.set_begin_index(t_start * self.scheduler.order)
return timesteps, num_inference_steps - t_start
- def prepare_img2img_latents(
+ def prepare_latents(
self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None
):
if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
@@ -361,6 +357,13 @@ def prepare_img2img_latents(
f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
)
+ # Check for latents_mean and latents_std in the VAE config
+ latents_mean = latents_std = None
+ if hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None:
+ latents_mean = torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1)
+ if hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None:
+ latents_std = torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1)
+
image = image.to(device=device, dtype=dtype)
batch_size = batch_size * num_images_per_prompt
@@ -404,26 +407,30 @@ def prepare_img2img_latents(
if isinstance(generator, list):
sample = torch.cat(
[
- torch.randn(
+ randn_tensor(
(1, *mean.shape[1:]),
generator=generator[i],
- device=generator[i].device if hasattr(generator[i], "device") else "cpu",
+ device=mean.device,
dtype=mean.dtype,
- ).to(mean.device)
+ )
for i in range(batch_size)
]
)
else:
# Single generator - use its device if it has one
- generator_device = getattr(generator, "device", "cpu") if generator is not None else "cpu"
- noise = torch.randn(mean.shape, generator=generator, device=generator_device, dtype=mean.dtype)
- sample = noise.to(mean.device)
+ sample = randn_tensor(mean.shape, generator=generator, device=mean.device, dtype=mean.dtype)
# Compute latents
latents = mean + std * sample
- # Scale latents
- latents = latents * self.vae.config.scaling_factor
+ # Apply standardization if VAE has mean and std defined in config
+ if latents_mean is not None and latents_std is not None:
+ latents_mean = latents_mean.to(device=device, dtype=dtype)
+ latents_std = latents_std.to(device=device, dtype=dtype)
+ latents = (latents - latents_mean) * self.vae.config.scaling_factor / latents_std
+ else:
+ # Scale latents
+ latents = latents * self.vae.config.scaling_factor
# get the original timestep using init_timestep
init_timestep = timestep # Use the passed timestep directly
@@ -433,21 +440,20 @@ def prepare_img2img_latents(
if isinstance(generator, list):
noise = torch.cat(
[
- torch.randn(
+ randn_tensor(
(1, *latents.shape[1:]),
generator=generator[i],
- device=generator[i].device if hasattr(generator[i], "device") else "cpu",
+ device=latents.device,
dtype=latents.dtype,
- ).to(latents.device)
+ )
for i in range(batch_size)
]
)
else:
# Single generator - use its device if it has one
- generator_device = getattr(generator, "device", "cpu") if generator is not None else "cpu"
- noise = torch.randn(
- latents.shape, generator=generator, device=generator_device, dtype=latents.dtype
- ).to(latents.device)
+ noise = randn_tensor(
+ latents.shape, generator=generator, device=latents.device, dtype=latents.dtype
+ )
latents = self.scheduler.scale_noise(latents, init_timestep, noise)
@@ -654,13 +660,29 @@ def __call__(
prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
# 5. Prepare timesteps
+ # Calculate shift parameter based on image dimensions
+ image_seq_len = (int(height) // self.vae_scale_factor // 2) * (int(width) // self.vae_scale_factor // 2)
+
+ # Calculate mu (shift parameter) based on image dimensions
+ mu = self.calculate_shift(
+ image_seq_len,
+ self.scheduler.config.get("base_image_seq_len", 256),
+ self.scheduler.config.get("max_image_seq_len", 4096),
+ self.scheduler.config.get("base_shift", 0.5),
+ self.scheduler.config.get("max_shift", 1.15),
+ )
+
+ # Set timesteps with shift parameter
+ self.scheduler.set_timesteps(num_inference_steps, device=device, mu=mu)
+
+ # Now adjust for strength
timesteps, num_inference_steps = self.get_timesteps(
num_inference_steps, strength, device
)
latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt) # Get the first timestep(s) for initial noise
# 6. Prepare latent variables
- latents = self.prepare_img2img_latents(
+ latents = self.prepare_latents(
image,
latent_timestep,
batch_size,
From 937502046951d656dd736f62fcdc13d1108f601b Mon Sep 17 00:00:00 2001
From: AstraliteHeart <81396681+AstraliteHeart@users.noreply.github.com>
Date: Thu, 1 May 2025 20:56:26 -0700
Subject: [PATCH 5/5] Update
src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
Co-authored-by: Bagheera <59658056+bghira@users.noreply.github.com>
---
.../aura_flow/pipeline_aura_flow_img2img.py | 127 ++++++------------
1 file changed, 39 insertions(+), 88 deletions(-)
diff --git a/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
index dae182c921f3..5cf9a820681d 100644
--- a/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
+++ b/src/diffusers/pipelines/aura_flow/pipeline_aura_flow_img2img.py
@@ -350,115 +350,66 @@ def get_timesteps(self, num_inference_steps, strength, device):
return timesteps, num_inference_steps - t_start
def prepare_latents(
- self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None
+ self,
+ image,
+ timestep,
+ batch_size,
+ num_images_per_prompt,
+ dtype,
+ device,
+ generator=None,
):
if not isinstance(image, (torch.Tensor, PIL.Image.Image, list)):
raise ValueError(
- f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
+ f"`image` must be `torch.Tensor`, `PIL.Image.Image` or list, got {type(image)}"
)
- # Check for latents_mean and latents_std in the VAE config
- latents_mean = latents_std = None
- if hasattr(self.vae.config, "latents_mean") and self.vae.config.latents_mean is not None:
- latents_mean = torch.tensor(self.vae.config.latents_mean).view(1, 4, 1, 1)
- if hasattr(self.vae.config, "latents_std") and self.vae.config.latents_std is not None:
- latents_std = torch.tensor(self.vae.config.latents_std).view(1, 4, 1, 1)
-
image = image.to(device=device, dtype=dtype)
-
batch_size = batch_size * num_images_per_prompt
if image.shape[1] == 4:
- latents = image
+ latents_0 = image
else:
- 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."
- )
-
- # Handle different batch size scenarios
- if image.shape[0] < batch_size:
- if batch_size % image.shape[0] == 0:
- # Duplicate the image to match the batch size
- additional_image_per_prompt = batch_size // image.shape[0]
- image = torch.cat([image] * additional_image_per_prompt, dim=0)
- else:
- raise ValueError(
- f"Cannot duplicate `image` of batch size {image.shape[0]} to {batch_size} text prompts."
- f" Batch size must be divisible by the image batch size."
- )
-
- # Temporarily move VAE to float32 for encoding
- vae_dtype = self.vae.dtype
- if vae_dtype != torch.float32:
+ # VAE ⇢ latents (ALWAYS on fp32 for numerical stability)
+ orig_dtype = self.vae.dtype
+ if orig_dtype != torch.float32:
self.vae.to(dtype=torch.float32)
- # encode the init image into latents and scale the latents
- # 1. Get VAE distribution parameters (on device)
latent_dist = self.vae.encode(image.to(dtype=torch.float32)).latent_dist
- mean, std = latent_dist.mean, latent_dist.std # Already on device
+ latents_0 = latent_dist.mean # ❶ deterministic!
- # Restore VAE dtype
- if vae_dtype != torch.float32:
- self.vae.to(dtype=vae_dtype)
+ if orig_dtype != torch.float32:
+ self.vae.to(dtype=orig_dtype)
- # 2. Sample noise for each batch element individually if using multiple generators
- if isinstance(generator, list):
- sample = torch.cat(
- [
- randn_tensor(
- (1, *mean.shape[1:]),
- generator=generator[i],
- device=mean.device,
- dtype=mean.dtype,
- )
- for i in range(batch_size)
- ]
- )
- else:
- # Single generator - use its device if it has one
- sample = randn_tensor(mean.shape, generator=generator, device=mean.device, dtype=mean.dtype)
+ # scale
+ latents_0 = latents_0 * self.vae.config.scaling_factor
- # Compute latents
- latents = mean + std * sample
-
- # Apply standardization if VAE has mean and std defined in config
- if latents_mean is not None and latents_std is not None:
- latents_mean = latents_mean.to(device=device, dtype=dtype)
- latents_std = latents_std.to(device=device, dtype=dtype)
- latents = (latents - latents_mean) * self.vae.config.scaling_factor / latents_std
- else:
- # Scale latents
- latents = latents * self.vae.config.scaling_factor
-
- # get the original timestep using init_timestep
- init_timestep = timestep # Use the passed timestep directly
-
- # add noise to latents using the timesteps
- # Handle noise generation with multiple generators if provided
- if isinstance(generator, list):
- noise = torch.cat(
- [
- randn_tensor(
- (1, *latents.shape[1:]),
- generator=generator[i],
- device=latents.device,
- dtype=latents.dtype,
- )
- for i in range(batch_size)
- ]
- )
- else:
- # Single generator - use its device if it has one
- noise = randn_tensor(
- latents.shape, generator=generator, device=latents.device, dtype=latents.dtype
+ # replicate to match `batch_size`
+ if latents_0.shape[0] != batch_size:
+ if batch_size % latents_0.shape[0] != 0:
+ raise ValueError(
+ f"Cannot duplicate image batch of size {latents_0.shape[0]} "
+ f"to effective batch size {batch_size}."
)
+ repeats = batch_size // latents_0.shape[0]
+ latents_0 = latents_0.repeat(repeats, 1, 1, 1)
+
+ noise = randn_tensor(
+ latents_0.shape,
+ generator=generator,
+ device=latents_0.device,
+ dtype=latents_0.dtype,
+ )
- latents = self.scheduler.scale_noise(latents, init_timestep, noise)
+ # make sure `timestep` is 1-D and matches batch
+ if isinstance(timestep, (int, float)):
+ timestep = torch.tensor([timestep], device=latents_0.device, dtype=latents_0.dtype)
+ timestep = timestep.expand(latents_0.shape[0])
+ latents = self.scheduler.scale_noise(latents_0, timestep, noise)
return latents
+
# Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.upcast_vae
def upcast_vae(self):
dtype = self.vae.dtype