Skip to content

GitLab

"docs/source/en/api/pipelines/latent_diffusion_uncond.mdx" did not exist on "f6fb3282b18f44f14bcb95a34a16203906df992a"
Unverified Commit 67b3d326 authored by Aryan V S's avatar Aryan V S Committed by GitHub
Browse files

Support img2img and inpaint in lpw-xl (#6114) 



* add img2img and inpaint support to lpw-xl

* update community README

---------
Co-authored-by: default avatarSayak Paul <spsayakpaul@gmail.com>
parent 4e770568
Show whitespace changes
Inline Side-by-side
Showing with 785 additions and 43 deletions
examples/community/README.md
View file @ 67b3d326
......@@ -41,7 +41,7 @@ If a community doesn't work as expected, please open an issue and ping the autho
| TensorRT Stable Diffusion Inpainting Pipeline | Accelerates the Stable Diffusion Inpainting Pipeline using TensorRT | [TensorRT Stable Diffusion Inpainting Pipeline](#tensorrt-inpainting-stable-diffusion-pipeline) | - | [Asfiya Baig](https://github.com/asfiyab-nvidia) |
| IADB Pipeline | Implementation of [Iterative α-(de)Blending: a Minimalist Deterministic Diffusion Model](https://arxiv.org/abs/2305.03486) | [IADB Pipeline](#iadb-pipeline) | - | [Thomas Chambon](https://github.com/tchambon)
| Zero1to3 Pipeline | Implementation of [Zero-1-to-3: Zero-shot One Image to 3D Object](https://arxiv.org/abs/2303.11328) | [Zero1to3 Pipeline](#Zero1to3-pipeline) | - | [Xin Kong](https://github.com/kxhit) |
Stable Diffusion XL Long Weighted Prompt Pipeline | A pipeline support unlimited length of prompt and negative prompt, use A1111 style of prompt weighting | [Stable Diffusion XL Long Weighted Prompt Pipeline](#stable-diffusion-xl-long-weighted-prompt-pipeline) | - | [Andrew Zhu](https://xhinker.medium.com/) |
| Stable Diffusion XL Long Weighted Prompt Pipeline | A pipeline support unlimited length of prompt and negative prompt, use A1111 style of prompt weighting | [Stable Diffusion XL Long Weighted Prompt Pipeline](#stable-diffusion-xl-long-weighted-prompt-pipeline) | [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1LsqilswLR40XLLcp6XFOl5nKb_wOe26W?usp=sharing) | [Andrew Zhu](https://xhinker.medium.com/) |
FABRIC - Stable Diffusion with feedback Pipeline | pipeline supports feedback from liked and disliked images | [Stable Diffusion Fabric Pipeline](#stable-diffusion-fabric-pipeline) | - | [Shauray Singh](https://shauray8.github.io/about_shauray/) |
sketch inpaint - Inpainting with non-inpaint Stable Diffusion | sketch inpaint much like in automatic1111 | [Masked Im2Im Stable Diffusion Pipeline](#stable-diffusion-masked-im2im) | - | [Anatoly Belikov](https://github.com/noskill) |
prompt-to-prompt | change parts of a prompt and retain image structure (see [paper page](https://prompt-to-prompt.github.io/)) | [Prompt2Prompt Pipeline](#prompt2prompt-pipeline) | - | [Umer H. Adil](https://twitter.com/UmerHAdil) |
......@@ -1619,10 +1619,11 @@ This approach is using (optional) CoCa model to avoid writing image description.
This SDXL pipeline support unlimited length prompt and negative prompt, compatible with A1111 prompt weighted style.
You can provide both `prompt` and `prompt_2`. if only one prompt is provided, `prompt_2` will be a copy of the provided `prompt`. Here is a sample code to use this pipeline.
You can provide both `prompt` and `prompt_2`. If only one prompt is provided, `prompt_2` will be a copy of the provided `prompt`. Here is a sample code to use this pipeline.
```python
from diffusers import DiffusionPipeline
from diffusers.utils import load_image
import torch
pipe = DiffusionPipeline.from_pretrained(
......@@ -1633,25 +1634,52 @@ pipe = DiffusionPipeline.from_pretrained(
, custom_pipeline = "lpw_stable_diffusion_xl",
)
prompt = "photo of a cute (white) cat running on the grass"*20
prompt2 = "chasing (birds:1.5)"*20
prompt = "photo of a cute (white) cat running on the grass" * 20
prompt2 = "chasing (birds:1.5)" * 20
prompt = f"{prompt},{prompt2}"
neg_prompt = "blur, low quality, carton, animate"
pipe.to("cuda")
images = pipe(
prompt = prompt
, negative_prompt = neg_prompt
).images[0]
# text2img
t2i_images = pipe(
prompt=prompt,
negative_prompt=neg_prompt,
).images # alternatively, you can call the .text2img() function
# img2img
input_image = load_image("/path/to/local/image.png") # or URL to your input image
i2i_images = pipe.img2img(
prompt=prompt,
negative_prompt=neg_prompt,
image=input_image,
strength=0.8, # higher strength will result in more variation compared to original image
).images
# inpaint
input_mask = load_image("/path/to/local/mask.png") # or URL to your input inpainting mask
inpaint_images = pipe.inpaint(
prompt="photo of a cute (black) cat running on the grass" * 20,
negative_prompt=neg_prompt,
image=input_image,
mask=input_mask,
strength=0.6, # higher strength will result in more variation compared to original image
).images
pipe.to("cpu")
torch.cuda.empty_cache()
images
from IPython.display import display # assuming you are using this code in a notebook
display(t2i_images[0])
display(i2i_images[0])
display(inpaint_images[0])
```
In the above code, the `prompt2` is appended to the `prompt`, which is more than 77 tokens. "birds" are showing up in the result.
![Stable Diffusion XL Long Weighted Prompt Pipeline sample](https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/sdxl_long_weighted_prompt.png)
For more results, checkout [PR #6114](https://github.com/huggingface/diffusers/pull/6114).
## Example Images Mixing (with CoCa)
```python
import requests
......
examples/community/lpw_stable_diffusion_xl.py
View file @ 67b3d326
......@@ -11,10 +11,11 @@ import os
from typing import Any, Callable, Dict, List, Optional, Tuple, Union
import torch
from PIL import Image
from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokenizer
from diffusers import DiffusionPipeline, StableDiffusionXLPipeline
from diffusers.image_processor import VaeImageProcessor
from diffusers.image_processor import PipelineImageInput, VaeImageProcessor
from diffusers.loaders import FromSingleFileMixin, LoraLoaderMixin, TextualInversionLoaderMixin
from diffusers.models import AutoencoderKL, UNet2DConditionModel
from diffusers.models.attention_processor import (
......@@ -23,7 +24,7 @@ from diffusers.models.attention_processor import (
LoRAXFormersAttnProcessor,
XFormersAttnProcessor,
)
from diffusers.pipelines.stable_diffusion_xl import StableDiffusionXLPipelineOutput
from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
from diffusers.schedulers import KarrasDiffusionSchedulers
from diffusers.utils import (
is_accelerate_available,
......@@ -461,6 +462,65 @@ def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
return noise_cfg
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_img2img.retrieve_latents
def retrieve_latents(
encoder_output: torch.Tensor, generator: Optional[torch.Generator] = None, sample_mode: str = "sample"
):
if hasattr(encoder_output, "latent_dist") and sample_mode == "sample":
return encoder_output.latent_dist.sample(generator)
elif hasattr(encoder_output, "latent_dist") and sample_mode == "argmax":
return encoder_output.latent_dist.mode()
elif hasattr(encoder_output, "latents"):
return encoder_output.latents
else:
raise AttributeError("Could not access latents of provided encoder_output")
# 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,
**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 support arbitrary spacing between timesteps. If `None`, then the default
timestep spacing strategy of the scheduler is used. If `timesteps` is passed, `num_inference_steps`
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:
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)
else:
scheduler.set_timesteps(num_inference_steps, device=device, **kwargs)
timesteps = scheduler.timesteps
return timesteps, num_inference_steps
class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, LoraLoaderMixin):
r"""
Pipeline for text-to-image generation using Stable Diffusion XL.
......@@ -526,6 +586,9 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
self.vae_scale_factor = 2 ** (len(self.vae.config.block_out_channels) - 1)
self.image_processor = VaeImageProcessor(vae_scale_factor=self.vae_scale_factor)
self.mask_processor = VaeImageProcessor(
vae_scale_factor=self.vae_scale_factor, do_normalize=False, do_binarize=True, do_convert_grayscale=True
)
self.default_sample_size = self.unet.config.sample_size
add_watermarker = add_watermarker if add_watermarker is not None else is_invisible_watermark_available()
......@@ -813,6 +876,7 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
prompt_2,
height,
width,
strength,
callback_steps,
negative_prompt=None,
negative_prompt_2=None,
......@@ -824,6 +888,9 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
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 strength < 0 or strength > 1:
raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
if (callback_steps is None) or (
callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
):
......@@ -880,8 +947,64 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
"If `negative_prompt_embeds` are provided, `negative_pooled_prompt_embeds` also have to be passed. Make sure to generate `negative_pooled_prompt_embeds` from the same text encoder that was used to generate `negative_prompt_embeds`."
)
# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_latents
def prepare_latents(self, batch_size, num_channels_latents, height, width, dtype, device, generator, latents=None):
def get_timesteps(self, num_inference_steps, strength, device, denoising_start=None):
# get the original timestep using init_timestep
if denoising_start is None:
init_timestep = min(int(num_inference_steps * strength), num_inference_steps)
t_start = max(num_inference_steps - init_timestep, 0)
else:
t_start = 0
timesteps = self.scheduler.timesteps[t_start * self.scheduler.order :]
# Strength is irrelevant if we directly request a timestep to start at;
# that is, strength is determined by the denoising_start instead.
if denoising_start is not None:
discrete_timestep_cutoff = int(
round(
self.scheduler.config.num_train_timesteps
- (denoising_start * self.scheduler.config.num_train_timesteps)
)
)
num_inference_steps = (timesteps < discrete_timestep_cutoff).sum().item()
if self.scheduler.order == 2 and num_inference_steps % 2 == 0:
# if the scheduler is a 2nd order scheduler we might have to do +1
# because `num_inference_steps` might be even given that every timestep
# (except the highest one) is duplicated. If `num_inference_steps` is even it would
# mean that we cut the timesteps in the middle of the denoising step
# (between 1st and 2nd devirative) which leads to incorrect results. By adding 1
# we ensure that the denoising process always ends after the 2nd derivate step of the scheduler
num_inference_steps = num_inference_steps + 1
# because t_n+1 >= t_n, we slice the timesteps starting from the end
timesteps = timesteps[-num_inference_steps:]
return timesteps, num_inference_steps
return timesteps, num_inference_steps - t_start
def prepare_latents(
self,
image,
mask,
width,
height,
num_channels_latents,
timestep,
batch_size,
num_images_per_prompt,
dtype,
device,
generator=None,
add_noise=True,
latents=None,
is_strength_max=True,
return_noise=False,
return_image_latents=False,
):
batch_size *= num_images_per_prompt
if image is None:
shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, width // self.vae_scale_factor)
if isinstance(generator, list) and len(generator) != batch_size:
raise ValueError(
......@@ -898,6 +1021,190 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
latents = latents * self.scheduler.init_noise_sigma
return latents
elif mask is None:
if not isinstance(image, (torch.Tensor, Image.Image, list)):
raise ValueError(
f"`image` has to be of type `torch.Tensor`, `PIL.Image.Image` or list but is {type(image)}"
)
# Offload text encoder if `enable_model_cpu_offload` was enabled
if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
self.text_encoder_2.to("cpu")
torch.cuda.empty_cache()
image = image.to(device=device, dtype=dtype)
if image.shape[1] == 4:
init_latents = image
else:
# make sure the VAE is in float32 mode, as it overflows in float16
if self.vae.config.force_upcast:
image = image.float()
self.vae.to(dtype=torch.float32)
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."
)
elif isinstance(generator, list):
init_latents = [
retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
for i in range(batch_size)
]
init_latents = torch.cat(init_latents, dim=0)
else:
init_latents = retrieve_latents(self.vae.encode(image), generator=generator)
if self.vae.config.force_upcast:
self.vae.to(dtype)
init_latents = init_latents.to(dtype)
init_latents = self.vae.config.scaling_factor * init_latents
if batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] == 0:
# expand init_latents for batch_size
additional_image_per_prompt = batch_size // init_latents.shape[0]
init_latents = torch.cat([init_latents] * additional_image_per_prompt, dim=0)
elif batch_size > init_latents.shape[0] and batch_size % init_latents.shape[0] != 0:
raise ValueError(
f"Cannot duplicate `image` of batch size {init_latents.shape[0]} to {batch_size} text prompts."
)
else:
init_latents = torch.cat([init_latents], dim=0)
if add_noise:
shape = init_latents.shape
noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
# get latents
init_latents = self.scheduler.add_noise(init_latents, noise, timestep)
latents = init_latents
return latents
else:
shape = (batch_size, num_channels_latents, height // self.vae_scale_factor, 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."
)
if (image is None or timestep is None) and not is_strength_max:
raise ValueError(
"Since strength < 1. initial latents are to be initialised as a combination of Image + Noise."
"However, either the image or the noise timestep has not been provided."
)
if image.shape[1] == 4:
image_latents = image.to(device=device, dtype=dtype)
image_latents = image_latents.repeat(batch_size // image_latents.shape[0], 1, 1, 1)
elif return_image_latents or (latents is None and not is_strength_max):
image = image.to(device=device, dtype=dtype)
image_latents = self._encode_vae_image(image=image, generator=generator)
image_latents = image_latents.repeat(batch_size // image_latents.shape[0], 1, 1, 1)
if latents is None and add_noise:
noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
# if strength is 1. then initialise the latents to noise, else initial to image + noise
latents = noise if is_strength_max else self.scheduler.add_noise(image_latents, noise, timestep)
# if pure noise then scale the initial latents by the Scheduler's init sigma
latents = latents * self.scheduler.init_noise_sigma if is_strength_max else latents
elif add_noise:
noise = latents.to(device)
latents = noise * self.scheduler.init_noise_sigma
else:
noise = randn_tensor(shape, generator=generator, device=device, dtype=dtype)
latents = image_latents.to(device)
outputs = (latents,)
if return_noise:
outputs += (noise,)
if return_image_latents:
outputs += (image_latents,)
return outputs
def _encode_vae_image(self, image: torch.Tensor, generator: torch.Generator):
dtype = image.dtype
if self.vae.config.force_upcast:
image = image.float()
self.vae.to(dtype=torch.float32)
if isinstance(generator, list):
image_latents = [
retrieve_latents(self.vae.encode(image[i : i + 1]), generator=generator[i])
for i in range(image.shape[0])
]
image_latents = torch.cat(image_latents, dim=0)
else:
image_latents = retrieve_latents(self.vae.encode(image), generator=generator)
if self.vae.config.force_upcast:
self.vae.to(dtype)
image_latents = image_latents.to(dtype)
image_latents = self.vae.config.scaling_factor * image_latents
return image_latents
def prepare_mask_latents(
self, mask, masked_image, batch_size, height, width, dtype, device, generator, do_classifier_free_guidance
):
# resize the mask to latents shape as we concatenate the mask to the latents
# we do that before converting to dtype to avoid breaking in case we're using cpu_offload
# and half precision
mask = torch.nn.functional.interpolate(
mask, size=(height // self.vae_scale_factor, width // self.vae_scale_factor)
)
mask = mask.to(device=device, dtype=dtype)
# duplicate mask and masked_image_latents for each generation per prompt, using mps friendly method
if mask.shape[0] < batch_size:
if not batch_size % mask.shape[0] == 0:
raise ValueError(
"The passed mask and the required batch size don't match. Masks are supposed to be duplicated to"
f" a total batch size of {batch_size}, but {mask.shape[0]} masks were passed. Make sure the number"
" of masks that you pass is divisible by the total requested batch size."
)
mask = mask.repeat(batch_size // mask.shape[0], 1, 1, 1)
mask = torch.cat([mask] * 2) if do_classifier_free_guidance else mask
if masked_image is not None and masked_image.shape[1] == 4:
masked_image_latents = masked_image
else:
masked_image_latents = None
if masked_image is not None:
if masked_image_latents is None:
masked_image = masked_image.to(device=device, dtype=dtype)
masked_image_latents = self._encode_vae_image(masked_image, generator=generator)
if masked_image_latents.shape[0] < batch_size:
if not batch_size % masked_image_latents.shape[0] == 0:
raise ValueError(
"The passed images and the required batch size don't match. Images are supposed to be duplicated"
f" to a total batch size of {batch_size}, but {masked_image_latents.shape[0]} images were passed."
" Make sure the number of images that you pass is divisible by the total requested batch size."
)
masked_image_latents = masked_image_latents.repeat(
batch_size // masked_image_latents.shape[0], 1, 1, 1
)
masked_image_latents = (
torch.cat([masked_image_latents] * 2) if do_classifier_free_guidance else masked_image_latents
)
# aligning device to prevent device errors when concating it with the latent model input
masked_image_latents = masked_image_latents.to(device=device, dtype=dtype)
return mask, masked_image_latents
def _get_add_time_ids(self, original_size, crops_coords_top_left, target_size, dtype):
add_time_ids = list(original_size + crops_coords_top_left + target_size)
......@@ -934,15 +1241,52 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
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 guidance_rescale(self):
return self._guidance_rescale
# 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.
@property
def do_classifier_free_guidance(self):
return self._guidance_scale > 1 and self.unet.config.time_cond_proj_dim is None
@property
def cross_attention_kwargs(self):
return self._cross_attention_kwargs
@property
def denoising_end(self):
return self._denoising_end
@property
def denoising_start(self):
return self._denoising_start
@property
def num_timesteps(self):
return self._num_timesteps
@torch.no_grad()
@replace_example_docstring(EXAMPLE_DOC_STRING)
def __call__(
self,
prompt: str = None,
prompt_2: Optional[str] = None,
image: Optional[PipelineImageInput] = None,
mask_image: Optional[PipelineImageInput] = None,
masked_image_latents: Optional[torch.FloatTensor] = None,
height: Optional[int] = None,
width: Optional[int] = None,
strength: float = 0.8,
num_inference_steps: int = 50,
timesteps: List[int] = None,
denoising_start: Optional[float] = None,
denoising_end: Optional[float] = None,
guidance_scale: float = 5.0,
negative_prompt: Optional[str] = None,
......@@ -975,20 +1319,46 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
prompt_2 (`str`):
The prompt to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
used in both text-encoders
image (`PipelineImageInput`, *optional*):
`Image`, or tensor representing an image batch, that will be used as the starting point for the
process.
mask_image (`PipelineImageInput`, *optional*):
`Image`, or tensor representing an image batch, to mask `image`. White pixels in the mask will be
replaced by noise and therefore repainted, while black pixels will be preserved. If `mask_image` is a
PIL image, it will be converted to a single channel (luminance) before use. If it's a tensor, it should
contain one color channel (L) instead of 3, so the expected shape would be `(B, H, W, 1)`.
height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
The height in pixels of the generated image.
width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
The width in pixels of the generated image.
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`.
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.
denoising_start (`float`, *optional*):
When specified, indicates the fraction (between 0.0 and 1.0) of the total denoising process to be
bypassed before it is initiated. Consequently, the initial part of the denoising process is skipped and
it is assumed that the passed `image` is a partly denoised image. Note that when this is specified,
strength will be ignored. The `denoising_start` parameter is particularly beneficial when this pipeline
is integrated into a "Mixture of Denoisers" multi-pipeline setup, as detailed in [**Refine Image
Quality**](https://huggingface.co/docs/diffusers/using-diffusers/sdxl#refine-image-quality).
denoising_end (`float`, *optional*):
When specified, determines the fraction (between 0.0 and 1.0) of the total denoising process to be
completed before it is intentionally prematurely terminated. As a result, the returned sample will
still retain a substantial amount of noise as determined by the discrete timesteps selected by the
scheduler. The denoising_end parameter should ideally be utilized when this pipeline forms a part of a
"Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refining the Image
Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output)
still retain a substantial amount of noise (ca. final 20% of timesteps still needed) and should be
denoised by a successor pipeline that has `denoising_start` set to 0.8 so that it only denoises the
final 20% of the scheduler. The denoising_end parameter should ideally be utilized when this pipeline
forms a part of a "Mixture of Denoisers" multi-pipeline setup, as elaborated in [**Refine Image
Quality**](https://huggingface.co/docs/diffusers/using-diffusers/sdxl#refine-image-quality).
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
......@@ -1084,6 +1454,7 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
prompt_2,
height,
width,
strength,
callback_steps,
negative_prompt,
negative_prompt_2,
......@@ -1093,6 +1464,12 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
negative_pooled_prompt_embeds,
)
self._guidance_scale = guidance_scale
self._guidance_rescale = guidance_rescale
self._cross_attention_kwargs = cross_attention_kwargs
self._denoising_end = denoising_end
self._denoising_start = denoising_start
# 2. Define call parameters
if prompt is not None and isinstance(prompt, str):
batch_size = 1
......@@ -1121,28 +1498,126 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
) = get_weighted_text_embeddings_sdxl(
pipe=self, prompt=prompt, neg_prompt=negative_prompt, num_images_per_prompt=num_images_per_prompt
)
dtype = prompt_embeds.dtype
if isinstance(image, Image.Image):
image = self.image_processor.preprocess(image, height=height, width=width)
if image is not None:
image = image.to(device=self.device, dtype=dtype)
if isinstance(mask_image, Image.Image):
mask = self.mask_processor.preprocess(mask_image, height=height, width=width)
else:
mask = mask_image
if mask_image is not None:
mask = mask.to(device=self.device, dtype=dtype)
if masked_image_latents is not None:
masked_image = masked_image_latents
elif image.shape[1] == 4:
# if image is in latent space, we can't mask it
masked_image = None
else:
masked_image = image * (mask < 0.5)
else:
mask = None
# 4. Prepare timesteps
self.scheduler.set_timesteps(num_inference_steps, device=device)
def denoising_value_valid(dnv):
return isinstance(self.denoising_end, float) and 0 < dnv < 1
timesteps, num_inference_steps = retrieve_timesteps(self.scheduler, num_inference_steps, device, timesteps)
if image is not None:
timesteps, num_inference_steps = self.get_timesteps(
num_inference_steps,
strength,
device,
denoising_start=self.denoising_start if denoising_value_valid else None,
)
timesteps = self.scheduler.timesteps
# check that number of inference steps is not < 1 - as this doesn't make sense
if num_inference_steps < 1:
raise ValueError(
f"After adjusting the num_inference_steps by strength parameter: {strength}, the number of pipeline"
f"steps is {num_inference_steps} which is < 1 and not appropriate for this pipeline."
)
latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
is_strength_max = strength == 1.0
add_noise = True if self.denoising_start is None else False
# 5. Prepare latent variables
num_channels_latents = self.unet.config.in_channels
num_channels_latents = self.vae.config.latent_channels
num_channels_unet = self.unet.config.in_channels
return_image_latents = num_channels_unet == 4
latents = self.prepare_latents(
batch_size * num_images_per_prompt,
num_channels_latents,
height,
width,
prompt_embeds.dtype,
device,
generator,
latents,
image=image,
mask=mask,
width=width,
height=height,
num_channels_latents=num_channels_unet,
timestep=latent_timestep,
batch_size=batch_size,
num_images_per_prompt=num_images_per_prompt,
dtype=prompt_embeds.dtype,
device=device,
generator=generator,
add_noise=add_noise,
latents=latents,
is_strength_max=is_strength_max,
return_noise=True,
return_image_latents=return_image_latents,
)
if mask is not None:
if return_image_latents:
latents, noise, image_latents = latents
else:
latents, noise = latents
# 5.1. Prepare mask latent variables
if mask is not None:
mask, masked_image_latents = self.prepare_mask_latents(
mask=mask,
masked_image=masked_image,
batch_size=batch_size * num_images_per_prompt,
height=height,
width=width,
dtype=prompt_embeds.dtype,
device=device,
generator=generator,
do_classifier_free_guidance=self.do_classifier_free_guidance,
)
# 8. Check that sizes of mask, masked image and latents match
if num_channels_unet == 9:
# default case for runwayml/stable-diffusion-inpainting
num_channels_mask = mask.shape[1]
num_channels_masked_image = masked_image_latents.shape[1]
if num_channels_latents + num_channels_mask + num_channels_masked_image != num_channels_unet:
raise ValueError(
f"Incorrect configuration settings! The config of `pipeline.unet`: {self.unet.config} expects"
f" {self.unet.config.in_channels} but received `num_channels_latents`: {num_channels_latents} +"
f" `num_channels_mask`: {num_channels_mask} + `num_channels_masked_image`: {num_channels_masked_image}"
f" = {num_channels_latents+num_channels_masked_image+num_channels_mask}. Please verify the config of"
" `pipeline.unet` or your `mask_image` or `image` input."
)
elif num_channels_unet != 4:
raise ValueError(
f"The unet {self.unet.__class__} should have either 4 or 9 input channels, not {self.unet.config.in_channels}."
)
# 6. Prepare extra step kwargs. TODO: Logic should ideally just be moved out of the pipeline
extra_step_kwargs = self.prepare_extra_step_kwargs(generator, eta)
height, width = latents.shape[-2:]
height = height * self.vae_scale_factor
width = width * self.vae_scale_factor
original_size = original_size or (height, width)
target_size = target_size or (height, width)
# 7. Prepare added time ids & embeddings
add_text_embeds = pooled_prompt_embeds
add_time_ids = self._get_add_time_ids(
......@@ -1158,20 +1633,41 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
add_text_embeds = add_text_embeds.to(device)
add_time_ids = add_time_ids.to(device).repeat(batch_size * num_images_per_prompt, 1)
# 8. Denoising loop
num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
# 7.1 Apply denoising_end
if denoising_end is not None and isinstance(denoising_end, float) and denoising_end > 0 and denoising_end < 1:
if (
self.denoising_end is not None
and self.denoising_start is not None
and denoising_value_valid(self.denoising_end)
and denoising_value_valid(self.denoising_start)
and self.denoising_start >= self.denoising_end
):
raise ValueError(
f"`denoising_start`: {self.denoising_start} cannot be larger than or equal to `denoising_end`: "
+ f" {self.denoising_end} when using type float."
)
elif self.denoising_end is not None and denoising_value_valid(self.denoising_end):
discrete_timestep_cutoff = int(
round(
self.scheduler.config.num_train_timesteps
- (denoising_end * self.scheduler.config.num_train_timesteps)
- (self.denoising_end * self.scheduler.config.num_train_timesteps)
)
)
num_inference_steps = len(list(filter(lambda ts: ts >= discrete_timestep_cutoff, timesteps)))
timesteps = timesteps[:num_inference_steps]
# 8. Optionally get Guidance Scale Embedding
timestep_cond = None
if self.unet.config.time_cond_proj_dim is not None:
guidance_scale_tensor = torch.tensor(self.guidance_scale - 1).repeat(batch_size * num_images_per_prompt)
timestep_cond = self.get_guidance_scale_embedding(
guidance_scale_tensor, embedding_dim=self.unet.config.time_cond_proj_dim
).to(device=device, dtype=latents.dtype)
self._num_timesteps = len(timesteps)
# 9. Denoising loop
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
......@@ -1179,13 +1675,17 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
if mask is not None and num_channels_unet == 9:
latent_model_input = torch.cat([latent_model_input, mask, masked_image_latents], dim=1)
# predict the noise residual
added_cond_kwargs = {"text_embeds": add_text_embeds, "time_ids": add_time_ids}
noise_pred = self.unet(
latent_model_input,
t,
encoder_hidden_states=prompt_embeds,
cross_attention_kwargs=cross_attention_kwargs,
timestep_cond=timestep_cond,
cross_attention_kwargs=self.cross_attention_kwargs,
added_cond_kwargs=added_cond_kwargs,
return_dict=False,
)[0]
......@@ -1202,6 +1702,22 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
# compute the previous noisy sample x_t -> x_t-1
latents = self.scheduler.step(noise_pred, t, latents, **extra_step_kwargs, return_dict=False)[0]
if mask is not None and num_channels_unet == 4:
init_latents_proper = image_latents
if self.do_classifier_free_guidance:
init_mask, _ = mask.chunk(2)
else:
init_mask = mask
if i < len(timesteps) - 1:
noise_timestep = timesteps[i + 1]
init_latents_proper = self.scheduler.add_noise(
init_latents_proper, noise, torch.tensor([noise_timestep])
)
latents = (1 - init_mask) * init_latents_proper + init_mask * latents
# 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()
......@@ -1241,6 +1757,204 @@ class SDXLLongPromptWeightingPipeline(DiffusionPipeline, FromSingleFileMixin, Lo
return StableDiffusionXLPipelineOutput(images=image)
def text2img(
self,
prompt: str = None,
prompt_2: Optional[str] = None,
height: Optional[int] = None,
width: Optional[int] = None,
num_inference_steps: int = 50,
timesteps: List[int] = None,
denoising_start: Optional[float] = None,
denoising_end: Optional[float] = None,
guidance_scale: float = 5.0,
negative_prompt: Optional[str] = None,
negative_prompt_2: Optional[str] = None,
num_images_per_prompt: Optional[int] = 1,
eta: float = 0.0,
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
latents: Optional[torch.FloatTensor] = None,
prompt_embeds: Optional[torch.FloatTensor] = None,
negative_prompt_embeds: Optional[torch.FloatTensor] = None,
pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
output_type: Optional[str] = "pil",
return_dict: bool = True,
callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
callback_steps: int = 1,
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
guidance_rescale: float = 0.0,
original_size: Optional[Tuple[int, int]] = None,
crops_coords_top_left: Tuple[int, int] = (0, 0),
target_size: Optional[Tuple[int, int]] = None,
):
return self.__call__(
prompt=prompt,
prompt_2=prompt_2,
height=height,
width=width,
num_inference_steps=num_inference_steps,
timesteps=timesteps,
denoising_start=denoising_start,
denoising_end=denoising_end,
guidance_scale=guidance_scale,
negative_prompt=negative_prompt,
negative_prompt_2=negative_prompt_2,
num_images_per_prompt=num_images_per_prompt,
eta=eta,
generator=generator,
latents=latents,
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
output_type=output_type,
return_dict=return_dict,
callback=callback,
callback_steps=callback_steps,
cross_attention_kwargs=cross_attention_kwargs,
guidance_rescale=guidance_rescale,
original_size=original_size,
crops_coords_top_left=crops_coords_top_left,
target_size=target_size,
)
def img2img(
self,
prompt: str = None,
prompt_2: Optional[str] = None,
image: Optional[PipelineImageInput] = None,
height: Optional[int] = None,
width: Optional[int] = None,
strength: float = 0.8,
num_inference_steps: int = 50,
timesteps: List[int] = None,
denoising_start: Optional[float] = None,
denoising_end: Optional[float] = None,
guidance_scale: float = 5.0,
negative_prompt: Optional[str] = None,
negative_prompt_2: Optional[str] = None,
num_images_per_prompt: Optional[int] = 1,
eta: float = 0.0,
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
latents: Optional[torch.FloatTensor] = None,
prompt_embeds: Optional[torch.FloatTensor] = None,
negative_prompt_embeds: Optional[torch.FloatTensor] = None,
pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
output_type: Optional[str] = "pil",
return_dict: bool = True,
callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
callback_steps: int = 1,
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
guidance_rescale: float = 0.0,
original_size: Optional[Tuple[int, int]] = None,
crops_coords_top_left: Tuple[int, int] = (0, 0),
target_size: Optional[Tuple[int, int]] = None,
):
return self.__call__(
prompt=prompt,
prompt_2=prompt_2,
image=image,
height=height,
width=width,
strength=strength,
num_inference_steps=num_inference_steps,
timesteps=timesteps,
denoising_start=denoising_start,
denoising_end=denoising_end,
guidance_scale=guidance_scale,
negative_prompt=negative_prompt,
negative_prompt_2=negative_prompt_2,
num_images_per_prompt=num_images_per_prompt,
eta=eta,
generator=generator,
latents=latents,
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
output_type=output_type,
return_dict=return_dict,
callback=callback,
callback_steps=callback_steps,
cross_attention_kwargs=cross_attention_kwargs,
guidance_rescale=guidance_rescale,
original_size=original_size,
crops_coords_top_left=crops_coords_top_left,
target_size=target_size,
)
def inpaint(
self,
prompt: str = None,
prompt_2: Optional[str] = None,
image: Optional[PipelineImageInput] = None,
mask_image: Optional[PipelineImageInput] = None,
masked_image_latents: Optional[torch.FloatTensor] = None,
height: Optional[int] = None,
width: Optional[int] = None,
strength: float = 0.8,
num_inference_steps: int = 50,
timesteps: List[int] = None,
denoising_start: Optional[float] = None,
denoising_end: Optional[float] = None,
guidance_scale: float = 5.0,
negative_prompt: Optional[str] = None,
negative_prompt_2: Optional[str] = None,
num_images_per_prompt: Optional[int] = 1,
eta: float = 0.0,
generator: Optional[Union[torch.Generator, List[torch.Generator]]] = None,
latents: Optional[torch.FloatTensor] = None,
prompt_embeds: Optional[torch.FloatTensor] = None,
negative_prompt_embeds: Optional[torch.FloatTensor] = None,
pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
negative_pooled_prompt_embeds: Optional[torch.FloatTensor] = None,
output_type: Optional[str] = "pil",
return_dict: bool = True,
callback: Optional[Callable[[int, int, torch.FloatTensor], None]] = None,
callback_steps: int = 1,
cross_attention_kwargs: Optional[Dict[str, Any]] = None,
guidance_rescale: float = 0.0,
original_size: Optional[Tuple[int, int]] = None,
crops_coords_top_left: Tuple[int, int] = (0, 0),
target_size: Optional[Tuple[int, int]] = None,
):
return self.__call__(
prompt=prompt,
prompt_2=prompt_2,
image=image,
mask_image=mask_image,
masked_image_latents=masked_image_latents,
height=height,
width=width,
strength=strength,
num_inference_steps=num_inference_steps,
timesteps=timesteps,
denoising_start=denoising_start,
denoising_end=denoising_end,
guidance_scale=guidance_scale,
negative_prompt=negative_prompt,
negative_prompt_2=negative_prompt_2,
num_images_per_prompt=num_images_per_prompt,
eta=eta,
generator=generator,
latents=latents,
prompt_embeds=prompt_embeds,
negative_prompt_embeds=negative_prompt_embeds,
pooled_prompt_embeds=pooled_prompt_embeds,
negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
output_type=output_type,
return_dict=return_dict,
callback=callback,
callback_steps=callback_steps,
cross_attention_kwargs=cross_attention_kwargs,
guidance_rescale=guidance_rescale,
original_size=original_size,
crops_coords_top_left=crops_coords_top_left,
target_size=target_size,
)
# Overrride to properly handle the loading and unloading of the additional text encoder.
def load_lora_weights(self, pretrained_model_name_or_path_or_dict: Union[str, Dict[str, torch.Tensor]], **kwargs):
# We could have accessed the unet config from `lora_state_dict()` too. We pass
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment