diff --git a/CHANGELOG.md b/CHANGELOG.md
index 900f38912..018c81608 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,6 +1,6 @@
# Change Log for SD.Next
-## Update for 2024-02-19
+## Update for 2024-02-20
- **Improvements**:
- **IP Adapter** major refactor
@@ -13,27 +13,32 @@
- unified interface in txt2img, img2img and control
- enhanced xyz grid support
- **FaceID** now also works with multiple input images!
- - **Regional prompting** as a built-in solution
- simply enable from scripts -> regional prompting
- supports sd-15 models
- note that usage is same as original implementation from @hako-mikan
+ - [Differential diffusion](https://github.com/exx8/differential-diffusion)
+ img2img generation where you control strength of each pixel or image area
+ can be used with manually created masks or with auto-generated depth-maps
+ simply enable from *img2img -> scripts -> differential diffusion*
+ *note*: supports sd-xl models only
+ - [Regional prompting](https://github.com/huggingface/diffusers/blob/main/examples/community/README.md#regional-prompting-pipeline) as a built-in solution
+ usage is same as original implementation from @hako-mikan
click on title to open docs and see examples of full syntax on how to use it
- - **Clip-skip** reworked completely, thanks @AI-Casanova & @Disty0
- now clip-skip range is 0-12 where previously lowest value was 1 (default is still 1)
- values can also be decimal to interpolate between different layers, for example `clip-skip: 1.5`, thanks @AI-Casanova
- - **CFG End** new param to control image generation guidance, thanks @AI-Casanova
- sometimes you want strong control over composition, but you want it to stop at some point
- for example, when used with ip-adapters or controlnet, high cfg scale can overpower the guided image
- - **Outpaint** control outpaint now uses new alghorithm: noised-edge-extend
- new method allows for much larger outpaint areas in a single pass, even outpaint 512->1024 works well
- note that denoise strength should be increased for larger the outpaint areas, for example outpainting 512->1024 works well with denoise 0.75
- outpaint can run in *img2img* mode (default) and *inpaint* mode where original image is masked (if inpaint masked only is selected)
+ simply enable from *scripts -> regional prompting*
+ *note*: supports sd-15 models only
- [DeepCache](https://github.com/horseee/DeepCache) model acceleration
it can produce massive speedups (2x-5x) with no overhead, but with some loss of quality
*settings -> compute -> model compile -> deep-cache* and *settings -> compute -> model compile -> cache interval*
- [ZLUDA](https://github.com/vosen/ZLUDA) experimental support, thanks @lshqqytiger
- ZLUDA is CUDA wrapper that can be used for GPUs without native support
- best use case is *AMD GPUs on Windows*, see [wiki](https://github.com/vladmandic/automatic/wiki/ZLUDA) for details
+ - **Outpaint** control outpaint now uses new alghorithm: noised-edge-extend
+ new method allows for much larger outpaint areas in a single pass, even outpaint 512->1024 works well
+ note that denoise strength should be increased for larger the outpaint areas, for example outpainting 512->1024 works well with denoise 0.75
+ outpaint can run in *img2img* mode (default) and *inpaint* mode where original image is masked (if inpaint masked only is selected)
+ - **Clip-skip** reworked completely, thanks @AI-Casanova & @Disty0
+ now clip-skip range is 0-12 where previously lowest value was 1 (default is still 1)
+ values can also be decimal to interpolate between different layers, for example `clip-skip: 1.5`, thanks @AI-Casanova
+ - **CFG End** new param to control image generation guidance, thanks @AI-Casanova
+ sometimes you want strong control over composition, but you want it to stop at some point
+ for example, when used with ip-adapters or controlnet, high cfg scale can overpower the guided image
- **Control**
- when performing inpainting, you can specify processing resolution using **size->mask**
- units now have extra option to re-use current preview image as processor input
diff --git a/extensions-builtin/Lora/extra_networks_lora.py b/extensions-builtin/Lora/extra_networks_lora.py
index c8771e4de..57df5b010 100644
--- a/extensions-builtin/Lora/extra_networks_lora.py
+++ b/extensions-builtin/Lora/extra_networks_lora.py
@@ -67,7 +67,10 @@ class ExtraNetworkLora(extra_networks.ExtraNetwork):
if 'CLIP' in shared.sd_model.text_encoder.__class__.__name__ and not (shared.compiled_model_state is not None and shared.compiled_model_state.is_compiled is True):
if shared.opts.lora_fuse_diffusers:
shared.sd_model.unfuse_lora()
- shared.sd_model.unload_lora_weights()
+ try:
+ shared.sd_model.unload_lora_weights() # fails for non-CLIP models
+ except Exception:
+ pass
if not self.active and getattr(networks, "originals", None ) is not None:
networks.originals.undo() # remove patches
if networks.debug:
diff --git a/modules/control/proc/dpt.py b/modules/control/proc/dpt.py
index f726bf58f..c2c82386b 100644
--- a/modules/control/proc/dpt.py
+++ b/modules/control/proc/dpt.py
@@ -10,16 +10,21 @@ image_processor: AutoImageProcessor = None
class DPTDetector:
- def __init__(self, model=None, processor=None):
+ def __init__(self, model=None, processor=None, model_path=None):
self.model = model
self.processor = processor
+ self.model_path = model_path or "Intel/dpt-large"
- def __call__(self, input_image=None):
+ def __call__(self, input_image=None, model_path=None):
from modules.control.processors import cache_dir
+ if model_path is not None and model_path != self.model_path:
+ self.model_path = model_path
+ self.processor = None
+ self.model = None
if self.processor is None:
- self.processor = AutoImageProcessor.from_pretrained("Intel/dpt-large", cache_dir=cache_dir)
+ self.processor = AutoImageProcessor.from_pretrained(self.model_path, cache_dir=cache_dir)
if self.model is None:
- self.model = DPTForDepthEstimation.from_pretrained("Intel/dpt-large", cache_dir=cache_dir)
+ self.model = DPTForDepthEstimation.from_pretrained(self.model_path, cache_dir=cache_dir)
self.model.to(devices.device)
with devices.inference_context():
diff --git a/modules/processing_diffusers.py b/modules/processing_diffusers.py
index c28b893bd..16c86024b 100644
--- a/modules/processing_diffusers.py
+++ b/modules/processing_diffusers.py
@@ -266,32 +266,15 @@ def process_diffusers(p: processing.StableDiffusionProcessing):
clean.pop('callback_steps', None)
clean.pop('callback_on_step_end', None)
clean.pop('callback_on_step_end_tensor_inputs', None)
- if 'latents' in clean:
- clean['latents'] = clean['latents'].shape
- if 'image' in clean:
- clean['image'] = type(clean['image'])
- if 'mask_image' in clean:
- clean['mask_image'] = type(clean['mask_image'])
- if 'masked_image_latents' in clean:
- clean['masked_image_latents'] = type(clean['masked_image_latents'])
- if 'ip_adapter_image' in clean:
- clean['ip_adapter_image'] = type(clean['ip_adapter_image'])
if 'prompt' in clean:
clean['prompt'] = len(clean['prompt'])
if 'negative_prompt' in clean:
clean['negative_prompt'] = len(clean['negative_prompt'])
- if 'prompt_embeds' in clean:
- clean['prompt_embeds'] = clean['prompt_embeds'].shape if torch.is_tensor(clean['prompt_embeds']) else type(clean['prompt_embeds'])
- if 'pooled_prompt_embeds' in clean:
- clean['pooled_prompt_embeds'] = clean['pooled_prompt_embeds'].shape if torch.is_tensor(clean['pooled_prompt_embeds']) else type(clean['pooled_prompt_embeds'])
- if 'negative_prompt_embeds' in clean:
- clean['negative_prompt_embeds'] = clean['negative_prompt_embeds'].shape if torch.is_tensor(clean['negative_prompt_embeds']) else type(clean['negative_prompt_embeds'])
- if 'negative_pooled_prompt_embeds' in clean:
- clean['negative_pooled_prompt_embeds'] = clean['negative_pooled_prompt_embeds'].shape if torch.is_tensor(clean['negative_pooled_prompt_embeds']) else type(clean['negative_pooled_prompt_embeds'])
- if 'image_embeds' in clean:
- clean['image_embeds'] = clean['image_embeds'].shape if torch.is_tensor(clean['image_embeds']) else type(clean['image_embeds'])
clean['generator'] = generator_device
clean['parser'] = parser
+ for k, v in clean.items():
+ if isinstance(v, torch.Tensor):
+ clean[k] = v.shape
shared.log.debug(f'Diffuser pipeline: {model.__class__.__name__} task={sd_models.get_diffusers_task(model)} set={clean}')
if p.hdr_clamp or p.hdr_maximize or p.hdr_brightness != 0 or p.hdr_color != 0 or p.hdr_sharpen != 0:
txt = 'HDR:'
diff --git a/scripts/differential_diffusion.py b/scripts/differential_diffusion.py
new file mode 100644
index 000000000..ee9d02964
--- /dev/null
+++ b/scripts/differential_diffusion.py
@@ -0,0 +1,1147 @@
+"""
+credits: https://github.com/exx8/differential-diffusion
+code from: https://github.com/exx8/differential-diffusion/blob/main/SDXL/diff_pipe.py
+sdnext implementation follows after pipeline-end
+"""
+
+### pipeline start
+
+import inspect
+import hashlib
+from typing import Any, Callable, Dict, List, Optional, Tuple, Union
+
+import numpy as np
+import PIL.Image
+import torch
+from transformers import CLIPTextModel, CLIPTextModelWithProjection, CLIPTokenizer
+import torchvision
+
+from diffusers.image_processor import VaeImageProcessor
+from diffusers.loaders import FromSingleFileMixin, LoraLoaderMixin, TextualInversionLoaderMixin
+from diffusers.models import AutoencoderKL, UNet2DConditionModel
+from diffusers.models.attention_processor import (
+ AttnProcessor2_0,
+ LoRAAttnProcessor2_0,
+ LoRAXFormersAttnProcessor,
+ XFormersAttnProcessor,
+)
+from diffusers.schedulers import KarrasDiffusionSchedulers
+from diffusers.utils import (
+ is_accelerate_available,
+ is_accelerate_version,
+ logging,
+ replace_example_docstring,
+)
+from diffusers.utils.torch_utils import randn_tensor
+from diffusers.pipelines.pipeline_utils import DiffusionPipeline
+from diffusers.pipelines.stable_diffusion_xl.pipeline_output import StableDiffusionXLPipelineOutput
+
+
+logger = logging.get_logger(__name__) # pylint: disable=invalid-name
+
+EXAMPLE_DOC_STRING = """
+ Examples:
+ ```py
+ >>> import torch
+ >>> from diffusers import StableDiffusionXLImg2ImgPipeline
+ >>> from diffusers.utils import load_image
+
+ >>> pipe = StableDiffusionXLImg2ImgPipeline.from_pretrained(
+ ... "stabilityai/stable-diffusion-xl-refiner-1.0", torch_dtype=torch.float16
+ ... )
+ >>> pipe = pipe.to("cuda")
+ >>> url = "https://huggingface.co/datasets/patrickvonplaten/images/resolve/main/aa_xl/000000009.png"
+
+ >>> init_image = load_image(url).convert("RGB")
+ >>> prompt = "a photo of an astronaut riding a horse on mars"
+ >>> image = pipe(prompt, image=init_image).images[0]
+ ```
+"""
+
+
+# Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.rescale_noise_cfg
+def rescale_noise_cfg(noise_cfg, noise_pred_text, guidance_rescale=0.0):
+ """
+ Rescale `noise_cfg` according to `guidance_rescale`. Based on findings of [Common Diffusion Noise Schedules and
+ Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf). See Section 3.4
+ """
+ std_text = noise_pred_text.std(dim=list(range(1, noise_pred_text.ndim)), keepdim=True)
+ std_cfg = noise_cfg.std(dim=list(range(1, noise_cfg.ndim)), keepdim=True)
+ # rescale the results from guidance (fixes overexposure)
+ noise_pred_rescaled = noise_cfg * (std_text / std_cfg)
+ # mix with the original results from guidance by factor guidance_rescale to avoid "plain looking" images
+ noise_cfg = guidance_rescale * noise_pred_rescaled + (1 - guidance_rescale) * noise_cfg
+ return noise_cfg
+
+
+class StableDiffusionXLDiffImg2ImgPipeline(DiffusionPipeline, FromSingleFileMixin, LoraLoaderMixin):
+ r"""
+ Pipeline for text-to-image generation using Stable Diffusion XL.
+
+ This model inherits from [`DiffusionPipeline`]. Check the superclass documentation for the generic methods the
+ library implements for all the pipelines (such as downloading or saving, running on a particular device, etc.)
+
+ In addition the pipeline inherits the following loading methods:
+ - *Textual-Inversion*: [`loaders.TextualInversionLoaderMixin.load_textual_inversion`]
+ - *LoRA*: [`loaders.LoraLoaderMixin.load_lora_weights`]
+ - *Ckpt*: [`loaders.FromSingleFileMixin.from_single_file`]
+
+ as well as the following saving methods:
+ - *LoRA*: [`loaders.LoraLoaderMixin.save_lora_weights`]
+
+ Args:
+ vae ([`AutoencoderKL`]):
+ Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
+ text_encoder ([`CLIPTextModel`]):
+ Frozen text-encoder. Stable Diffusion XL uses the text portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModel), specifically
+ the [clip-vit-large-patch14](https://huggingface.co/openai/clip-vit-large-patch14) variant.
+ text_encoder_2 ([` CLIPTextModelWithProjection`]):
+ Second frozen text-encoder. Stable Diffusion XL uses the text and pool portion of
+ [CLIP](https://huggingface.co/docs/transformers/model_doc/clip#transformers.CLIPTextModelWithProjection),
+ specifically the
+ [laion/CLIP-ViT-bigG-14-laion2B-39B-b160k](https://huggingface.co/laion/CLIP-ViT-bigG-14-laion2B-39B-b160k)
+ variant.
+ tokenizer (`CLIPTokenizer`):
+ Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ tokenizer_2 (`CLIPTokenizer`):
+ Second Tokenizer of class
+ [CLIPTokenizer](https://huggingface.co/docs/transformers/v4.21.0/en/model_doc/clip#transformers.CLIPTokenizer).
+ unet ([`UNet2DConditionModel`]): Conditional U-Net architecture to denoise the encoded image latents.
+ scheduler ([`SchedulerMixin`]):
+ A scheduler to be used in combination with `unet` to denoise the encoded image latents. Can be one of
+ [`DDIMScheduler`], [`LMSDiscreteScheduler`], or [`PNDMScheduler`].
+ """
+ _optional_components = ["tokenizer", "text_encoder"]
+
+ def __init__(
+ self,
+ vae: AutoencoderKL,
+ text_encoder: CLIPTextModel,
+ text_encoder_2: CLIPTextModelWithProjection,
+ tokenizer: CLIPTokenizer,
+ tokenizer_2: CLIPTokenizer,
+ unet: UNet2DConditionModel,
+ scheduler: KarrasDiffusionSchedulers,
+ requires_aesthetics_score: bool = False,
+ force_zeros_for_empty_prompt: bool = True,
+ add_watermarker: Optional[bool] = None, # pylint: disable=unused-argument
+ ):
+ super().__init__()
+
+ self.register_modules(
+ vae=vae,
+ text_encoder=text_encoder,
+ text_encoder_2=text_encoder_2,
+ tokenizer=tokenizer,
+ tokenizer_2=tokenizer_2,
+ unet=unet,
+ scheduler=scheduler,
+ )
+ self.register_to_config(force_zeros_for_empty_prompt=force_zeros_for_empty_prompt)
+ self.register_to_config(requires_aesthetics_score=requires_aesthetics_score)
+ 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.watermark = None
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_slicing
+ def enable_vae_slicing(self):
+ r"""
+ Enable sliced VAE decoding. When this option is enabled, the VAE will split the input tensor in slices to
+ compute decoding in several steps. This is useful to save some memory and allow larger batch sizes.
+ """
+ self.vae.enable_slicing()
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_vae_slicing
+ def disable_vae_slicing(self):
+ r"""
+ Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
+ computing decoding in one step.
+ """
+ self.vae.disable_slicing()
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.enable_vae_tiling
+ def enable_vae_tiling(self):
+ r"""
+ Enable tiled VAE decoding. When this option is enabled, the VAE will split the input tensor into tiles to
+ compute decoding and encoding in several steps. This is useful for saving a large amount of memory and to allow
+ processing larger images.
+ """
+ self.vae.enable_tiling()
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.disable_vae_tiling
+ def disable_vae_tiling(self):
+ r"""
+ Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
+ computing decoding in one step.
+ """
+ self.vae.disable_tiling()
+
+ def enable_model_cpu_offload(self, gpu_id=0): # pylint: disable=arguments-differ
+ r"""
+ Offloads all models to CPU using accelerate, reducing memory usage with a low impact on performance. Compared
+ to `enable_sequential_cpu_offload`, this method moves one whole model at a time to the GPU when its `forward`
+ method is called, and the model remains in GPU until the next model runs. Memory savings are lower than with
+ `enable_sequential_cpu_offload`, but performance is much better due to the iterative execution of the `unet`.
+ """
+ if is_accelerate_available() and is_accelerate_version(">=", "0.17.0.dev0"):
+ from accelerate import cpu_offload_with_hook
+ else:
+ raise ImportError("`enable_model_cpu_offload` requires `accelerate v0.17.0` or higher.")
+
+ device = torch.device(f"cuda:{gpu_id}")
+
+ if self.device.type != "cpu":
+ self.to("cpu", silence_dtype_warnings=True)
+ torch.cuda.empty_cache() # otherwise we don't see the memory savings (but they probably exist)
+
+ model_sequence = (
+ [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2]
+ )
+ model_sequence.extend([self.unet, self.vae])
+
+ hook = None
+ for cpu_offloaded_model in model_sequence:
+ _, hook = cpu_offload_with_hook(cpu_offloaded_model, device, prev_module_hook=hook)
+
+ # We'll offload the last model manually.
+ self.final_offload_hook = hook # pylint: disable=attribute-defined-outside-init
+
+ # Copied from diffusers.pipelines.stable_diffusion_xl.pipeline_stable_diffusion_xl.StableDiffusionXLPipeline.encode_prompt
+ def encode_prompt(
+ self,
+ prompt: str,
+ prompt_2: Optional[str] = None,
+ device: Optional[torch.device] = None,
+ num_images_per_prompt: int = 1,
+ do_classifier_free_guidance: bool = True,
+ negative_prompt: Optional[str] = None,
+ negative_prompt_2: Optional[str] = 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,
+ lora_scale: Optional[float] = None,
+ ):
+ r"""
+ Encodes the prompt into text encoder hidden states.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ prompt to be encoded
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ device: (`torch.device`):
+ torch device
+ num_images_per_prompt (`int`):
+ number of images that should be generated per prompt
+ do_classifier_free_guidance (`bool`):
+ whether to use classifier free guidance or not
+ 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`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ 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.
+ 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.
+ pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ lora_scale (`float`, *optional*):
+ A lora scale that will be applied to all LoRA layers of the text encoder if LoRA layers are loaded.
+ """
+ device = device or self._execution_device
+
+ # set lora scale so that monkey patched LoRA
+ # function of text encoder can correctly access it
+ if lora_scale is not None and isinstance(self, LoraLoaderMixin):
+ self._lora_scale = lora_scale # pylint: disable=attribute-defined-outside-init
+
+ 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]
+
+ # Define tokenizers and text encoders
+ tokenizers = [self.tokenizer, self.tokenizer_2] if self.tokenizer is not None else [self.tokenizer_2]
+ text_encoders = (
+ [self.text_encoder, self.text_encoder_2] if self.text_encoder is not None else [self.text_encoder_2]
+ )
+
+ if prompt_embeds is None:
+ prompt_2 = prompt_2 or prompt
+ # textual inversion: procecss multi-vector tokens if necessary
+ prompt_embeds_list = []
+ prompts = [prompt, prompt_2]
+ for prompt, tokenizer, text_encoder in zip(prompts, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ prompt = self.maybe_convert_prompt(prompt, tokenizer)
+
+ text_inputs = tokenizer(
+ prompt,
+ padding="max_length",
+ max_length=tokenizer.model_max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ text_input_ids = text_inputs.input_ids
+ untruncated_ids = tokenizer(prompt, padding="longest", return_tensors="pt").input_ids
+ untruncated_ids = 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 = tokenizer.batch_decode(untruncated_ids[:, tokenizer.model_max_length - 1 : -1])
+ logger.warning(
+ "The following part of your input was truncated because CLIP can only handle sequences up to"
+ f" {tokenizer.model_max_length} tokens: {removed_text}"
+ )
+
+ prompt_embeds = text_encoder(
+ text_input_ids.to(device),
+ output_hidden_states=True,
+ )
+
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ pooled_prompt_embeds = prompt_embeds[0]
+ prompt_embeds = prompt_embeds.hidden_states[-2]
+
+ prompt_embeds_list.append(prompt_embeds)
+
+ prompt_embeds = torch.concat(prompt_embeds_list, dim=-1)
+
+ # get unconditional embeddings for classifier free guidance
+ zero_out_negative_prompt = negative_prompt is None and self.config.force_zeros_for_empty_prompt # pylint: disable=no-member
+ if do_classifier_free_guidance and negative_prompt_embeds is None and zero_out_negative_prompt:
+ negative_prompt_embeds = torch.zeros_like(prompt_embeds)
+ negative_pooled_prompt_embeds = torch.zeros_like(pooled_prompt_embeds)
+ elif do_classifier_free_guidance and negative_prompt_embeds is None:
+ negative_prompt = negative_prompt or ""
+ negative_prompt_2 = negative_prompt_2 or negative_prompt
+
+ uncond_tokens: List[str]
+ if prompt is not None and type(prompt) is not type(negative_prompt):
+ raise TypeError(
+ f"`negative_prompt` should be the same type to `prompt`, but got {type(negative_prompt)} !="
+ f" {type(prompt)}."
+ )
+ elif isinstance(negative_prompt, str):
+ uncond_tokens = [negative_prompt, negative_prompt_2]
+ elif batch_size != len(negative_prompt):
+ raise ValueError(
+ f"`negative_prompt`: {negative_prompt} has batch size {len(negative_prompt)}, but `prompt`:"
+ f" {prompt} has batch size {batch_size}. Please make sure that passed `negative_prompt` matches"
+ " the batch size of `prompt`."
+ )
+ else:
+ uncond_tokens = [negative_prompt, negative_prompt_2]
+
+ negative_prompt_embeds_list = []
+ for negative_prompt, tokenizer, text_encoder in zip(uncond_tokens, tokenizers, text_encoders):
+ if isinstance(self, TextualInversionLoaderMixin):
+ negative_prompt = self.maybe_convert_prompt(negative_prompt, tokenizer)
+
+ max_length = prompt_embeds.shape[1]
+ uncond_input = tokenizer(
+ negative_prompt,
+ padding="max_length",
+ max_length=max_length,
+ truncation=True,
+ return_tensors="pt",
+ )
+
+ negative_prompt_embeds = text_encoder(
+ uncond_input.input_ids.to(device),
+ output_hidden_states=True,
+ )
+ # We are only ALWAYS interested in the pooled output of the final text encoder
+ negative_pooled_prompt_embeds = negative_prompt_embeds[0]
+ negative_prompt_embeds = negative_prompt_embeds.hidden_states[-2]
+
+ negative_prompt_embeds_list.append(negative_prompt_embeds)
+
+ negative_prompt_embeds = torch.concat(negative_prompt_embeds_list, dim=-1)
+
+ prompt_embeds = prompt_embeds.to(dtype=self.text_encoder_2.dtype, device=device)
+ bs_embed, seq_len, _ = prompt_embeds.shape
+ # duplicate text embeddings for each generation per prompt, using mps friendly method
+ prompt_embeds = prompt_embeds.repeat(1, num_images_per_prompt, 1)
+ prompt_embeds = prompt_embeds.view(bs_embed * num_images_per_prompt, seq_len, -1)
+
+ 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=self.text_encoder_2.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)
+
+ pooled_prompt_embeds = pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+ if do_classifier_free_guidance:
+ negative_pooled_prompt_embeds = negative_pooled_prompt_embeds.repeat(1, num_images_per_prompt).view(
+ bs_embed * num_images_per_prompt, -1
+ )
+
+ return prompt_embeds, negative_prompt_embeds, pooled_prompt_embeds, negative_pooled_prompt_embeds
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion.StableDiffusionPipeline.prepare_extra_step_kwargs
+ def prepare_extra_step_kwargs(self, generator, eta):
+ # prepare extra kwargs for the scheduler step, since not all schedulers have the same signature
+ # eta (η) is only used with the DDIMScheduler, it will be ignored for other schedulers.
+ # eta corresponds to η in DDIM paper: https://arxiv.org/abs/2010.02502
+ # and should be between [0, 1]
+
+ accepts_eta = "eta" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ extra_step_kwargs = {}
+ if accepts_eta:
+ extra_step_kwargs["eta"] = eta
+
+ # check if the scheduler accepts generator
+ accepts_generator = "generator" in set(inspect.signature(self.scheduler.step).parameters.keys())
+ if accepts_generator:
+ extra_step_kwargs["generator"] = generator
+ return extra_step_kwargs
+
+ def check_inputs(
+ self,
+ prompt,
+ prompt_2,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ negative_prompt=None,
+ negative_prompt_2=None,
+ prompt_embeds=None,
+ negative_prompt_embeds=None,
+ ):
+ if strength < 0 or strength > 1:
+ raise ValueError(f"The value of strength should in [0.0, 1.0] but is {strength}")
+ if num_inference_steps is None:
+ raise ValueError("`num_inference_steps` cannot be None.")
+ elif not isinstance(num_inference_steps, int) or num_inference_steps <= 0:
+ raise ValueError(
+ f"`num_inference_steps` has to be a positive integer but is {num_inference_steps} of type"
+ f" {type(num_inference_steps)}."
+ )
+ if (callback_steps is None) or (
+ callback_steps is not None and (not isinstance(callback_steps, int) or callback_steps <= 0)
+ ):
+ raise ValueError(
+ f"`callback_steps` has to be a positive integer but is {callback_steps} of type"
+ f" {type(callback_steps)}."
+ )
+
+ if prompt is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt`: {prompt} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt_2 is not None and prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `prompt_2`: {prompt_2} and `prompt_embeds`: {prompt_embeds}. Please make sure to"
+ " only forward one of the two."
+ )
+ elif prompt is None and prompt_embeds is None:
+ raise ValueError(
+ "Provide either `prompt` or `prompt_embeds`. Cannot leave both `prompt` and `prompt_embeds` undefined."
+ )
+ elif prompt is not None and (not isinstance(prompt, str) and not isinstance(prompt, list)):
+ raise ValueError(f"`prompt` has to be of type `str` or `list` but is {type(prompt)}")
+ elif prompt_2 is not None and (not isinstance(prompt_2, str) and not isinstance(prompt_2, list)):
+ raise ValueError(f"`prompt_2` has to be of type `str` or `list` but is {type(prompt_2)}")
+
+ if 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."
+ )
+ elif negative_prompt_2 is not None and negative_prompt_embeds is not None:
+ raise ValueError(
+ f"Cannot forward both `negative_prompt_2`: {negative_prompt_2} 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 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}."
+ )
+
+ def get_timesteps(self, num_inference_steps, strength, device, denoising_start=None): # pylint: disable=unused-argument
+ # 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)
+ )
+ )
+ timesteps = list(filter(lambda ts: ts < discrete_timestep_cutoff, timesteps))
+ return torch.tensor(timesteps), len(timesteps)
+
+ return timesteps, num_inference_steps - t_start
+
+ def prepare_latents(
+ self, image, timestep, batch_size, num_images_per_prompt, dtype, device, generator=None, add_noise=True
+ ):
+ 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)}"
+ )
+
+ # 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)
+
+ batch_size = batch_size * num_images_per_prompt
+
+ 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 = [
+ self.vae.encode(image[i : i + 1]).latent_dist.sample(generator[i]) for i in range(batch_size)
+ ]
+ init_latents = torch.cat(init_latents, dim=0)
+ else:
+ init_latents = self.vae.encode(image).latent_dist.sample(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
+
+ def _get_add_time_ids(
+ self, original_size, crops_coords_top_left, target_size, aesthetic_score, negative_aesthetic_score, dtype
+ ):
+ if self.config.requires_aesthetics_score: # pylint: disable=no-member
+ add_time_ids = list(original_size + crops_coords_top_left + (aesthetic_score,))
+ add_neg_time_ids = list(original_size + crops_coords_top_left + (negative_aesthetic_score,))
+ else:
+ add_time_ids = list(original_size + crops_coords_top_left + target_size)
+ add_neg_time_ids = list(original_size + crops_coords_top_left + target_size)
+
+ passed_add_embed_dim = (
+ self.unet.config.addition_time_embed_dim * len(add_time_ids) + self.text_encoder_2.config.projection_dim
+ )
+ expected_add_embed_dim = self.unet.add_embedding.linear_1.in_features
+
+ if (
+ expected_add_embed_dim > passed_add_embed_dim
+ and (expected_add_embed_dim - passed_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to enable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=True)` to make sure `aesthetic_score` {aesthetic_score} and `negative_aesthetic_score` {negative_aesthetic_score} is correctly used by the model."
+ )
+ elif (
+ expected_add_embed_dim < passed_add_embed_dim
+ and (passed_add_embed_dim - expected_add_embed_dim) == self.unet.config.addition_time_embed_dim
+ ):
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. Please make sure to disable `requires_aesthetics_score` with `pipe.register_to_config(requires_aesthetics_score=False)` to make sure `target_size` {target_size} is correctly used by the model."
+ )
+ elif expected_add_embed_dim != passed_add_embed_dim:
+ raise ValueError(
+ f"Model expects an added time embedding vector of length {expected_add_embed_dim}, but a vector of {passed_add_embed_dim} was created. The model has an incorrect config. Please check `unet.config.time_embedding_type` and `text_encoder_2.config.projection_dim`."
+ )
+
+ add_time_ids = torch.tensor([add_time_ids], dtype=dtype)
+ add_neg_time_ids = torch.tensor([add_neg_time_ids], dtype=dtype)
+
+ return add_time_ids, add_neg_time_ids
+
+ # Copied from diffusers.pipelines.stable_diffusion.pipeline_stable_diffusion_upscale.StableDiffusionUpscalePipeline.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,
+ LoRAXFormersAttnProcessor,
+ LoRAAttnProcessor2_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)
+
+ @torch.no_grad()
+ @replace_example_docstring(EXAMPLE_DOC_STRING)
+ def __call__(
+ self,
+ prompt: Union[str, List[str]] = None,
+ prompt_2: Optional[Union[str, List[str]]] = None,
+ image: Union[
+ torch.FloatTensor,
+ PIL.Image.Image,
+ np.ndarray,
+ List[torch.FloatTensor],
+ List[PIL.Image.Image],
+ List[np.ndarray],
+ ] = None,
+ strength: float = 0.3,
+ num_inference_steps: int = 50,
+ denoising_start: Optional[float] = None,
+ denoising_end: Optional[float] = None,
+ guidance_scale: float = 5.0,
+ negative_prompt: Optional[Union[str, List[str]]] = None,
+ negative_prompt_2: Optional[Union[str, List[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: Tuple[int, int] = None,
+ crops_coords_top_left: Tuple[int, int] = (0, 0),
+ target_size: Tuple[int, int] = None,
+ aesthetic_score: float = 6.0,
+ negative_aesthetic_score: float = 2.5,
+ map: torch.FloatTensor = None, # pylint: disable=redefined-builtin
+ original_image: Union[
+ torch.FloatTensor,
+ PIL.Image.Image,
+ np.ndarray,
+ List[torch.FloatTensor],
+ List[PIL.Image.Image],
+ List[np.ndarray],
+ ] = None,
+ ):
+ r"""
+ Function invoked when calling the pipeline for generation.
+
+ Args:
+ prompt (`str` or `List[str]`, *optional*):
+ The prompt or prompts to guide the image generation. If not defined, one has to pass `prompt_embeds`.
+ instead.
+ prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts to be sent to the `tokenizer_2` and `text_encoder_2`. If not defined, `prompt` is
+ used in both text-encoders
+ image (`torch.FloatTensor` or `PIL.Image.Image` or `np.ndarray` or `List[torch.FloatTensor]` or `List[PIL.Image.Image]` or `List[np.ndarray]`):
+ The image(s) to modify with the pipeline.
+ strength (`float`, *optional*, defaults to 0.3):
+ 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`. Note that in the case of
+ `denoising_start` being declared as an integer, the value of `strength` will be ignored.
+ 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.
+ 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 [**Refining the Image
+ Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output).
+ 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 (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 [**Refining the Image
+ Output**](https://huggingface.co/docs/diffusers/api/pipelines/stable_diffusion/stable_diffusion_xl#refining-the-image-output).
+ guidance_scale (`float`, *optional*, defaults to 7.5):
+ 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.
+ 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`).
+ negative_prompt_2 (`str` or `List[str]`, *optional*):
+ The prompt or prompts not to guide the image generation to be sent to `tokenizer_2` and
+ `text_encoder_2`. If not defined, `negative_prompt` is used in both text-encoders
+ num_images_per_prompt (`int`, *optional*, defaults to 1):
+ The number of images to generate per prompt.
+ eta (`float`, *optional*, defaults to 0.0):
+ Corresponds to parameter eta (η) in the DDIM paper: https://arxiv.org/abs/2010.02502. Only applies to
+ [`schedulers.DDIMScheduler`], will be ignored for others.
+ 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.
+ 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.
+ pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt weighting.
+ If not provided, pooled text embeddings will be generated from `prompt` input argument.
+ negative_pooled_prompt_embeds (`torch.FloatTensor`, *optional*):
+ Pre-generated negative pooled text embeddings. Can be used to easily tweak text inputs, *e.g.* prompt
+ weighting. If not provided, pooled negative_prompt_embeds will be generated from `negative_prompt`
+ input argument.
+ output_type (`str`, *optional*, defaults to `"pil"`):
+ The output format of the generate image. Choose between
+ [PIL](https://pillow.readthedocs.io/en/stable/): `PIL.Image.Image` or `np.array`.
+ return_dict (`bool`, *optional*, defaults to `True`):
+ Whether or not to return a [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] instead of a
+ plain tuple.
+ callback (`Callable`, *optional*):
+ A function that will be called every `callback_steps` steps during inference. The function will be
+ called with the following arguments: `callback(step: int, timestep: int, latents: torch.FloatTensor)`.
+ callback_steps (`int`, *optional*, defaults to 1):
+ The frequency at which the `callback` function will be called. If not specified, the callback will be
+ called at every step.
+ cross_attention_kwargs (`dict`, *optional*):
+ A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under
+ `self.processor` in
+ [diffusers.cross_attention](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/cross_attention.py).
+ guidance_rescale (`float`, *optional*, defaults to 0.7):
+ Guidance rescale factor proposed by [Common Diffusion Noise Schedules and Sample Steps are
+ Flawed](https://arxiv.org/pdf/2305.08891.pdf) `guidance_scale` is defined as `φ` in equation 16. of
+ [Common Diffusion Noise Schedules and Sample Steps are Flawed](https://arxiv.org/pdf/2305.08891.pdf).
+ Guidance rescale factor should fix overexposure when using zero terminal SNR.
+ original_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ If `original_size` is not the same as `target_size` the image will appear to be down- or upsampled.
+ `original_size` defaults to `(width, height)` if not specified. Part of SDXL's micro-conditioning as
+ explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ crops_coords_top_left (`Tuple[int]`, *optional*, defaults to (0, 0)):
+ `crops_coords_top_left` can be used to generate an image that appears to be "cropped" from the position
+ `crops_coords_top_left` downwards. Favorable, well-centered images are usually achieved by setting
+ `crops_coords_top_left` to (0, 0). Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ target_size (`Tuple[int]`, *optional*, defaults to (1024, 1024)):
+ For most cases, `target_size` should be set to the desired height and width of the generated image. If
+ not specified it will default to `(width, height)`. Part of SDXL's micro-conditioning as explained in
+ section 2.2 of [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ aesthetic_score (`float`, *optional*, defaults to 6.0):
+ Used to simulate an aesthetic score of the generated image by influencing the positive text condition.
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952).
+ negative_aesthetic_score (`float`, *optional*, defaults to 2.5):
+ Part of SDXL's micro-conditioning as explained in section 2.2 of
+ [https://huggingface.co/papers/2307.01952](https://huggingface.co/papers/2307.01952). Can be used to
+ simulate an aesthetic score of the generated image by influencing the negative text condition.
+
+ Examples:
+
+ Returns:
+ [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] or `tuple`:
+ [`~pipelines.stable_diffusion.StableDiffusionXLPipelineOutput`] if `return_dict` is True, otherwise a
+ `tuple. When returning a tuple, the first element is a list with the generated images.
+ """
+ # 1. Check inputs. Raise error if not correct
+ self.check_inputs(
+ prompt,
+ prompt_2,
+ strength,
+ num_inference_steps,
+ callback_steps,
+ negative_prompt,
+ negative_prompt_2,
+ prompt_embeds,
+ negative_prompt_embeds,
+ )
+
+ # 2. Define call parameters
+ if prompt is not None and isinstance(prompt, str):
+ batch_size = 1
+ elif prompt is not None and isinstance(prompt, list):
+ batch_size = len(prompt)
+ else:
+ batch_size = prompt_embeds.shape[0]
+
+ device = self._execution_device
+
+ # 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
+
+ # 3. Encode input prompt
+ text_encoder_lora_scale = (
+ cross_attention_kwargs.get("scale", None) if cross_attention_kwargs is not None else None
+ )
+ (
+ prompt_embeds,
+ negative_prompt_embeds,
+ pooled_prompt_embeds,
+ negative_pooled_prompt_embeds,
+ ) = self.encode_prompt(
+ prompt=prompt,
+ prompt_2=prompt_2,
+ device=device,
+ num_images_per_prompt=num_images_per_prompt,
+ do_classifier_free_guidance=do_classifier_free_guidance,
+ negative_prompt=negative_prompt,
+ negative_prompt_2=negative_prompt_2,
+ prompt_embeds=prompt_embeds,
+ negative_prompt_embeds=negative_prompt_embeds,
+ pooled_prompt_embeds=pooled_prompt_embeds,
+ negative_pooled_prompt_embeds=negative_pooled_prompt_embeds,
+ lora_scale=text_encoder_lora_scale,
+ )
+
+ # 4. Preprocess image
+ #image = self.image_processor.preprocess(image) #ideally we would have preprocess the image with diffusers, but for this POC we won't --- it throws a deprecated warning
+ map = torchvision.transforms.Resize(tuple(s // self.vae_scale_factor for s in original_image.shape[2:]),antialias=None)(map)
+ # 5. Prepare timesteps
+ def denoising_value_valid(dnv):
+ return type(denoising_end) == float and 0 < dnv < 1
+
+ self.scheduler.set_timesteps(num_inference_steps, device=device)
+ #begin diff diff change
+ total_time_steps = num_inference_steps
+ #end diff diff change
+ timesteps, num_inference_steps = self.get_timesteps(
+ num_inference_steps, strength, device, denoising_start=denoising_start if denoising_value_valid else None # pylint: disable=missing-parentheses-for-call-in-test, using-constant-test
+ )
+ latent_timestep = timesteps[:1].repeat(batch_size * num_images_per_prompt)
+
+ add_noise = True if denoising_start is None else False
+ # 6. Prepare latent variables
+ latents = self.prepare_latents(
+ image,
+ latent_timestep,
+ batch_size,
+ num_images_per_prompt,
+ prompt_embeds.dtype,
+ device,
+ generator,
+ add_noise,
+ )
+ # 7. Prepare extra step kwargs.
+ 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)
+
+ # 8. Prepare added time ids & embeddings
+ add_text_embeds = pooled_prompt_embeds
+ add_time_ids, add_neg_time_ids = self._get_add_time_ids(
+ original_size,
+ crops_coords_top_left,
+ target_size,
+ aesthetic_score,
+ negative_aesthetic_score,
+ dtype=prompt_embeds.dtype,
+ )
+ add_time_ids = add_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+
+ if do_classifier_free_guidance:
+ prompt_embeds = torch.cat([negative_prompt_embeds, prompt_embeds], dim=0)
+ add_text_embeds = torch.cat([negative_pooled_prompt_embeds, add_text_embeds], dim=0)
+ add_neg_time_ids = add_neg_time_ids.repeat(batch_size * num_images_per_prompt, 1)
+ add_time_ids = torch.cat([add_neg_time_ids, add_time_ids], dim=0)
+
+ prompt_embeds = prompt_embeds.to(device)
+ add_text_embeds = add_text_embeds.to(device)
+ add_time_ids = add_time_ids.to(device)
+
+ # 9. Denoising loop
+ num_warmup_steps = max(len(timesteps) - num_inference_steps * self.scheduler.order, 0)
+
+
+ # 9.1 Apply denoising_end
+ if (
+ denoising_end is not None
+ and denoising_start is not None
+ and denoising_value_valid(denoising_end)
+ and denoising_value_valid(denoising_start)
+ and denoising_start >= denoising_end
+ ):
+ raise ValueError(
+ f"`denoising_start`: {denoising_start} cannot be larger than or equal to `denoising_end`: "
+ + f" {denoising_end} when using type float."
+ )
+ elif denoising_end is not None and denoising_value_valid(denoising_end):
+ discrete_timestep_cutoff = int(
+ round(
+ self.scheduler.config.num_train_timesteps
+ - (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]
+
+ # prepartions for diff diff
+ original_with_noise = self.prepare_latents(
+ original_image, timesteps, batch_size, num_images_per_prompt, prompt_embeds.dtype, device, generator
+ )
+ thresholds = torch.arange(total_time_steps, dtype=map.dtype) / total_time_steps
+ thresholds = thresholds.unsqueeze(1).unsqueeze(1).to(device)
+ masks = map > (thresholds + (denoising_start or 0))
+ # end diff diff preparations
+ with self.progress_bar(total=num_inference_steps) as progress_bar:
+ for i, t in enumerate(timesteps):
+ # diff diff
+ if i==0 and denoising_start is None:
+ latents = original_with_noise[:1]
+ else:
+ mask = masks[i].unsqueeze(0)
+ # cast mask to the same type as latents etc
+ mask = mask.to(latents.dtype)
+ mask = mask.unsqueeze(1) # fit shape
+ latents = original_with_noise[i] * mask + latents * (1 - mask)
+ # end diff diff
+ # expand the latents if we are doing classifier free guidance
+ latent_model_input = torch.cat([latents] * 2) if do_classifier_free_guidance else latents
+
+ latent_model_input = self.scheduler.scale_model_input(latent_model_input, t)
+
+ # 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,
+ added_cond_kwargs=added_cond_kwargs,
+ 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)
+
+ if do_classifier_free_guidance and guidance_rescale > 0.0:
+ # Based on 3.4. in https://arxiv.org/pdf/2305.08891.pdf
+ noise_pred = rescale_noise_cfg(noise_pred, noise_pred_text, guidance_rescale=guidance_rescale)
+
+ # 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]
+
+ # 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 callback is not None and i % callback_steps == 0:
+ callback(i, t, latents)
+
+ # make sure the VAE is in float32 mode, as it overflows in float16
+ if self.vae.dtype == torch.float16 and self.vae.config.force_upcast:
+ self.upcast_vae()
+ latents = latents.to(next(iter(self.vae.post_quant_conv.parameters())).dtype)
+
+ if output_type != "latent":
+ image = self.vae.decode(latents / self.vae.config.scaling_factor, return_dict=False)[0]
+ else:
+ image = latents
+ return StableDiffusionXLPipelineOutput(images=image)
+
+ # apply watermark if available
+ if self.watermark is not None:
+ image = self.watermark.apply_watermark(image)
+
+ image = self.image_processor.postprocess(image, output_type=output_type)
+
+ # Offload last model to CPU
+ if hasattr(self, "final_offload_hook") and self.final_offload_hook is not None:
+ self.final_offload_hook.offload()
+
+ if not return_dict:
+ return (image,)
+
+ return StableDiffusionXLPipelineOutput(images=image)
+
+### pipeline end
+
+### script start
+
+import gradio as gr
+import diffusers
+from PIL import Image, ImageEnhance, ImageOps # pylint: disable=reimported
+from torchvision import transforms
+from modules import errors, shared, devices, scripts, processing, sd_models, images
+
+
+detector = None
+MODELS = {
+ 'DPT Tiny': 'Intel/dpt-swinv2-tiny-256',
+ 'DPT Hybrid': 'Intel/dpt-hybrid-midas',
+ 'DPT Large': 'Intel/dpt-large'
+}
+
+
+class Script(scripts.Script):
+ def title(self):
+ return 'Differential diffusion'
+
+ def show(self, is_img2img):
+ return is_img2img if shared.backend == shared.Backend.DIFFUSERS else False
+
+ def ui(self, _is_img2img):
+ with gr.Row():
+ gr.HTML(' Differential diffusion
Select a model for auto-preprocess or upload an image map
')
+ with gr.Row():
+ enabled = gr.Checkbox(label='Enabled', value=True)
+ invert = gr.Checkbox(label='Mask invert', value=False)
+ strength = gr.Slider(minimum=0.0, maximum=2.0, value=1.0, label='Mask strength')
+ model = gr.Dropdown(label='Model', choices=['None', 'DPT Tiny', 'DPT Hybrid', 'DPT Large'], value='None')
+ with gr.Row():
+ image = gr.Image(label="Image map", show_label=False, type="pil", source="upload", interactive=True, tool="editor", visible=True, image_mode='RGB')
+ return enabled, strength, invert, model, image
+
+ def depthmap(self, image_init: Image.Image, image_map: Image.Image, model: str, strength: float, invert: bool):
+ global detector # pylint: disable=global-statement
+ from modules.control.proc.dpt import DPTDetector
+ if image_init is None:
+ return None, None
+ image_map = None
+ if image_map is not None:
+ image_map = image_map.resize(image_init.size, Image.Resampling.LANCZOS)
+ image_init = image_map
+ if model != 'None':
+ if detector is None:
+ detector = DPTDetector()
+ image_map = detector(image_init, MODELS[model])
+ if image_map is not None:
+ if strength != 1.0:
+ enhancer = ImageEnhance.Brightness(image_map)
+ image_map = enhancer.enhance(strength)
+ image_map = image_map.convert('L')
+ if invert:
+ image_map = ImageOps.invert(image_map)
+ if shared.opts.save_init_img:
+ init_img_hash = hashlib.sha256(image_map.tobytes()).hexdigest()[0:8] # pylint: disable=attribute-defined-outside-init
+ images.save_image(image_map, path=shared.opts.outdir_init_images, basename=None, forced_filename=init_img_hash, suffix="-init-image")
+ image_map = transforms.ToTensor()(image_map)
+ image_init = 2 * transforms.ToTensor()(image_init) - 1
+ image_init = image_init.unsqueeze(0)
+ return image_init.to(devices.device), image_map.to(devices.device)
+
+ def run(self, p: processing.StableDiffusionProcessingImg2Img, enabled, strength, invert, model, image): # pylint: disable=arguments-differ
+ if not enabled:
+ return
+ if shared.sd_model_type != 'sdxl':
+ shared.log.error(f'Differential-diffusion: incorrect base model: {shared.sd_model.__class__.__name__}')
+ return
+ if not hasattr(p, 'init_images') or len(p.init_images) == 0:
+ shared.log.error('Differential-diffusion: no input images')
+ return
+
+ image_init, image_map = self.depthmap(p.init_images[0], image, model, strength, invert)
+ if image_map is None:
+ shared.log.error('Differential-diffusion: no image map')
+ return
+
+ orig_pipeline = shared.sd_model
+ pipe = None
+ try:
+ diffusers.pipelines.auto_pipeline.AUTO_IMAGE2IMAGE_PIPELINES_MAPPING["StableDiffusionXLDiffImg2ImgPipeline"] = StableDiffusionXLDiffImg2ImgPipeline
+ pipe = StableDiffusionXLDiffImg2ImgPipeline(
+ vae=shared.sd_model.vae,
+ text_encoder=shared.sd_model.text_encoder,
+ text_encoder_2=shared.sd_model.text_encoder_2,
+ tokenizer=shared.sd_model.tokenizer,
+ tokenizer_2=shared.sd_model.tokenizer_2,
+ unet=shared.sd_model.unet,
+ scheduler=shared.sd_model.scheduler,
+ )
+ sd_models.copy_diffuser_options(pipe, shared.sd_model)
+ sd_models.set_diffuser_options(pipe)
+ p.task_args['image'] = image_init
+ p.task_args['original_image'] = image_init
+ p.task_args['map'] = image_map
+ shared.log.debug(f'Differential-diffusion: pipeline={pipe.__class__.__name__} strength={strength} model={model} image={image_init.shape} map={image_map.shape}')
+ shared.sd_model = pipe
+ except Exception as e:
+ shared.log.error(f'Differential-diffusion: pipeline creation failed: {e}')
+ errors.display(e, 'Differential-diffusion: pipeline creation failed')
+ shared.sd_model = orig_pipeline
+
+ # run pipeline
+ processed: processing.Processed = processing.process_images(p) # runs processing using main loop
+
+ # restore pipeline and params
+ pipe = None
+ shared.sd_model = orig_pipeline
+ devices.torch_gc()
+ return processed
diff --git a/wiki b/wiki
index 7654276a0..83b5a6083 160000
--- a/wiki
+++ b/wiki
@@ -1 +1 @@
-Subproject commit 7654276a07723988aefc17da885de8686ee0c530
+Subproject commit 83b5a60831be6fe70707be9bb7103e4c837e21e9