Editing Implicit Assumptions in Text-to-Image Diffusion Models by Hadas Orgad, Bahjat Kawar, and Yonatan Belinkov.
The abstract of the paper is the following:
Text-to-image diffusion models often make implicit assumptions about the world when generating images. While some assumptions are useful (e.g., the sky is blue), they can also be outdated, incorrect, or reflective of social biases present in the training data. Thus, there is a need to control these assumptions without requiring explicit user input or costly re-training. In this work, we aim to edit a given implicit assumption in a pre-trained diffusion model. Our Text-to-Image Model Editing method, TIME for short, receives a pair of inputs: a “source” under-specified prompt for which the model makes an implicit assumption (e.g., “a pack of roses”), and a “destination” prompt that describes the same setting, but with a specified desired attribute (e.g., “a pack of blue roses”). TIME then updates the model’s cross-attention layers, as these layers assign visual meaning to textual tokens. We edit the projection matrices in these layers such that the source prompt is projected close to the destination prompt. Our method is highly efficient, as it modifies a mere 2.2% of the model’s parameters in under one second. To evaluate model editing approaches, we introduce TIMED (TIME Dataset), containing 147 source and destination prompt pairs from various domains. Our experiments (using Stable Diffusion) show that TIME is successful in model editing, generalizes well for related prompts unseen during editing, and imposes minimal effect on unrelated generations.
|StableDiffusionModelEditingPipeline||Text-to-Image Model Editing||🤗 Space)|
This pipeline enables editing the diffusion model weights, such that its assumptions on a given concept are changed. The resulting change is expected to take effect in all prompt generations pertaining to the edited concept.
import torch from diffusers import StableDiffusionModelEditingPipeline model_ckpt = "CompVis/stable-diffusion-v1-4" pipe = StableDiffusionModelEditingPipeline.from_pretrained(model_ckpt) pipe = pipe.to("cuda") source_prompt = "A pack of roses" destination_prompt = "A pack of blue roses" pipe.edit_model(source_prompt, destination_prompt) prompt = "A field of roses" image = pipe(prompt).images image.save("field_of_roses.png")
class diffusers.StableDiffusionModelEditingPipeline< source >
( vae: AutoencoderKL text_encoder: CLIPTextModel tokenizer: CLIPTokenizer unet: UNet2DConditionModel scheduler: SchedulerMixin safety_checker: StableDiffusionSafetyChecker feature_extractor: CLIPFeatureExtractor requires_safety_checker: bool = True with_to_k: bool = True with_augs: list = ['A photo of ', 'An image of ', 'A picture of '] )
- vae (AutoencoderKL) — Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
CLIPTextModel) — Frozen text-encoder. Stable Diffusion uses the text portion of CLIP, specifically the clip-vit-large-patch14 variant.
CLIPTokenizer) — Tokenizer of class CLIPTokenizer.
- unet (UNet2DConditionModel) — Conditional U-Net architecture to denoise the encoded image latents.
scheduler (SchedulerMixin) —
A scheduler to be used in combination with
unetto denoise the encoded image latents.
StableDiffusionSafetyChecker) — Classification module that estimates whether generated images could be considered offensive or harmful. Please, refer to the model card for details.
CLIPFeatureExtractor) — Model that extracts features from generated images to be used as inputs for the
bool) — Whether to edit the key projection matrices along wiht the value projection matrices.
list) — Textual augmentations to apply while editing the text-to-image model. Set to  for no augmentations.
Pipeline for text-to-image model editing using “Editing Implicit Assumptions in Text-to-Image Diffusion Models”.
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.).
__call__< source >
prompt: typing.Union[str, typing.List[str]] = None
height: typing.Optional[int] = None
width: typing.Optional[int] = None
num_inference_steps: int = 50
guidance_scale: float = 7.5
negative_prompt: typing.Union[str, typing.List[str], NoneType] = None
num_images_per_prompt: typing.Optional[int] = 1
eta: float = 0.0
generator: typing.Union[torch._C.Generator, typing.List[torch._C.Generator], NoneType] = None
latents: typing.Optional[torch.FloatTensor] = None
prompt_embeds: typing.Optional[torch.FloatTensor] = None
negative_prompt_embeds: typing.Optional[torch.FloatTensor] = None
output_type: typing.Optional[str] = 'pil'
return_dict: bool = True
callback: typing.Union[typing.Callable[[int, int, torch.FloatTensor], NoneType], NoneType] = None
callback_steps: int = 1
cross_attention_kwargs: typing.Union[typing.Dict[str, typing.Any], NoneType] = None
List[str], optional) — The prompt or prompts to guide the image generation. If not defined, one has to pass
int, optional, defaults to self.unet.config.sample_size * self.vae_scale_factor) — The height in pixels of the generated image.
int, optional, defaults to self.unet.config.sample_size * self.vae_scale_factor) — The width in pixels of the generated image.
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.
float, optional, defaults to 7.5) — Guidance scale as defined in Classifier-Free Diffusion Guidance.
guidance_scaleis defined as
wof equation 2. of Imagen Paper. 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.
List[str], optional) — The prompt or prompts not to guide the image generation. If not defined, one has to pass
negative_prompt_embedsinstead. Ignored when not using guidance (i.e., ignored if
guidance_scaleis less than
int, optional, defaults to 1) — The number of images to generate per prompt.
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.
List[torch.Generator], optional) — One or a list of torch generator(s) to make generation deterministic.
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
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
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
str, optional, defaults to
"pil") — The output format of the generate image. Choose between PIL:
bool, optional, defaults to
True) — Whether or not to return a StableDiffusionPipelineOutput instead of a plain tuple.
Callable, optional) — A function that will be called every
callback_stepssteps during inference. The function will be called with the following arguments:
callback(step: int, timestep: int, latents: torch.FloatTensor).
int, optional, defaults to 1) — The frequency at which the
callbackfunction will be called. If not specified, the callback will be called at every step.
dict, optional) — A kwargs dictionary that if specified is passed along to the
AttentionProcessoras defined under
return_dict is True, otherwise a
tuple. When returning a tuple, the first element is a list with the generated images, and the second element is a list of bool
s denoting whether the corresponding generated image likely represents "not-safe-for-work" (nsfw) content, according to the safety_checker`.
Function invoked when calling the pipeline for generation.
disable_vae_slicing< source >
Disable sliced VAE decoding. If
enable_vae_slicing was previously invoked, this method will go back to
computing decoding in one step.
edit_model< source >
( source_prompt: str destination_prompt: str lamb: float = 0.1 restart_params: bool = True )
str) — The source prompt containing the concept to be edited.
str) — The destination prompt. Must contain all words from source_prompt with additional ones to specify the target edit.
float, optional, defaults to 0.1) — The lambda parameter specifying the regularization intesity. Smaller values increase the editing power.
bool, optional, defaults to True) — Restart the model parameters to their pre-trained version before editing. This is done to avoid edit compounding. When it is False, edits accumulate.
Apply model editing via closed-form solution (see Eq. 5 in the TIME paper https://arxiv.org/abs/2303.08084)
enable_sequential_cpu_offload< source >
( gpu_id = 0 )
Offloads all models to CPU using accelerate, significantly reducing memory usage. When called, unet,
text_encoder, vae and safety checker have their state dicts saved to CPU and then are moved to a
torch.device('meta') and loaded to GPU only when their specific submodule has its forward
method called. Note that offloading happens on a submodule basis. Memory savings are higher than withenable_model_cpu_offload`, but performance is lower.
enable_vae_slicing< source >
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.