T2I-Adapter: Learning Adapters to Dig out More Controllable Ability for Text-to-Image Diffusion Models by Chong Mou, Xintao Wang, Liangbin Xie, Jian Zhang, Zhongang Qi, Ying Shan, Xiaohu Qie.
Using the pretrained models we can provide control images (for example, a depth map) to control Stable Diffusion text-to-image generation so that it follows the structure of the depth image and fills in the details.
The abstract of the paper is the following:
The incredible generative ability of large-scale text-to-image (T2I) models has demonstrated strong power of learning complex structures and meaningful semantics. However, relying solely on text prompts cannot fully take advantage of the knowledge learned by the model, especially when flexible and accurate structure control is needed. In this paper, we aim to “dig out” the capabilities that T2I models have implicitly learned, and then explicitly use them to control the generation more granularly. Specifically, we propose to learn simple and small T2I-Adapters to align internal knowledge in T2I models with external control signals, while freezing the original large T2I models. In this way, we can train various adapters according to different conditions, and achieve rich control and editing effects. Further, the proposed T2I-Adapters have attractive properties of practical value, such as composability and generalization ability. Extensive experiments demonstrate that our T2I-Adapter has promising generation quality and a wide range of applications.
Pipeline | Tasks | Demo |
---|---|---|
StableDiffusionAdapterPipeline | Text-to-Image Generation with T2I-Adapter Conditioning | - |
In the following we give a simple example of how to use a T2IAdapter checkpoint with Diffusers for inference. The inference pipeline is the same for all pipelines:
Let’s have a look at a simple example using the Color Adapter.
from diffusers.utils import load_image
image = load_image("https://huggingface.co/RzZ/sd-v1-4-adapter-color/resolve/main/color_ref.png")
Then we can create our color palette by simply resize it to 8 by 8 pixels then scale it back to original size.
from PIL import Image
color_palette = image.resize((8, 8))
color_palette = color_palette.resize((512, 512), resample=Image.Resampling.NEAREST)
Let’s take a look at the processed image.
After we having color_palette
in hand, we can create the StableDiffusionAdapterPipeline with pretrained checkpoint.
import torch
from diffusers import StableDiffusionAdapterPipeline, T2IAdapter
adapter = T2IAdapter.from_pretrained("RzZ/sd-v1-4-adapter-color")
pipe = StableDiffusionAdapterPipeline.from_pretrained(
"CompVis/stable-diffusion-v1-4",
adapter=adapter,
torch_dtype=torch.float16,
)
pipe.to("cuda")
And finally we feed the data to the pipelien and wait for the result!
# fix the random seed, so you will get the same result as the example
generator = torch.manual_seed(7)
out_image = pipe(
["At night, glowing cubes in front of the beach"],
image=[color_palette],
generator=generator,
).images[0]
This should take only few seconds on GPU (depending on hardware). The output image then looks as follows:
Note: To see how to run all other Adapter checkpoints, please have a look at T2I-Adapter with Stable Diffusion 1.4
Adapter requires a control image in addition to the text-to-image prompt. Each pretrained model is trained using a different conditioning method that requires different images for conditioning the generated outputs. For example, Canny edge conditioning requires the control image to be the output of a Canny filter, while depth conditioning requires the control image to be a depth map. See the overview and image examples below to know more.
All official checkpoints can be found under the authors’ namespace TencentARC/T2I-Adapter.
Model Name | Control Image Overview | Control Image Example | Generated Image Example |
---|---|---|---|
RzZ/sd-v1.4-adapter-color Trained with spatial color palette |
A image with 8x8 color palette. | ![]() |
![]() |
RzZ/sd-v1.4-adapter-canny Trained with canny edge detection |
A monochrome image with white edges on a black background. | ![]() |
![]() |
RzZ/sd-v1.4-adapter-sketch Trained with PidiNet edge detection |
A hand-drawn monochrome image with white outlines on a black background. | ![]() |
![]() |
RzZ/sd-v1.4-adapter-depth Trained with Midas depth estimation |
A grayscale image with black representing deep areas and white representing shallow areas. | ![]() |
![]() |
RzZ/sd-v1.4-adapter-openpose Trained with OpenPose bone image |
A OpenPose bone image. | ![]() |
![]() |
RzZ/sd-v1.4-adapter-keypose Trained with mmpose skeleton image |
A mmpose skeleton image. | ![]() |
![]() |
RzZ/sd-v1.4-adapter-seg Trained with semantic segmentation |
An custom segmentation protocol image. | ![]() |
![]() |
StableDiffusionAdapterPipeline also support using multiple type of control image at once with combination with MultiAdapter. Here is a example of using keypose adapter for character posture control and depth adapter for outlining background.
Just like the previous example, we will first prepare the control image for inference. One big difference when using MultiAdapter is that the control image we will send to pipeline is
combined from multiple images. In this example we stack two 3 channels RGB image(cond_keypose
, cond_depth
) together to create a 6 channels image tensor(cond
).
import torch
from PIL import Image
from diffusers.utils import load_image
cond_keypose = load_image(
"https://huggingface.co/RzZ/sd-v1-4-adapter-keypose-depth/resolve/main/sample_input_keypose.png"
)
cond_depth = load_image("https://huggingface.co/RzZ/sd-v1-4-adapter-keypose-depth/resolve/main/sample_input_depth.png")
cond = [[cond_keypose, cond_depth]]
prompt = ["A man waling in an office room with nice view"]
Two control image should look like follows:
Now we can using from_adapters
method combine keypose and depth adapter into one, then pass our newly created MultiAdapter to
StableDiffusionAdapterPipeline. You can also play around the value of adapter_conditioning_scale
to balance the control between adapters.
from diffusers import StableDiffusionAdapterPipeline, MultiAdapter
adapters = MultiAdapter(
[
T2IAdapter.from_pretrained("RzZ/sd-v1-4-adapter-keypose"),
T2IAdapter.from_pretrained("RzZ/sd-v1-4-adapter-depth"),
]
)
adapters = adapters.to(torch.float16)
pipe = StableDiffusionAdapterPipeline.from_pretrained(
"CompVis/stable-diffusion-v1-4",
torch_dtype=torch.float16,
adapter=adapters,
)
images = pipe(prompt, cond, adapter_conditioning_scale=[0.8, 0.8])
After prompt and image is processed by pipeline we should get the result looks like:
T2I-Adapter is similar to ControlNet. However, T2i-Adapter uses a smaller auxiliary network which is only run once for the entire diffusion process. T2I-Adapter performs slightly worse than ControlNet. However, T2I-Adapter is cheaper to run and is cheaper to run multiple auxiliary networks.
( vae: AutoencoderKL text_encoder: CLIPTextModel tokenizer: CLIPTokenizer unet: UNet2DConditionModel adapter: typing.Union[diffusers.models.adapter.T2IAdapter, diffusers.models.adapter.MultiAdapter, typing.List[diffusers.models.adapter.T2IAdapter]] scheduler: KarrasDiffusionSchedulers safety_checker: StableDiffusionSafetyChecker feature_extractor: CLIPFeatureExtractor adapter_weights: typing.Optional[typing.List[float]] = None requires_safety_checker: bool = True )
Parameters
List[T2IAdapter]
) —
Provides additional conditioning to the unet during the denoising process. If you set multiple Adapter as a
list, the outputs from each Adapter are added together to create one combined additional conditioning.
List[float]
, optional, defaults to None) —
List of floats representing the weight which will be multiply to each adapter’s output before adding them
together.
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
to denoise the encoded image latents. Can be one of
DDIMScheduler, LMSDiscreteScheduler, or PNDMScheduler.
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 safety_checker
.
Pipeline for text-to-image generation using Stable Diffusion augmented with T2I-Adapter
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.)
(
prompt: typing.Union[str, typing.List[str]] = None
image: typing.Union[torch.Tensor, PIL.Image.Image, typing.List[PIL.Image.Image]] = 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
adapter_conditioning_scale: typing.Union[float, typing.List[float]] = 1.0
)
→
StableDiffusionPipelineOutput or tuple
Parameters
str
or List[str]
, optional) —
The prompt or prompts to guide the image generation. If not defined, one has to pass prompt_embeds
.
instead.
torch.FloatTensor
, PIL.Image.Image
, List[torch.FloatTensor]
or List[PIL.Image.Image]
or List[List[PIL.Image.Image]]
) —
The Adapter input condition. Adapter uses this input condition to generate guidance to Unet. If the
type is specified as Torch.FloatTensor
, it is passed to Adapter as is. PIL.Image.Image` can also be
accepted as an image. The control image is automatically resized to fit the output image.
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_scale
is defined as w
of 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.
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. 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
).
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.
torch.Generator
or 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 generator
.
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.
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.
str
, optional, defaults to "pil"
) —
The output format of the generate image. Choose between
PIL: PIL.Image.Image
or np.array
.
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_steps
steps 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 callback
function 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 AttnProcessor
as defined under
self.processor
in
diffusers.cross_attention.
float
or List[float]
, optional, defaults to 1.0) —
The outputs of the adapter are multiplied by adapter_conditioning_scale
before they are added to the
residual in the original unet. If multiple adapters are specified in init, you can set the
corresponding scale as a list.
Returns
StableDiffusionPipelineOutput or tuple
StableDiffusionPipelineOutput if 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
bools 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.
Examples:
>>> from PIL import Image
>>> from diffusers.utils import load_image
>>> image = load_image("https://huggingface.co/RzZ/sd-v1-4-adapter-color/resolve/main/color_ref.png")
>>> color_palette = image.resize((8, 8))
>>> color_palette = color_palette.resize((512, 512), resample=Image.Resampling.NEAREST)
>>> import torch
>>> from diffusers import StableDiffusionAdapterPipeline, T2IAdapter
>>> adapter = T2IAdapter.from_pretrained("RzZ/sd-v1-4-adapter-color")
>>> pipe = StableDiffusionAdapterPipeline.from_pretrained(
... "CompVis/stable-diffusion-v1-4",
... adapter=adapter,
... torch_dtype=torch.float16,
... )
>>> pipe.to("cuda")
>>> out_image = pipe(
... "At night, glowing cubes in front of the beach",
... image=color_palette,
... generator=generator,
... ).images[0]
( slice_size: typing.Union[str, int, NoneType] = 'auto' )
Parameters
str
or int
, optional, defaults to "auto"
) —
When "auto"
, halves the input to the attention heads, so attention will be computed in two steps. If
"max"
, maximum amount of memory will be saved by running only one slice at a time. If a number is
provided, uses as many slices as attention_head_dim // slice_size
. In this case, attention_head_dim
must be a multiple of slice_size
.
Enable sliced attention computation.
When this option is enabled, the attention module will split the input tensor in slices, to compute attention in several steps. This is useful to save some memory in exchange for a small speed decrease.
Disable sliced attention computation. If enable_attention_slicing
was previously invoked, this method will go
back to computing attention in one step.
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.
Disable sliced VAE decoding. If enable_vae_slicing
was previously invoked, this method will go back to
computing decoding in one step.
( attention_op: typing.Optional[typing.Callable] = None )
Parameters
Callable
, optional) —
Override the default None
operator for use as op
argument to the
memory_efficient_attention()
function of xFormers.
Enable memory efficient attention as implemented in xformers.
When this option is enabled, you should observe lower GPU memory usage and a potential speed up at inference time. Speed up at training time is not guaranteed.
Warning: When Memory Efficient Attention and Sliced attention are both enabled, the Memory Efficient Attention is used.
Examples:
>>> import torch
>>> from diffusers import DiffusionPipeline
>>> from xformers.ops import MemoryEfficientAttentionFlashAttentionOp
>>> pipe = DiffusionPipeline.from_pretrained("stabilityai/stable-diffusion-2-1", torch_dtype=torch.float16)
>>> pipe = pipe.to("cuda")
>>> pipe.enable_xformers_memory_efficient_attention(attention_op=MemoryEfficientAttentionFlashAttentionOp)
>>> # Workaround for not accepting attention shape using VAE for Flash Attention
>>> pipe.vae.enable_xformers_memory_efficient_attention(attention_op=None)
Disable memory efficient attention as implemented in xformers.
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
.
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
forwardmethod called. Note that offloading happens on a submodule basis. Memory savings are higher than with
enable_model_cpu_offload`, but performance is lower.