🤗 Diffusers provides pretrained diffusion models across multiple modalities, such as vision and audio, and serves
as a modular toolbox for inference and training of diffusion models.
More precisely, 🤗 Diffusers offers:
- State-of-the-art diffusion pipelines that can be run in inference with just a couple of lines of code (see [src/diffusers/pipelines](https://github.com/huggingface/diffusers/tree/main/src/diffusers/pipelines)).
- Various noise schedulers that can be used interchangeably for the prefered speed vs. quality trade-off in inference (see [src/diffusers/schedulers](https://github.com/huggingface/diffusers/tree/main/src/diffusers/schedulers)).
- Multiple types of models, such as UNet, that can be used as building blocks in an end-to-end diffusion system (see [src/diffusers/models](https://github.com/huggingface/diffusers/tree/main/src/diffusers/models)).
- Training examples to show how to train the most popular diffusion models (see [examples](https://github.com/huggingface/diffusers/tree/main/examples)).
## Definitions
**Models**: Single neural network that models p_θ(x_t-1|x_t) and is trained to “denoise” to image
*Examples: UNet, Conditioned UNet, 3D UNet, Transformer UNet*
**Models**: Neural network that models $p_\theta(\mathbf{x}_{t-1}|\mathbf{x}_t)$ (see image below) and is trained end-to-end to *denoise* a noisy input to an image.
*Examples*: UNet, Conditioned UNet, 3D UNet, Transformer UNet
**Schedulers**: Algorithm to compute previous image according to alpha, beta schedule and to sample noise. Should be used for both *training* and *inference*.
*Example: Gaussian DDPM, DDIM, PMLS, DEIN*
**Schedulers**: Algorithm class for both **inference** and **training**.
The class provides functionality to compute previous image according to alpha, beta schedule as well as predict noise for training.
- Readability and clarity is prefered over highly optimized code. A strong importance is put on providing readable, intuitive and elementary code design. *E.g.*, the provided [schedulers](https://github.com/huggingface/diffusers/tree/main/src/diffusers/schedulers) are separated from the provided [models](https://github.com/huggingface/diffusers/tree/main/src/diffusers/models) and provide well-commented code that can be read alongside the original paper.
- Diffusers is **modality independent** and focusses on providing pretrained models and tools to build systems that generate **continous outputs**, *e.g.* vision and audio.
- Diffusion models and schedulers are provided as consise, elementary building blocks whereas diffusion pipelines are a collection of end-to-end diffusion systems that can be used out-of-the-box, should stay as close as possible to their original implementation and can include components of other library, such as text-encoders. Examples for diffusion pipelines are [Glide](https://github.com/openai/glide-text2im) and [Latent Diffusion](https://github.com/CompVis/latent-diffusion).
pip install diffusers # should install diffusers 0.0.4
```
### 1. `diffusers` as a central modular diffusion and sampler library
### 1. `diffusers` as a toolbox for schedulers and models
`diffusers` is more modularized than `transformers`. The idea is that researchers and engineers can use only parts of the library easily for the own use cases.
It could become a central place for all kinds of models, schedulers, training utils and processors that one can mix and match for one's own use case.
Both models and schedulers should be load- and saveable from the Hub.
For more examples see [schedulers](https://github.com/huggingface/diffusers/tree/main/src/diffusers/schedulers) and [models](https://github.com/huggingface/diffusers/tree/main/src/diffusers/models)
#### **Example for [DDPM](https://arxiv.org/abs/2006.11239):**
### 2. `diffusers` as a collection of most important Diffusion systems (GLIDE, Dalle, ...)
`models` directory in repository hosts the complete code necessary for running a diffusion system as well as to train it. A `DiffusionPipeline` class allows to easily run the diffusion model in inference:
### 2. `diffusers` as a collection of popular Diffusion systems (GLIDE, Dalle, ...)
#### **Example image generation with DDPM**
For more examples see [pipelines](https://github.com/huggingface/diffusers/tree/main/src/diffusers/pipelines).
- Models: Neural network that models $p_\theta(\mathbf{x}_{t-1}|\mathbf{x}_t)$ (see image below) and is trained end-to-end to denoise a noisy input to an image. Examples: UNet, Conditioned UNet, 3D UNet, Transformer UNet
- Pipelines are a collection of end-to-end diffusion systems that can be used out-of-the-box
- Pipelines should stay as close as possible to their original implementation
- Pipelines can include components of other library, such as text-encoders.
## API
TODO(Patrick, Anton, Suraj)
## Examples
- DDPM for unconditional image generation in [pipeline_ddpm](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_ddpm.py).
- DDIM for unconditional image generation in [pipeline_ddim](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_ddim.py).
- PNDM for unconditional image generation in [pipeline_pndm](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_pndm.py).
- Latent diffusion for text to image generation / conditional image generation in [pipeline_ddpm](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_bddm.py).
- Glide for text to image generation / conditional image generation in [pipeline_ddpm](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_bddm.py).
- BDDM for spectrogram-to-sound vocoding in [pipeline_ddpm](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_bddm.py).
- Grad-TTS for text to audio generation / conditional audio generation in [pipeline_ddpm](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_bddm.py).
- Pipelines are a collection of end-to-end diffusion systems that can be used out-of-the-box
- Pipelines should stay as close as possible to their original implementation
- Pipelines can include components of other library, such as text-encoders.
## API
TODO(Patrick, Anton, Suraj)
## Examples
- DDPM for unconditional image generation in [pipeline_ddpm](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_ddpm.py).
- DDIM for unconditional image generation in [pipeline_ddim](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_ddim.py).
- PNDM for unconditional image generation in [pipeline_pndm](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_pndm.py).
- Latent diffusion for text to image generation / conditional image generation in [pipeline_latent_diffusion](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_latent_diffusion.py).
- Glide for text to image generation / conditional image generation in [pipeline_glide](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_glide.py).
- BDDM for spectrogram-to-sound vocoding in [pipeline_bddm](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_bddm.py).
- Grad-TTS for text to audio generation / conditional audio generation in [pipeline_grad_tts](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_grad_tts.py).
- Schedulers are the algorithms to use diffusion models in inference as well as for training. They include the noise schedules and define algorithm-specific diffusion steps.
- Schedulers can be used interchangable between diffusion models in inference to find the preferred tradef-off between speed and generation quality.
- Schedulers are available in numpy, but can easily be transformed into PyTorch.
## API
- Schedulers should provide one or more `def step(...)` functions that should be called iteratively to unroll the diffusion loop during
the forward pass.
- Schedulers should be framework-agonstic, but provide a simple functionality to convert the scheduler into a specific framework, such as PyTorch
with a `set_format(...)` method.
## Examples
- The DDPM scheduler was proposed in [Denoising Diffusion Probabilistic Models](https://arxiv.org/abs/2006.11239) and can be found in [scheduling_ddpm.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/schedulers/scheduling_ddpm.py). An example of how to use this scheduler can be found in [pipeline_ddpm.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_ddpm.py).
- The DDIM scheduler was proposed in [Denoising Diffusion Implicit Models](https://arxiv.org/abs/2010.02502) and can be found in [scheduling_ddim.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/schedulers/scheduling_ddim.py). An example of how to use this scheduler can be found in [pipeline_ddim.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_ddim.py).
- The PNMD scheduler was proposed in [Pseudo Numerical Methods for Diffusion Models on Manifolds](https://arxiv.org/abs/2202.09778) and can be found in [scheduling_pndm.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/schedulers/scheduling_pndm.py). An example of how to use this scheduler can be found in [pipeline_pndm.py](https://github.com/huggingface/diffusers/blob/main/src/diffusers/pipelines/pipeline_pndm.py).
