bi\ d dl Z d dlmZmZmZmZmZ d dlZd dl Z d dl
m Z ddl m
Z
mZ ddlmZ ddlmZ ddlmZ dd lmZmZmZmZ dd
lmZ d d l m Z m Z e rd
dl m Z e rdZ!ndZ! ejD e# Z$dZ% ddee& deee'e jP f deee& deee) fdZ* G d de Z+y) N)CallableDictListOptionalUnion)LlamaTokenizer )PipelineImageInputVaeImageProcessor)
AutoencoderKL)OmniGenTransformer2DModel) FlowMatchEulerDiscreteScheduler)is_torch_xla_availableis_torchvision_availableloggingreplace_example_docstring) randn_tensor )DiffusionPipelineImagePipelineOutput )OmniGenMultiModalProcessorTFa[
Examples:
```py
>>> import torch
>>> from diffusers import OmniGenPipeline
>>> pipe = OmniGenPipeline.from_pretrained("Shitao/OmniGen-v1-diffusers", torch_dtype=torch.bfloat16)
>>> pipe.to("cuda")
>>> prompt = "A cat holding a sign that says hello world"
>>> # Depending on the variant being used, the pipeline call will slightly vary.
>>> # Refer to the pipeline documentation for more details.
>>> image = pipe(prompt, num_inference_steps=50, guidance_scale=2.5).images[0]
>>> image.save("output.png")
```
num_inference_stepsdevice timestepssigmasc |
| t d |dt t j | j j
j
v }|st d| j d | j d
||d| | j }t | }||fS |dt t j | j j
j
v }|st d| j d | j d
||d| | j }t | }||fS | j |fd |i| | j }||fS ) a
Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles
custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`.
Args:
scheduler (`SchedulerMixin`):
The scheduler to get timesteps from.
num_inference_steps (`int`):
The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps`
must be `None`.
device (`str` or `torch.device`, *optional*):
The device to which the timesteps should be moved to. If `None`, the timesteps are not moved.
timesteps (`List[int]`, *optional*):
Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed,
`num_inference_steps` and `sigmas` must be `None`.
sigmas (`List[float]`, *optional*):
Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed,
`num_inference_steps` and `timesteps` must be `None`.
Returns:
`Tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the
second element is the number of inference steps.
zYOnly one of `timesteps` or `sigmas` can be passed. Please choose one to set custom valuesr z The current scheduler class zx's `set_timesteps` does not support custom timestep schedules. Please check whether you are using the correct scheduler.)r r r zv's `set_timesteps` does not support custom sigmas schedules. Please check whether you are using the correct scheduler.)r r r )
ValueErrorsetinspect signature
set_timesteps
parameterskeys __class__r len) schedulerr r r r kwargsaccepts_timesteps
accept_sigmass g/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/diffusers/pipelines/omnigen/pipeline_omnigen.pyretrieve_timestepsr- < s > !3tuu '3w/@/@AXAX/Y/d/d/i/i/k+l l .y/B/B.C Da b
M)FMfM '' !)n ) ))
C(9(9):Q:Q(R(](](b(b(d$ee
.y/B/B.C D_ `
GvfGG '' !)n ) )) 3MFMfM '' ) )) c ' f e Zd ZdZdZg ZdgZdedede de
f fd Z d0d
e e
j d ee
j d ee
j" fd
Z d1dZd Zd Zd Zd Z d1dZed Zed Zed Z e
j: e e d d d ddd dddd d d d d d dgfd e e!e e! f d e e"e e" f d!ee# d"ee# d#e#d$e#d%e e# d&e$d'e$d(e%d)ee# d*ee e
jL e e
jL f dee
j d+ee! d,e%d-ee'e#e#e(gd f d.e e! f"d/ Z) xZ*S )2OmniGenPipelinea
The OmniGen pipeline for multimodal-to-image generation.
Reference: https://huggingface.co/papers/2409.11340
Args:
transformer ([`OmniGenTransformer2DModel`]):
Autoregressive Transformer architecture for OmniGen.
scheduler ([`FlowMatchEulerDiscreteScheduler`]):
A scheduler to be used in combination with `transformer` to denoise the encoded image latents.
vae ([`AutoencoderKL`]):
Variational Auto-Encoder (VAE) Model to encode and decode images to and from latent representations.
tokenizer (`LlamaTokenizer`):
Text tokenizer of class.
[LlamaTokenizer](https://huggingface.co/docs/transformers/main/model_doc/llama#transformers.LlamaTokenizer).
ztransformer->vaelatents transformerr( vae tokenizerc t | | j |||| t | dd /dt | j
j j dz
z nd| _ t | j dz | _
t |d | _ t | d r"| j | j j nd
| _ d | _ y ) N)r3 r4 r2 r( r3 r r )vae_scale_factor max_image_sizer4 i )super__init__register_modulesgetattrr' r3 configblock_out_channelsr7 r image_processorr multimodal_processorhasattrr4 model_max_lengthtokenizer_max_lengthdefault_sample_size)selfr2 r( r3 r4 r& s r, r= zOmniGenPipeline.__init__ s #
CJ$PUW[B\BhA#dhhoo889A= >no
1$BWBWZ[B[ \ $>yY]$^!/6t[/IdnnNhDNN + +nt
! $' r. Ninput_pixel_valuesr dtypec n |xs | j }|xs | j j }g }|D ] }| j j |j || j
j
j | j j j }|j | |S )z
get the continue embedding of input images by VAE
Args:
input_pixel_values: normalized pixel of input images
device:
Returns: torch.Tensor
) _execution_devicer3 rJ encodeto latent_distsamplemul_r@ scaling_factorappend)rH rI r rJ input_img_latentsimgs r, encode_input_imagesz#OmniGenPipeline.encode_input_images s 1411' % *C((//#&&"78DDKKMRRSWS[S[SbSbSqSqrC $ $S ) * ! r. c
|t | t k7 r$t dt dt | d t t | D ]J | t fdt t | D r4t d d| d | j dz z d k7 s| j dz z d k7 r,t
j
d
j dz d | d | d
|r||d t d |Mt fd|D s8t d j d|D cg c] }| j vs| c} y y c c}w )NzThe number of prompts: z, does not match the number of input images: .c 3 : K | ] }d |dz d v yw)z
![]()
<|image_r z|>Nr ).0kiprompts r, z/OmniGenPipeline.check_inputs.. s( qPQ q1ugX >&) K q zprompt `z9` doesn't have enough placeholders for the input images ``r r z-`height` and `width` have to be divisible by z but are z and z(. Dimensions will be resized accordinglyz`use_input_image_size_as_output` is set to True, but no input image was found. If you are performing a text-to-image task, please set it to False.c 3 : K | ] }|j v y wN)_callback_tensor_inputs)rZ r[ rH s r, r^ z/OmniGenPipeline.check_inputs.. s# F
23A-- -F
r_ z2`callback_on_step_end_tensor_inputs` has to be in z , but found )r' r rangeallr7 loggerwarningrc ) rH r] input_imagesheightwidth use_input_image_size_as_output"callback_on_step_end_tensor_inputsr[ r\ s `` @r, check_inputsz OmniGenPipeline.check_inputs s #< CK/ -c&k]:fgjkwgxfyyz{ 3|,-
?. qUZ[^_klm_n[oUp qq ( &vayk1jkwxykzj{{| }
T**Q. /1 4AVAVYZAZ8[_`8` NN?@U@UXY@Y?ZZcdjckkpqvpw x` a
*#|A'> i . 9# F
7YF
C
DTEaEaDbbn |^ pHvw bc ko kG kG bGpq pH oI J
C
9 pHs 9E
E c 8 | j j y)z
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.
N)r3 enable_slicingrH s r, enable_vae_slicingz"OmniGenPipeline.enable_vae_slicing
!r. c 8 | j j y)z
Disable sliced VAE decoding. If `enable_vae_slicing` was previously enabled, this method will go back to
computing decoding in one step.
N)r3 disable_slicingrp s r, disable_vae_slicingz#OmniGenPipeline.disable_vae_slicing s
"r. c 8 | j j y)a
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.
N)r3
enable_tilingrp s r, enable_vae_tilingz!OmniGenPipeline.enable_vae_tiling s
r. c 8 | j j y)z
Disable tiled VAE decoding. If `enable_vae_tiling` was previously enabled, this method will go back to
computing decoding in one step.
N)r3 disable_tilingrp s r, disable_vae_tilingz"OmniGenPipeline.disable_vae_tiling rr r. c $ ||j || S ||t | | j z t | | j z f} t |t r)t | |k7 rt
dt | d| d t | ||| }|S )N)r rJ z/You have passed a list of generators of length z+, but requested an effective batch size of z@. Make sure the batch size matches the length of the generators.) generatorr rJ )rN intr7
isinstancelistr' r r )
rH
batch_sizenum_channels_latentsri rj rJ r r} r1 shapes
r, prepare_latentsz OmniGenPipeline.prepare_latents s ::V5:9 9
K400 0 J$// /
i &3y>Z+G A#i.AQ R &<'gi
u &PUVr. c | j S rb )_guidance_scalerp s r, guidance_scalez OmniGenPipeline.guidance_scale& s ###r. c | j S rb )_num_timestepsrp s r,
num_timestepsz OmniGenPipeline.num_timesteps* s """r. c | j S rb )
_interruptrp s r, interruptzOmniGenPipeline.interrupt. s r. 2 r8 g @g?Fr pilTr] rh ri rj r max_input_image_sizer r img_guidance_scalerk num_images_per_promptr} output_type return_dictcallback_on_step_endrl c |xs | j | j z }|xs | j | j z }|dnd}|dnd}t |t r|g}|g}| j |||||
| || _ d| _ t | }| j }|| j j k7 r | j j | | j ||||||
| }|d j | |d < |d
j | |d
< |d j | |d < | j |d |
}t j dd|dz d| }t! | j" |||| \ }}t | | _ | j& j( }|
r|d d j* dd \ }}| j& j, j. }| j1 || z |||t2 j4 || |
}
| j7 | 5 }t9 | D ]W \ } } t3 j: |
g|dz z } | j | } | j= | j* d } | j' | | |d ||d |d
|d d d } |dk( r=t3 j> | t | dz d \ }!}"}#|"| |#|"z
z z ||!|#z
z z } n2t3 j> | t | dz d \ }!}"|"||!|"z
z z } | j" jA | | |
d d }
|6i }$|D ] }%tC |% |$|%< || | | |$ }&|&jE d|
}
|jG Z ddd |dk( s|
j | jH j( }
|
| jH j, jJ z }
| jH jM |
d d }'| jN jQ |'| }'n|
}'| jS |s|'fS tU |' S # 1 sw Y xY w)a
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 the input includes images, need to add
placeholders `![]()
<|image_i|>` in the prompt to indicate the position of the i-th images.
input_images (`PipelineImageInput` or `List[PipelineImageInput]`, *optional*):
The list of input images. We will replace the "<|image_i|>" in prompt with the i-th image in list.
height (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
The height in pixels of the generated image. This is set to 1024 by default for the best results.
width (`int`, *optional*, defaults to self.unet.config.sample_size * self.vae_scale_factor):
The width in pixels of the generated image. This is set to 1024 by default for the best results.
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.
max_input_image_size (`int`, *optional*, defaults to 1024):
the maximum size of input image, which will be used to crop the input image to the maximum size
timesteps (`List[int]`, *optional*):
Custom timesteps to use for the denoising process with schedulers which support a `timesteps` argument
in their `set_timesteps` method. If not defined, the default behavior when `num_inference_steps` is
passed will be used. Must be in descending order.
guidance_scale (`float`, *optional*, defaults to 2.5):
Guidance scale as defined in [Classifier-Free Diffusion
Guidance](https://huggingface.co/papers/2207.12598). `guidance_scale` is defined as `w` of equation 2.
of [Imagen Paper](https://huggingface.co/papers/2205.11487). 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.
img_guidance_scale (`float`, *optional*, defaults to 1.6):
Defined as equation 3 in [Instrucpix2pix](https://huggingface.co/papers/2211.09800).
use_input_image_size_as_output (bool, defaults to False):
whether to use the input image size as the output image size, which can be used for single-image input,
e.g., image editing task
num_images_per_prompt (`int`, *optional*, defaults to 1):
The number of images to generate per prompt.
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.Tensor`, *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`.
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.flux.FluxPipelineOutput`] instead of a plain tuple.
callback_on_step_end (`Callable`, *optional*):
A function that calls at the end of each denoising steps during the inference. The function is called
with the following arguments: `callback_on_step_end(self: DiffusionPipeline, step: int, timestep: int,
callback_kwargs: Dict)`. `callback_kwargs` will include a list of all tensors as specified by
`callback_on_step_end_tensor_inputs`.
callback_on_step_end_tensor_inputs (`List`, *optional*):
The list of tensor inputs for the `callback_on_step_end` function. The tensors specified in the list
will be passed as `callback_kwargs` argument. You will only be able to include variables listed in the
`._callback_tensor_inputs` attribute of your pipeline class.
Examples:
Returns: [`~pipelines.ImagePipelineOutput`] or `tuple`:
If `return_dict` is `True`, [`~pipelines.ImagePipelineOutput`] is returned, otherwise a `tuple` is returned
where the first element is a list with the generated images.
Nr r TF)rl r9 )ri rj use_img_cfgrk r input_idsattention_mask position_idsrI )r r )r )totalinput_image_sizes)
hidden_statestimestepr rT r r r r r )dim)r r1 latent)r )images)+rG r7 r strrm r r r' rL rC r: reset_max_image_sizerN rV nplinspacer- r( r r2 rJ r r@ in_channelsr torchfloat32 progress_bar enumeratecatexpandsplitsteplocalspopupdater3 rR decoderB postprocessmaybe_free_model_hooksr )(rH r] rh ri rj r r r r r rk r r} r1 r r r rl num_cfgr r r processed_datarT r transformer_dtypelatent_channelsr r\ tlatent_model_inputr
noise_predconduncondimg_condcallback_kwargsr[ callback_outputsimages( r, __call__zOmniGenPipeline.__call__2 s l K433d6K6KKI11D4I4II#/!Q * 6dE fc " XF (>L
*/Q
. [
'' 4#<#<#K#K K % % : :J^ : _ 22 #+I"7 3
'5[&A&D&DV&L{#+9:J+K+N+Nv+V'()7)G)J)J6)R~& ! 4 4^DX5Ybh 4 i Q#6#:; >>GHHOOGO?BE((44U 4TEE
##%8O"%00m& & & &s
E'P44P=)NNrb )+__name__
__module__ __qualname____doc__model_cpu_offload_seq_optional_componentsrc r
r r r r= r r Tensorr r rJ rV rm rq ru rx r{ r propertyr r r no_gradr EXAMPLE_DOC_STRINGr r r
r~ floatbool Generatorr r r
__classcell__)r& s @r, r0 r0 w s~ " / (k'.' 3' '
" '> *.'+ ! .! &! $ !< ,0%N"#!"" > $ $ # # U]]_ 12 MQ $ ##%$( # #$'/4/0MQ*.%* KO9B %N1c49n%N1 .5G0H HIN1
N1
} N1 !
N1 "N1 9N1 N1 "N1 )-N1 (}N1 E%//43H"H IJN1 %,,' N1 c] N1 !N1" 'xc40@$0F'G H#N1$ -1I%N1 3 N1r. r0 )NNNN),r! typingr r r r r numpyr r transformersr rB r
r models.autoencodersr models.transformersr
schedulersr utilsr r r r utils.torch_utilsr pipeline_utilsr r processor_omnigenr
XLA_AVAILABLE
get_loggerr rf r r~ r r r r- r0 r r. r, r s 8 8 ' D 0 < 9 i i - C =MM H % * *.15%)$( 8*!# 8*
U3 ,- .8* S " 8*
T%[ ! 8*vK1K1r.