File size: 36,079 Bytes
4684d79 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 | # Copyright 2026 Krea AI and The HuggingFace Team. All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import torch
from diffusers.utils import logging
from diffusers.modular_pipelines.modular_pipeline import AutoPipelineBlocks, SequentialPipelineBlocks
from diffusers.modular_pipelines.modular_pipeline_utils import InputParam, InsertableDict, OutputParam
from .before_denoise import (
Krea2CreateMaskLatentsStep,
Krea2PrepareLatentsStep,
Krea2PrepareLatentsWithStrengthStep,
Krea2RoPEInputsStep,
Krea2SetTimestepsStep,
Krea2SetTimestepsWithStrengthStep,
)
from .decoders import (
Krea2AfterDenoiseStep,
Krea2DecoderStep,
Krea2InpaintProcessImagesOutputStep,
Krea2ProcessImagesOutputStep,
)
from .denoise import (
Krea2DenoiseStep,
Krea2InpaintDenoiseStep,
)
from .encoders import (
Krea2InpaintProcessImagesInputStep,
Krea2ProcessImagesInputStep,
Krea2TextEncoderStep,
Krea2VaeEncoderStep,
)
from .inputs import (
Krea2AdditionalInputsStep,
Krea2TextInputsStep,
)
logger = logging.get_logger(__name__)
# ====================
# 1. TEXT ENCODER
# ====================
# auto_docstring
class Krea2AutoTextEncoderStep(AutoPipelineBlocks):
"""
Text encoder step that encodes the text prompt into a text embedding. This is an auto pipeline block.
- `Krea2TextEncoderStep` (text_encoder) is used when `prompt` is provided.
- if `prompt` is not provided, step will be skipped.
Components:
text_encoder (`Qwen3VLModel`): The text encoder to use tokenizer (`Qwen2Tokenizer`): The tokenizer to use
guider (`ClassifierFreeGuidance`)
Inputs:
prompt (`str`, *optional*):
The prompt or prompts to guide image generation.
negative_prompt (`str`, *optional*):
The prompt or prompts not to guide the image generation.
max_sequence_length (`int`, *optional*, defaults to 512):
Maximum sequence length for prompt encoding.
Outputs:
prompt_embeds (`Tensor`):
The prompt embeddings.
prompt_embeds_mask (`Tensor`):
The encoder attention mask.
negative_prompt_embeds (`Tensor`):
The negative prompt embeddings.
negative_prompt_embeds_mask (`Tensor`):
The negative prompt embeddings mask.
"""
model_name = "krea2"
block_classes = [Krea2TextEncoderStep()]
block_names = ["text_encoder"]
block_trigger_inputs = ["prompt"]
@property
def description(self) -> str:
return (
"Text encoder step that encodes the text prompt into a text embedding. This is an auto pipeline block.\n"
" - `Krea2TextEncoderStep` (text_encoder) is used when `prompt` is provided.\n"
" - if `prompt` is not provided, step will be skipped."
)
# ====================
# 2. VAE ENCODER
# ====================
# auto_docstring
class Krea2InpaintVaeEncoderStep(SequentialPipelineBlocks):
"""
This step is used for processing image and mask inputs for inpainting tasks. It:
- Resizes the image to the target size, based on `height` and `width`.
- Processes and updates `image` and `mask_image`.
- Creates `image_latents`.
Components:
image_mask_processor (`InpaintProcessor`) vae (`AutoencoderKLQwenImage`)
Inputs:
mask_image (`Image`):
Mask image for inpainting.
image (`Image | list`):
Reference image(s) for denoising. Can be a single image or list of images.
height (`int`, *optional*):
The height in pixels of the generated image.
width (`int`, *optional*):
The width in pixels of the generated image.
padding_mask_crop (`int`, *optional*):
Padding for mask cropping in inpainting.
generator (`Generator`, *optional*):
Torch generator for deterministic generation.
Outputs:
processed_image (`Tensor`):
The processed image
processed_mask_image (`Tensor`):
The processed mask image
mask_overlay_kwargs (`dict`):
The kwargs for the postprocess step to apply the mask overlay
image_latents (`Tensor`):
The latent representation of the input image.
"""
model_name = "krea2"
block_classes = [Krea2InpaintProcessImagesInputStep(), Krea2VaeEncoderStep()]
block_names = ["preprocess", "encode"]
@property
def description(self) -> str:
return (
"This step is used for processing image and mask inputs for inpainting tasks. It:\n"
" - Resizes the image to the target size, based on `height` and `width`.\n"
" - Processes and updates `image` and `mask_image`.\n"
" - Creates `image_latents`."
)
# auto_docstring
class Krea2Img2ImgVaeEncoderStep(SequentialPipelineBlocks):
"""
Vae encoder step that preprocess and encode the image inputs into their latent representations.
Components:
image_processor (`VaeImageProcessor`) vae (`AutoencoderKLQwenImage`)
Inputs:
image (`Image | list`):
Reference image(s) for denoising. Can be a single image or list of images.
height (`int`, *optional*):
The height in pixels of the generated image.
width (`int`, *optional*):
The width in pixels of the generated image.
generator (`Generator`, *optional*):
Torch generator for deterministic generation.
Outputs:
processed_image (`Tensor`):
The processed image
image_latents (`Tensor`):
The latent representation of the input image.
"""
model_name = "krea2"
block_classes = [Krea2ProcessImagesInputStep(), Krea2VaeEncoderStep()]
block_names = ["preprocess", "encode"]
@property
def description(self) -> str:
return "Vae encoder step that preprocess and encode the image inputs into their latent representations."
class Krea2AutoVaeEncoderStep(AutoPipelineBlocks):
model_name = "krea2"
block_classes = [Krea2InpaintVaeEncoderStep, Krea2Img2ImgVaeEncoderStep]
block_names = ["inpaint", "img2img"]
block_trigger_inputs = ["mask_image", "image"]
@property
def description(self):
return (
"Vae encoder step that encode the image inputs into their latent representations.\n"
+ "This is an auto pipeline block.\n"
+ " - `Krea2InpaintVaeEncoderStep` (inpaint) is used when `mask_image` is provided.\n"
+ " - `Krea2Img2ImgVaeEncoderStep` (img2img) is used when `image` is provided.\n"
+ " - if `mask_image` or `image` is not provided, step will be skipped."
)
# ====================
# 3. DENOISE (input -> prepare_latents -> set_timesteps -> prepare_rope_inputs -> denoise -> after_denoise)
# ====================
# assemble input steps
# auto_docstring
class Krea2Img2ImgInputStep(SequentialPipelineBlocks):
"""
Input step that prepares the inputs for the img2img denoising step. It:
- make sure the text embeddings have consistent batch size as well as the additional inputs (`image_latents`).
- update height/width based `image_latents`, patchify `image_latents`.
Components:
pachifier (`Krea2Pachifier`)
Inputs:
num_images_per_prompt (`int`, *optional*, defaults to 1):
The number of images to generate per prompt.
prompt_embeds (`Tensor`):
text embeddings used to guide the image generation. Can be generated from text_encoder step.
prompt_embeds_mask (`Tensor`):
mask for the text embeddings. Can be generated from text_encoder step.
negative_prompt_embeds (`Tensor`, *optional*):
negative text embeddings used to guide the image generation. Can be generated from text_encoder step.
negative_prompt_embeds_mask (`Tensor`, *optional*):
mask for the negative text embeddings. Can be generated from text_encoder step.
height (`int`, *optional*):
The height in pixels of the generated image.
width (`int`, *optional*):
The width in pixels of the generated image.
image_latents (`Tensor`):
image latents used to guide the image generation. Can be generated from vae_encoder step.
Outputs:
batch_size (`int`):
The batch size of the prompt embeddings
dtype (`dtype`):
The data type of the prompt embeddings
prompt_embeds (`Tensor`):
The prompt embeddings. (batch-expanded)
prompt_embeds_mask (`Tensor`):
The encoder attention mask. (batch-expanded)
negative_prompt_embeds (`Tensor`):
The negative prompt embeddings. (batch-expanded)
negative_prompt_embeds_mask (`Tensor`):
The negative prompt embeddings mask. (batch-expanded)
image_height (`int`):
The image height calculated from the image latents dimension
image_width (`int`):
The image width calculated from the image latents dimension
height (`int`):
if not provided, updated to image height
width (`int`):
if not provided, updated to image width
image_latents (`Tensor`):
image latents used to guide the image generation. Can be generated from vae_encoder step. (patchified and
batch-expanded)
"""
model_name = "krea2"
block_classes = [Krea2TextInputsStep(), Krea2AdditionalInputsStep()]
block_names = ["text_inputs", "additional_inputs"]
@property
def description(self):
return (
"Input step that prepares the inputs for the img2img denoising step. It:\n"
" - make sure the text embeddings have consistent batch size as well as the additional inputs (`image_latents`).\n"
" - update height/width based `image_latents`, patchify `image_latents`."
)
# auto_docstring
class Krea2InpaintInputStep(SequentialPipelineBlocks):
"""
Input step that prepares the inputs for the inpainting denoising step. It:
- make sure the text embeddings have consistent batch size as well as the additional inputs (`image_latents` and
`processed_mask_image`).
- update height/width based `image_latents`, patchify `image_latents`.
Components:
pachifier (`Krea2Pachifier`)
Inputs:
num_images_per_prompt (`int`, *optional*, defaults to 1):
The number of images to generate per prompt.
prompt_embeds (`Tensor`):
text embeddings used to guide the image generation. Can be generated from text_encoder step.
prompt_embeds_mask (`Tensor`):
mask for the text embeddings. Can be generated from text_encoder step.
negative_prompt_embeds (`Tensor`, *optional*):
negative text embeddings used to guide the image generation. Can be generated from text_encoder step.
negative_prompt_embeds_mask (`Tensor`, *optional*):
mask for the negative text embeddings. Can be generated from text_encoder step.
height (`int`, *optional*):
The height in pixels of the generated image.
width (`int`, *optional*):
The width in pixels of the generated image.
image_latents (`Tensor`, *optional*):
image latents used to guide the image generation. Can be generated from vae_encoder step.
processed_mask_image (`Tensor`, *optional*):
The processed mask image
Outputs:
batch_size (`int`):
The batch size of the prompt embeddings
dtype (`dtype`):
The data type of the prompt embeddings
prompt_embeds (`Tensor`):
The prompt embeddings. (batch-expanded)
prompt_embeds_mask (`Tensor`):
The encoder attention mask. (batch-expanded)
negative_prompt_embeds (`Tensor`):
The negative prompt embeddings. (batch-expanded)
negative_prompt_embeds_mask (`Tensor`):
The negative prompt embeddings mask. (batch-expanded)
image_height (`int`):
The image height calculated from the image latents dimension
image_width (`int`):
The image width calculated from the image latents dimension
height (`int`):
if not provided, updated to image height
width (`int`):
if not provided, updated to image width
image_latents (`Tensor`):
image latents used to guide the image generation. Can be generated from vae_encoder step. (patchified and
batch-expanded)
processed_mask_image (`Tensor`):
The processed mask image (batch-expanded)
"""
model_name = "krea2"
block_classes = [
Krea2TextInputsStep(),
Krea2AdditionalInputsStep(
additional_batch_inputs=[
InputParam(name="processed_mask_image", type_hint=torch.Tensor, description="The processed mask image")
]
),
]
block_names = ["text_inputs", "additional_inputs"]
@property
def description(self):
return (
"Input step that prepares the inputs for the inpainting denoising step. It:\n"
" - make sure the text embeddings have consistent batch size as well as the additional inputs (`image_latents` and `processed_mask_image`).\n"
" - update height/width based `image_latents`, patchify `image_latents`."
)
# assemble prepare latents steps
# auto_docstring
class Krea2InpaintPrepareLatentsStep(SequentialPipelineBlocks):
"""
This step prepares the latents/image_latents and mask inputs for the inpainting denoising step. It:
- Add noise to the image latents to create the latents input for the denoiser.
- Create the patchified latents `mask` based on the processed mask image.
Components:
scheduler (`FlowMatchEulerDiscreteScheduler`) pachifier (`Krea2Pachifier`)
Inputs:
latents (`Tensor`):
The initial random noised, can be generated in prepare latent step.
image_latents (`Tensor`):
image latents used to guide the image generation. Can be generated from vae_encoder step. (Can be
generated from vae encoder and updated in input step.)
timesteps (`Tensor`):
The timesteps to use for the denoising process. Can be generated in set_timesteps step.
processed_mask_image (`Tensor`):
The processed mask to use for the inpainting process.
height (`int`):
The height in pixels of the generated image.
width (`int`):
The width in pixels of the generated image.
dtype (`dtype`, *optional*, defaults to torch.float32):
The dtype of the model inputs, can be generated in input step.
Outputs:
initial_noise (`Tensor`):
The initial random noised used for inpainting denoising.
latents (`Tensor`):
The scaled noisy latents to use for inpainting/image-to-image denoising.
mask (`Tensor`):
The mask to use for the inpainting process.
"""
model_name = "krea2"
block_classes = [Krea2PrepareLatentsWithStrengthStep(), Krea2CreateMaskLatentsStep()]
block_names = ["add_noise_to_latents", "create_mask_latents"]
@property
def description(self) -> str:
return (
"This step prepares the latents/image_latents and mask inputs for the inpainting denoising step. It:\n"
" - Add noise to the image latents to create the latents input for the denoiser.\n"
" - Create the patchified latents `mask` based on the processed mask image.\n"
)
# assemble denoising steps
# Krea 2 (text2image)
# auto_docstring
class Krea2CoreDenoiseStep(SequentialPipelineBlocks):
"""
step that denoise noise into image for text2image task. It includes the denoise loop, as well as prepare the inputs
(timesteps, latents, rope inputs etc.).
Components:
pachifier (`Krea2Pachifier`) scheduler (`FlowMatchEulerDiscreteScheduler`) guider (`ClassifierFreeGuidance`)
transformer (`Krea2Transformer2DModel`)
Inputs:
num_images_per_prompt (`int`, *optional*, defaults to 1):
The number of images to generate per prompt.
prompt_embeds (`Tensor`):
text embeddings used to guide the image generation. Can be generated from text_encoder step.
prompt_embeds_mask (`Tensor`):
mask for the text embeddings. Can be generated from text_encoder step.
negative_prompt_embeds (`Tensor`, *optional*):
negative text embeddings used to guide the image generation. Can be generated from text_encoder step.
negative_prompt_embeds_mask (`Tensor`, *optional*):
mask for the negative text embeddings. Can be generated from text_encoder step.
latents (`Tensor`, *optional*):
Pre-generated noisy latents for image generation.
height (`int`, *optional*):
The height in pixels of the generated image.
width (`int`, *optional*):
The width in pixels of the generated image.
generator (`Generator`, *optional*):
Torch generator for deterministic generation.
num_inference_steps (`int`, *optional*, defaults to 28):
The number of denoising steps.
sigmas (`list`, *optional*):
Custom sigmas for the denoising process.
mu (`float`, *optional*):
Fixed timestep shift for the scheduler. Pass `1.15` for the few-step distilled (TDM/turbo) checkpoint; if
not provided, computed from the image sequence length (base checkpoint behavior).
**denoiser_input_fields (`None`, *optional*):
conditional model inputs for the denoiser: e.g. prompt_embeds, negative_prompt_embeds, etc.
Outputs:
latents (`Tensor`):
Denoised latents.
"""
model_name = "krea2"
block_classes = [
Krea2TextInputsStep(),
Krea2PrepareLatentsStep(),
Krea2SetTimestepsStep(),
Krea2RoPEInputsStep(),
Krea2DenoiseStep(),
Krea2AfterDenoiseStep(),
]
block_names = [
"input",
"prepare_latents",
"set_timesteps",
"prepare_rope_inputs",
"denoise",
"after_denoise",
]
@property
def description(self):
return "step that denoise noise into image for text2image task. It includes the denoise loop, as well as prepare the inputs (timesteps, latents, rope inputs etc.)."
@property
def outputs(self):
return [
OutputParam.template("latents"),
]
# Krea 2 (inpainting)
# auto_docstring
class Krea2InpaintCoreDenoiseStep(SequentialPipelineBlocks):
"""
step that denoise noise into image for inpaint task. It includes the denoise loop, as well as prepare the inputs
(timesteps, latents, rope inputs etc.).
Components:
pachifier (`Krea2Pachifier`) scheduler (`FlowMatchEulerDiscreteScheduler`) guider (`ClassifierFreeGuidance`)
transformer (`Krea2Transformer2DModel`)
Inputs:
num_images_per_prompt (`int`, *optional*, defaults to 1):
The number of images to generate per prompt.
prompt_embeds (`Tensor`):
text embeddings used to guide the image generation. Can be generated from text_encoder step.
prompt_embeds_mask (`Tensor`):
mask for the text embeddings. Can be generated from text_encoder step.
negative_prompt_embeds (`Tensor`, *optional*):
negative text embeddings used to guide the image generation. Can be generated from text_encoder step.
negative_prompt_embeds_mask (`Tensor`, *optional*):
mask for the negative text embeddings. Can be generated from text_encoder step.
height (`int`, *optional*):
The height in pixels of the generated image.
width (`int`, *optional*):
The width in pixels of the generated image.
image_latents (`Tensor`, *optional*):
image latents used to guide the image generation. Can be generated from vae_encoder step.
processed_mask_image (`Tensor`, *optional*):
The processed mask image
latents (`Tensor`, *optional*):
Pre-generated noisy latents for image generation.
generator (`Generator`, *optional*):
Torch generator for deterministic generation.
num_inference_steps (`int`, *optional*, defaults to 28):
The number of denoising steps.
sigmas (`list`, *optional*):
Custom sigmas for the denoising process.
mu (`float`, *optional*):
Fixed timestep shift for the scheduler. Pass `1.15` for the few-step distilled (TDM/turbo) checkpoint; if
not provided, computed from the image sequence length (base checkpoint behavior).
strength (`float`, *optional*, defaults to 0.9):
Strength for img2img/inpainting.
**denoiser_input_fields (`None`, *optional*):
conditional model inputs for the denoiser: e.g. prompt_embeds, negative_prompt_embeds, etc.
Outputs:
latents (`Tensor`):
Denoised latents.
"""
model_name = "krea2"
block_classes = [
Krea2InpaintInputStep(),
Krea2PrepareLatentsStep(),
Krea2SetTimestepsWithStrengthStep(),
Krea2InpaintPrepareLatentsStep(),
Krea2RoPEInputsStep(),
Krea2InpaintDenoiseStep(),
Krea2AfterDenoiseStep(),
]
block_names = [
"input",
"prepare_latents",
"set_timesteps",
"prepare_inpaint_latents",
"prepare_rope_inputs",
"denoise",
"after_denoise",
]
@property
def description(self):
return "step that denoise noise into image for inpaint task. It includes the denoise loop, as well as prepare the inputs (timesteps, latents, rope inputs etc.)."
@property
def outputs(self):
return [
OutputParam.template("latents"),
]
# Krea 2 (image2image)
# auto_docstring
class Krea2Img2ImgCoreDenoiseStep(SequentialPipelineBlocks):
"""
step that denoise noise into image for img2img task. It includes the denoise loop, as well as prepare the inputs
(timesteps, latents, rope inputs etc.).
Components:
pachifier (`Krea2Pachifier`) scheduler (`FlowMatchEulerDiscreteScheduler`) guider (`ClassifierFreeGuidance`)
transformer (`Krea2Transformer2DModel`)
Inputs:
num_images_per_prompt (`int`, *optional*, defaults to 1):
The number of images to generate per prompt.
prompt_embeds (`Tensor`):
text embeddings used to guide the image generation. Can be generated from text_encoder step.
prompt_embeds_mask (`Tensor`):
mask for the text embeddings. Can be generated from text_encoder step.
negative_prompt_embeds (`Tensor`, *optional*):
negative text embeddings used to guide the image generation. Can be generated from text_encoder step.
negative_prompt_embeds_mask (`Tensor`, *optional*):
mask for the negative text embeddings. Can be generated from text_encoder step.
height (`int`, *optional*):
The height in pixels of the generated image.
width (`int`, *optional*):
The width in pixels of the generated image.
image_latents (`Tensor`):
image latents used to guide the image generation. Can be generated from vae_encoder step.
latents (`Tensor`, *optional*):
Pre-generated noisy latents for image generation.
generator (`Generator`, *optional*):
Torch generator for deterministic generation.
num_inference_steps (`int`, *optional*, defaults to 28):
The number of denoising steps.
sigmas (`list`, *optional*):
Custom sigmas for the denoising process.
mu (`float`, *optional*):
Fixed timestep shift for the scheduler. Pass `1.15` for the few-step distilled (TDM/turbo) checkpoint; if
not provided, computed from the image sequence length (base checkpoint behavior).
strength (`float`, *optional*, defaults to 0.9):
Strength for img2img/inpainting.
**denoiser_input_fields (`None`, *optional*):
conditional model inputs for the denoiser: e.g. prompt_embeds, negative_prompt_embeds, etc.
Outputs:
latents (`Tensor`):
Denoised latents.
"""
model_name = "krea2"
block_classes = [
Krea2Img2ImgInputStep(),
Krea2PrepareLatentsStep(),
Krea2SetTimestepsWithStrengthStep(),
Krea2PrepareLatentsWithStrengthStep(),
Krea2RoPEInputsStep(),
Krea2DenoiseStep(),
Krea2AfterDenoiseStep(),
]
block_names = [
"input",
"prepare_latents",
"set_timesteps",
"prepare_img2img_latents",
"prepare_rope_inputs",
"denoise",
"after_denoise",
]
@property
def description(self):
return "step that denoise noise into image for img2img task. It includes the denoise loop, as well as prepare the inputs (timesteps, latents, rope inputs etc.)."
@property
def outputs(self):
return [
OutputParam.template("latents"),
]
# Auto denoise step for Krea 2
class Krea2AutoCoreDenoiseStep(AutoPipelineBlocks):
model_name = "krea2"
block_classes = [
Krea2InpaintCoreDenoiseStep,
Krea2Img2ImgCoreDenoiseStep,
Krea2CoreDenoiseStep,
]
block_names = [
"inpaint",
"img2img",
"text2image",
]
block_trigger_inputs = ["processed_mask_image", "image_latents", None]
@property
def description(self):
return (
"Core step that performs the denoising process. \n"
+ " - `Krea2InpaintCoreDenoiseStep` (inpaint) is used when `processed_mask_image` is provided.\n"
+ " - `Krea2Img2ImgCoreDenoiseStep` (img2img) is used when `image_latents` is provided.\n"
+ " - `Krea2CoreDenoiseStep` (text2image) is used otherwise.\n"
+ "This step support text-to-image, image-to-image, and inpainting tasks for Krea 2:\n"
+ " - for image-to-image generation, you need to provide `image_latents`\n"
+ " - for inpainting, you need to provide `processed_mask_image` and `image_latents`\n"
+ " - for text-to-image generation, all you need to provide is prompt embeddings"
)
@property
def outputs(self):
return [
OutputParam.template("latents"),
]
# ====================
# 4. DECODE
# ====================
# standard decode step works for most tasks except for inpaint
# auto_docstring
class Krea2DecodeStep(SequentialPipelineBlocks):
"""
Decode step that decodes the latents to images and postprocess the generated image.
Components:
vae (`AutoencoderKLQwenImage`) image_processor (`VaeImageProcessor`)
Inputs:
latents (`Tensor`):
The denoised latents to decode, can be generated in the denoise step and unpacked in the after denoise
step.
output_type (`str`, *optional*, defaults to pil):
Output format: 'pil', 'np', 'pt'.
Outputs:
images (`list`):
Generated images. (tensor output of the vae decoder.)
"""
model_name = "krea2"
block_classes = [Krea2DecoderStep(), Krea2ProcessImagesOutputStep()]
block_names = ["decode", "postprocess"]
@property
def description(self):
return "Decode step that decodes the latents to images and postprocess the generated image."
# Inpaint decode step
# auto_docstring
class Krea2InpaintDecodeStep(SequentialPipelineBlocks):
"""
Decode step that decodes the latents to images and postprocess the generated image, optionally apply the mask
overlay to the original image.
Components:
vae (`AutoencoderKLQwenImage`) image_mask_processor (`InpaintProcessor`)
Inputs:
latents (`Tensor`):
The denoised latents to decode, can be generated in the denoise step and unpacked in the after denoise
step.
output_type (`str`, *optional*, defaults to pil):
Output format: 'pil', 'np', 'pt'.
mask_overlay_kwargs (`dict`, *optional*):
The kwargs for the postprocess step to apply the mask overlay. generated in
Krea2InpaintProcessImagesInputStep.
Outputs:
images (`list`):
Generated images. (tensor output of the vae decoder.)
"""
model_name = "krea2"
block_classes = [Krea2DecoderStep(), Krea2InpaintProcessImagesOutputStep()]
block_names = ["decode", "postprocess"]
@property
def description(self):
return "Decode step that decodes the latents to images and postprocess the generated image, optionally apply the mask overlay to the original image."
# Auto decode step for Krea 2
class Krea2AutoDecodeStep(AutoPipelineBlocks):
model_name = "krea2"
block_classes = [Krea2InpaintDecodeStep, Krea2DecodeStep]
block_names = ["inpaint_decode", "decode"]
block_trigger_inputs = ["mask", None]
@property
def description(self):
return (
"Decode step that decode the latents into images. \n"
" This is an auto pipeline block that works for inpaint/text2image/img2img tasks.\n"
+ " - `Krea2InpaintDecodeStep` (inpaint_decode) is used when `mask` is provided.\n"
+ " - `Krea2DecodeStep` (decode) is used when `mask` is not provided.\n"
)
# ====================
# 5. AUTO BLOCKS & PRESETS
# ====================
AUTO_BLOCKS = InsertableDict(
[
("text_encoder", Krea2AutoTextEncoderStep()),
("vae_encoder", Krea2AutoVaeEncoderStep()),
("denoise", Krea2AutoCoreDenoiseStep()),
("decode", Krea2AutoDecodeStep()),
]
)
# auto_docstring
class Krea2AutoBlocks(SequentialPipelineBlocks):
"""
Auto Modular pipeline for text-to-image, image-to-image, and inpainting tasks using Krea 2.
Supported workflows:
- `text2image`: requires `prompt`
- `image2image`: requires `prompt`, `image`
- `inpainting`: requires `prompt`, `mask_image`, `image`
Components:
text_encoder (`Qwen3VLModel`): The text encoder to use tokenizer (`Qwen2Tokenizer`): The tokenizer to use
guider (`ClassifierFreeGuidance`) image_mask_processor (`InpaintProcessor`) vae (`AutoencoderKLQwenImage`)
image_processor (`VaeImageProcessor`) pachifier (`Krea2Pachifier`) scheduler
(`FlowMatchEulerDiscreteScheduler`) transformer (`Krea2Transformer2DModel`)
Inputs:
prompt (`str`, *optional*):
The prompt or prompts to guide image generation.
negative_prompt (`str`, *optional*):
The prompt or prompts not to guide the image generation.
max_sequence_length (`int`, *optional*, defaults to 512):
Maximum sequence length for prompt encoding.
mask_image (`Image`, *optional*):
Mask image for inpainting.
image (`Image | list`, *optional*):
Reference image(s) for denoising. Can be a single image or list of images.
height (`int`, *optional*):
The height in pixels of the generated image.
width (`int`, *optional*):
The width in pixels of the generated image.
padding_mask_crop (`int`, *optional*):
Padding for mask cropping in inpainting.
generator (`Generator`, *optional*):
Torch generator for deterministic generation.
num_images_per_prompt (`int`, *optional*, defaults to 1):
The number of images to generate per prompt.
prompt_embeds (`Tensor`):
text embeddings used to guide the image generation. Can be generated from text_encoder step.
prompt_embeds_mask (`Tensor`):
mask for the text embeddings. Can be generated from text_encoder step.
negative_prompt_embeds (`Tensor`, *optional*):
negative text embeddings used to guide the image generation. Can be generated from text_encoder step.
negative_prompt_embeds_mask (`Tensor`, *optional*):
mask for the negative text embeddings. Can be generated from text_encoder step.
image_latents (`Tensor`, *optional*):
image latents used to guide the image generation. Can be generated from vae_encoder step.
processed_mask_image (`Tensor`, *optional*):
The processed mask image
latents (`Tensor`):
Pre-generated noisy latents for image generation.
num_inference_steps (`int`):
The number of denoising steps.
sigmas (`list`, *optional*):
Custom sigmas for the denoising process.
mu (`float`, *optional*):
Fixed timestep shift for the scheduler. Pass `1.15` for the few-step distilled (TDM/turbo) checkpoint; if
not provided, computed from the image sequence length (base checkpoint behavior).
strength (`float`, *optional*, defaults to 0.9):
Strength for img2img/inpainting.
**denoiser_input_fields (`None`, *optional*):
conditional model inputs for the denoiser: e.g. prompt_embeds, negative_prompt_embeds, etc.
output_type (`str`, *optional*, defaults to pil):
Output format: 'pil', 'np', 'pt'.
mask_overlay_kwargs (`dict`, *optional*):
The kwargs for the postprocess step to apply the mask overlay. generated in
Krea2InpaintProcessImagesInputStep.
Outputs:
images (`list`):
Generated images.
"""
model_name = "krea2"
block_classes = AUTO_BLOCKS.values()
block_names = AUTO_BLOCKS.keys()
# Workflow map defines the trigger conditions for each workflow.
# How to define:
# - Only include required inputs and trigger inputs (inputs that determine which blocks run)
# - currently, only supports `True` means the workflow triggers when the input is not None
_workflow_map = {
"text2image": {"prompt": True},
"image2image": {"prompt": True, "image": True},
"inpainting": {"prompt": True, "mask_image": True, "image": True},
}
@property
def description(self):
return "Auto Modular pipeline for text-to-image, image-to-image, and inpainting tasks using Krea 2."
@property
def outputs(self):
return [OutputParam.template("images")]
|