3j} SSKrSSKrSSKrSSKJrJr SSKJr SSK J r J r SSK J r Jr SSKJrJrJr S S K JrJrJr S.S \S \S \S\4SjjrS/S\S-S\\R4-S-S\\S-S\\S-4SjjrSr"SS\ 5r"SS\ 5r"SS\ 5r "SS\ 5r!"SS\ 5r""S S!\ 5r#"S"S#\ 5r$"S$S%\ 5r%"S&S'\ 5r&"S(S)\ 5r'"S*S+\ 5r("S,S-\ 5r)g)0N)QwenImageControlNetModelQwenImageMultiControlNetModel)FlowMatchEulerDiscreteScheduler) randn_tensor unwrap_module)ModularPipelineBlocks PipelineState) ComponentSpec InputParam OutputParam)QwenImageLayeredPachifierQwenImageModularPipelineQwenImagePachifier base_seq_len max_seq_len base_shift max_shiftc4XC- X!- - nX5U-- nX-U-nU$)N) image_seq_lenrrrrmbmus n/home/wildlama/miniconda3/lib/python3.13/site-packages/diffusers/modular_pipelines/qwenimage/before_denoise.pycalculate_shiftrs3  K$>?A%%A  Q B Inum_inference_stepsdevice timestepssigmasc UbUb [S5eUbS[[R"UR5R R 55;nU(d[SURS35eUR"S X2S.UD6 URn[U5nX14$UbS[[R"UR5R R 55;nU(d[SURS35eUR"S XBS.UD6 URn[U5nX14$UR"U4S U0UD6 URnX14$) a Calls the scheduler's `set_timesteps` method and retrieves timesteps from the scheduler after the call. Handles custom timesteps. Any kwargs will be supplied to `scheduler.set_timesteps`. Args: scheduler (`SchedulerMixin`): The scheduler to get timesteps from. num_inference_steps (`int`): The number of diffusion steps used when generating samples with a pre-trained model. If used, `timesteps` must be `None`. device (`str` or `torch.device`, *optional*): The device to which the timesteps should be moved to. If `None`, the timesteps are not moved. timesteps (`list[int]`, *optional*): Custom timesteps used to override the timestep spacing strategy of the scheduler. If `timesteps` is passed, `num_inference_steps` and `sigmas` must be `None`. sigmas (`list[float]`, *optional*): Custom sigmas used to override the timestep spacing strategy of the scheduler. If `sigmas` is passed, `num_inference_steps` and `timesteps` must be `None`. Returns: `tuple[torch.Tensor, int]`: A tuple where the first element is the timestep schedule from the scheduler and the second element is the number of inference steps. zYOnly one of `timesteps` or `sigmas` can be passed. Please choose one to set custom valuesr"zThe current scheduler class zx's `set_timesteps` does not support custom timestep schedules. Please check whether you are using the correct scheduler.)r"r!r#zv's `set_timesteps` does not support custom sigmas schedules. Please check whether you are using the correct scheduler.)r#r!r!r) ValueErrorsetinspect signature set_timesteps parameterskeys __class__r"len) schedulerr r!r"r#kwargsaccepts_timesteps accept_sigmass rretrieve_timestepsr2+s}>!3tuu'3w/@/@AXAX/Y/d/d/i/i/k+ll .y/B/B.CDab  M)MfM'' !)n  ))   C(9(9):Q:Q(R(](](b(b(d$ee .y/B/B.CD_`  GvGG'' !)n  ))  3MFMfM''  ))rc[X-U5n[[X- S55nURX@R-Sn[ US5(aUR X@R-5 XQU- 4$)Nrset_begin_index)minintmaxr"orderhasattrr4)r.r strength init_timestept_startr"s r get_timestepsr=gss+68KLM#)91=>G##Goo$=$?@Iy+,,!!'OO";< G3 33rc\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r \S\ \4Sj5r\S 5r\R&"5S \S \S\4S j5rS rg)QwenImagePrepareLatentsStepya Prepare initial random noise for the generation process Components: pachifier (`QwenImagePachifier`) Inputs: latents (`Tensor`, *optional*): Pre-generated noisy latents for image generation. height (`int`, *optional*): The height in pixels of the generated image. width (`int`, *optional*): The width in pixels of the generated image. num_images_per_prompt (`int`, *optional*, defaults to 1): The number of images to generate per prompt. generator (`Generator`, *optional*): Torch generator for deterministic generation. batch_size (`int`, *optional*, defaults to 1): Number of prompts, the final batch size of model inputs should be batch_size * num_images_per_prompt. Can be generated in input step. dtype (`dtype`, *optional*, defaults to torch.float32): The dtype of the model inputs, can be generated in input step. Outputs: height (`int`): if not set, updated to default value width (`int`): if not set, updated to default value latents (`Tensor`): The initial latents to use for the denoising process qwenimagereturncg)Nz7Prepare initial random noise for the generation processrselfs r description'QwenImagePrepareLatentsStep.descriptionsHrc"[S[SS9/$N pachifier from_config)default_creation_methodr rrDs rexpected_components/QwenImagePrepareLatentsStep.expected_components +'9S` a  rc ,[R"S5[R"S5[R"S5[R"S5[R"S5[R"S5[R"S5/$)Nlatentsheightwidthnum_images_per_prompt generator batch_sizedtyper templaterDs rinputs"QwenImagePrepareLatentsStep.inputssq    *    )    (    7 8    ,    -    (  rcn[S[SS9[S[SS9[S[RSS9/$NrSz$if not set, updated to default valuename type_hintrFrTrRz4The initial latents to use for the denoising processrr6torchTensorrDs rintermediate_outputs0QwenImagePrepareLatentsStep.intermediate_outputs= XBh i WAg h ,,R   rcUbXS--S:wa[SUS-SU35eUb XS--S:wa[SUS-SU35eggNr rzHeight must be divisible by z but is zWidth must be divisible by r%rSrTvae_scale_factors r check_inputs(QwenImagePrepareLatentsStep.check_inputs{  &q,@"AQ"F;!?1!D:;Ka;O:PPXY^X_`a a"E r componentsstatecURU5nURURURURS9 UR nUR UR-nUR=(d URUlUR=(d URUlS[UR5URS---nS[UR5URS---nXQRSXg4n[UR[5(a>[UR5U:wa%[!S[UR5SUS35eUR"cM[%XRXCR&S9UlUR(R+UR"5UlUR-X#5 X4$Nrkr rz/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.)rVr!rX)get_block_statermrSrTrl_execution_devicerWrUdefault_height default_widthr6num_channels_latents isinstancerVlistr-r%rRrrXrJ pack_latentsset_block_state rErprq block_stater!rW latent_height latent_widthshapes r__call__$QwenImagePrepareLatentsStep.__call__s**51  %%##'88  -- ++k.O.OO )//L:3L3L '--I1I1I S!3!349T9TWX9XYZ C 1 12z7R7RUV7VWX <[!UR5U:wa%[#S[!UR5SUS35eUR$cM['XRXCR(S9UlUR*R-UR$5UlUR/X#5 X4$rs)rtrmrSrTrlrurWrUrvrwr6rrxryrVrzr-r%rRrrXrJr{r|r}s rr+QwenImageLayeredPrepareLatentsStep.__call__=s**51  %%##'88  -- ++k.O.OO )//L:3L3L '--I1I1I S!3!349T9TWX9XYZ C 1 12z7R7RUV7VWX //!3Z5T5TVcr k++T 2 2s;;P;P7QU_7_A#kF[F[B\A]^&<'gi     &".!6!6vM^M^#K #-"6"6"C"CKDWDW"XK  U0  rrNrrrrrrs D%J `S`` T-%8    Z(      d;&7    bb ]]_"!#;"!M"!Vc"!"!rrc\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r \S\ \4Sj5r\S 5r\R&"5S \S \S\4S j5rS rg)'QwenImagePrepareLatentsWithStrengthStepida Step that adds noise to image latents for image-to-image/inpainting. Should be run after set_timesteps, prepare_latents. Both noise and image latents should alreadybe patchified. Components: scheduler (`FlowMatchEulerDiscreteScheduler`) 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. 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. rArBcg)NzStep that adds noise to image latents for image-to-image/inpainting. Should be run after set_timesteps, prepare_latents. Both noise and image latents should alreadybe patchified.rrDs rrF3QwenImagePrepareLatentsWithStrengthStep.description~sDrc$[S[5/$Nr.r rrDs rrN;QwenImagePrepareLatentsWithStrengthStep.expected_components +'F G  rc [SS[RSS9[R"SSS9[SS[RS S9/$) NrRTzCThe initial random noised, can be generated in prepare latent step.r`requiredrarF image_latentsz>  % %8K8K  U0  rrNrrrrrrds.J DSDD T-%8   Z(  "  d;&7    ss ]]_!#;!M!Vc!!rrc\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r \S\ \4Sj5r\R""5S \S \S\4S j5rS rg )QwenImageCreateMaskLatentsStepia Step that creates mask latents from preprocessed mask_image by interpolating to latent space. Components: pachifier (`QwenImagePachifier`) Inputs: processed_mask_image (`Tensor`): The processed mask to use for the inpainting process. height (`int`): The height in pixels of the generated image. width (`int`): The width in pixels of the generated image. dtype (`dtype`, *optional*, defaults to torch.float32): The dtype of the model inputs, can be generated in input step. Outputs: mask (`Tensor`): The mask to use for the inpainting process. rArBcg)Nz]Step that creates mask latents from preprocessed mask_image by interpolating to latent space.rrDs rrF*QwenImageCreateMaskLatentsStep.descriptionsnrc"[S[SS9/$rIrMrDs rrN2QwenImageCreateMaskLatentsStep.expected_componentsrPrc[SS[RSS9[R"SSS9[R"SSS9[R"S5/$) Nprocessed_mask_imageTz5The processed mask to use for the inpainting process.rrSrrTrXrrDs rr[%QwenImageCreateMaskLatentsStep.inputssV +,,S     4 8   $ 7    (  rc6[S[RSS9/$)Nmaskz+The mask to use for the inpainting process.r_rrDs rre3QwenImageCreateMaskLatentsStep.intermediate_outputss" u||An   rrprqcURU5nURnS[UR5URS---nS[UR 5URS---n[ RRRURXV4S9Ul URRS5Ul URRSURSSS5Ul URRXCR S9Ul UR"R%UR5Ul UR'X#5 X4$)Nr )sizerr!rX)rtrur6rSrlrTrcnn functional interpolaterr unsqueezerrxtorXrJr{r|)rErprqr~r!height_latents width_latentss rr'QwenImageCreateMaskLatentsStep.__call__s9**51 -- c+"4"45*:U:UXY:YZ[S!2!23 8S8SVW8WXY  88..::  , , 0;  '++55a8 &++221j6U6UWXZ[]^_ &++..fDUDU.V %//<<[=M=MN  U0  rrN)rrrrrrrrrFrzr rNr r[rrercrrr rrrrrrrs*J oSoo T-%8    Z(     d;&7   ]]_!#;!M!Vc!!rrc\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r \S\ \4Sj5rS \S \S\4S jrS rg )QwenImageSetTimestepsStepi'aj Step that sets the scheduler's timesteps for text-to-image generation. Should be run after prepare latents step. Components: scheduler (`FlowMatchEulerDiscreteScheduler`) Inputs: num_inference_steps (`int`, *optional*, defaults to 50): The number of denoising steps. sigmas (`list`, *optional*): Custom sigmas for the denoising process. latents (`Tensor`): The initial random noised latents for the denoising process. Can be generated in prepare latents step. Outputs: timesteps (`Tensor`): The timesteps to use for the denoising process rArBcg)NzpStep that sets the scheduler's timesteps for text-to-image generation. Should be run after prepare latents step.rrDs rrF%QwenImageSetTimestepsStep.description=sBrc$[S[5/$rrrDs rrN-QwenImageSetTimestepsStep.expected_componentsArrc [R"S5[R"S5[SS[RSS9/$)Nr r#rRTzfThe initial random noised latents for the denoising process. Can be generated in prepare latents step.rr rZrcrdrDs rr[ QwenImageSetTimestepsStep.inputsGsF    5 6    ) ,,E    rc6[S[RSS9/$)Nr"z.The timesteps to use for the denoising processr_rrDs rre.QwenImageSetTimestepsStep.intermediate_outputsTs"  ELLFv   rrprqc URU5nURnURc/[R"SSUR - UR 5O URn[ URRSURRRSS5URRRSS5URRRSS5URRRS S 5S 9n[URUR UUUS 9uUl UlURRS 5 URX#5 X4$)N?rbase_image_seq_lenmax_image_seq_lenr?rffffff?rrrrrr.r r!r#rr)rtrur#nplinspacer rrRrr.configgetr2r"r4r|rErprqr~r!r#rs rr"QwenImageSetTimestepsStep.__call__\sP**51 --!!) KKQ!@!@@+BaBa b##  %--33A6#--44889MsS",,33778KTR!++2266|SI **1155k4H  BT ** + ? ? B > {> ,,Q/ U0  rrNrrrrrrrrrFrzr rNr r[rrerr rrrrrrr's&J BSBB T-%8    Z(     d;&7  !#;!M!Vc!rrc\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r \S\ \4Sj5r\R""5S \S\4S j5rS rg ) QwenImageLayeredSetTimestepsStepi}aO Set timesteps step for QwenImage Layered with custom mu calculation based on image_latents. Components: scheduler (`FlowMatchEulerDiscreteScheduler`) Inputs: num_inference_steps (`int`, *optional*, defaults to 50): The number of denoising steps. sigmas (`list`, *optional*): Custom sigmas for the denoising process. image_latents (`Tensor`): image latents used to guide the image generation. Can be generated from vae_encoder step. Outputs: timesteps (`Tensor`): The timesteps to use for the denoising process. rrBcg)Nz[Set timesteps step for QwenImage Layered with custom mu calculation based on image_latents.rrDs rrF,QwenImageLayeredSetTimestepsStep.descriptionslrc$[S[5/$rrrDs rrN4QwenImageLayeredSetTimestepsStep.expected_componentsrrc[R"S5[R"S5[R"S5/$)Nr r#rrYrDs rr['QwenImageLayeredSetTimestepsStep.inputss9    5 6    )    0  rc6[S[RSS9/$)Nr"/The timesteps to use for the denoising process.r_rrDs rre5QwenImageLayeredSetTimestepsStep.intermediate_outputss"  ELLFw   rrqcURU5nURnSnURRSU- S-nURc/[ R "SSUR- UR5O URn[URURUUUS9uUl UlURRS5 URX#5 X4$)Ngp@rrr)r#rr) rtrurrr#rrr r2r.r"r4r|)rErprqr~r! base_seqlenrr#s rr)QwenImageLayeredSetTimestepsStep.__call__s**51 --* ''--a0;>3 F !!) KKQ!@!@@+BaBa b##  BT   + +  B > {> ,,Q/ U0  rrN)rrrrrrrrrFrzr rNr r[rrercrr rrrrrrr}s&%J mSmm T-%8   Z(   d;&7   ]]_!-!M!!rrc\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r \S\ \4Sj5rS \S \S\4S jrS rg )%QwenImageSetTimestepsWithStrengthStepiaS Step that sets the scheduler's timesteps for image-to-image generation, and inpainting. Should be run after prepare latents step. Components: scheduler (`FlowMatchEulerDiscreteScheduler`) Inputs: num_inference_steps (`int`, *optional*, defaults to 50): The number of denoising steps. sigmas (`list`, *optional*): Custom sigmas for the denoising process. latents (`Tensor`): The latents to use for the denoising process. Can be generated in prepare latents step. strength (`float`, *optional*, defaults to 0.9): Strength for img2img/inpainting. Outputs: timesteps (`Tensor`): The timesteps to use for the denoising process. num_inference_steps (`int`): The number of denoising steps to perform at inference time. Updated based on strength. rArBcg)NzStep that sets the scheduler's timesteps for image-to-image generation, and inpainting. Should be run after prepare latents step.rrDs rrF1QwenImageSetTimestepsWithStrengthStep.descriptionsSrc$[S[5/$rrrDs rrN9QwenImageSetTimestepsWithStrengthStep.expected_componentsrrc [R"S5[R"S5[SS[RSS9[R"SSS 9/$) Nr r#rRTzWThe latents to use for the denoising process. Can be generated in prepare latents step.)rrarFr:g?)defaultrrDs rr[,QwenImageSetTimestepsWithStrengthStep.inputssU    5 6    ) ,,u      C 8  rcR[S[RSS9[S[SS9/$)Nr"rr_r zVThe number of denoising steps to perform at inference time. Updated based on strength.)rrcrdr6rDs rre:QwenImageSetTimestepsWithStrengthStep.intermediate_outputss6  ,,M  *t   rrprqc $URU5nURnURc/[R"SSUR - UR 5O URn[ URRSURRRSS5URRRSS5URRRSS5URRRS S 5S 9n[URUR UUUS 9uUl Ul[URUR URS 9uUl UlUR!X#5 X4$)Nrrrrrrrrrrrr)r.r r:)rtrur#rrr rrRrr.rrr2r"r=r:r|rs rr.QwenImageSetTimestepsWithStrengthStep.__call__sq**51 --!!) KKQ!@!@@+BaBa b##  %--33A6#--44889MsS",,33778KTR!++2266|SI **1155k4H  BT ** + ? ? B > {>BO ** + ? ? ))B > {> U0  rrNrrrrrrs0J SSSS T-%8    Z(      d;&7    !!#;!!M!!Vc!!rrc\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r S\S \S\4S jrS rg ) QwenImageRoPEInputsStepi;aS Step that prepares the RoPE inputs for the denoising process. Should be place after prepare_latents step Inputs: batch_size (`int`, *optional*, defaults to 1): Number of prompts, the final batch size of model inputs should be batch_size * num_images_per_prompt. Can be generated in input step. height (`int`): The height in pixels of the generated image. width (`int`): The width in pixels of the generated image. prompt_embeds_mask (`Tensor`): mask for the text embeddings. 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. Outputs: img_shapes (`list`): The shapes of the images latents, used for RoPE calculation rArBcgNzhStep that prepares the RoPE inputs for the denoising process. Should be place after prepare_latents steprrDs rrF#QwenImageRoPEInputsStep.descriptionS  w rc[R"S5[R"SSS9[R"SSS9[R"S5[R"S5/$)NrWrSTrrTprompt_embeds_masknegative_prompt_embeds_maskrYrDs rr[QwenImageRoPEInputsStep.inputsYsZ    -   4 8   $ 7    4 5    = >   rc d[SS[[[[[[4SS9/$N img_shapesdenoiser_input_fieldsz;The shapes of the images latents, used for RoPE calculationr` kwargs_typerarFrrztupler6rDs rre,QwenImageRoPEInputsStep.intermediate_outputsc7 !3tE#sC-$89:Y    rrprqcURU5nSURUR-S-URUR-S-4//UR-UlUR X#5 X4$Nrr )rtrSrlrTrWrr|rErprqr~s rr QwenImageRoPEInputsStep.__call__ns**51 &&**E*EEJ%%)D)DDI "   " ""#  U0  rrNrrrrrrrrrFrzr r[rrerr rrrrrrr;s*J  S   Z(   d;&7  !#;!M!Vc!rrc\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r S\S \S\4S jrS rg ) QwenImageEditRoPEInputsStepia@ Step that prepares the RoPE inputs for denoising process. This is used in QwenImage Edit. Should be placed after prepare_latents step Inputs: batch_size (`int`, *optional*, defaults to 1): Number of prompts, the final batch size of model inputs should be batch_size * num_images_per_prompt. Can be generated in input step. image_height (`int`): The height of the reference image. Can be generated in input step. image_width (`int`): The width of the reference image. Can be generated in input step. height (`int`): The height in pixels of the generated image. width (`int`): The width in pixels of the generated image. prompt_embeds_mask (`Tensor`): mask for the text embeddings. 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. Outputs: img_shapes (`list`): The shapes of the images latents, used for RoPE calculation rArBcg)NzStep that prepares the RoPE inputs for denoising process. This is used in QwenImage Edit. Should be placed after prepare_latents steprrDs rrF'QwenImageEditRoPEInputsStep.descriptionsWrc [R"S5[SS[SS9[SS[SS9[R"SSS 9[R"S SS 9[R"S 5[R"S 5/$) NrW image_heightTzBThe height of the reference image. Can be generated in input step.r image_widthzAThe width of the reference image. Can be generated in input step.rSrrTrr)r rZr6rDs rr["QwenImageEditRoPEInputsStep.inputss    - #`   "_     4 8   $ 7    4 5    = >#  rc d[SS[[[[[[4SS9/$rrrDs rre0QwenImageEditRoPEInputsStep.intermediate_outputsr!rrprqcVURU5nSURUR-S-URUR-S-4SURUR-S-UR UR-S-4//UR -UlURX#5 X4$r#) rtrSrlrTr,r-rWrr|r$s rr$QwenImageEditRoPEInputsStep.__call__s**51 &&**E*EEJ%%)D)DDI ,, 0K0KKqP++z/J/JJaO  "   " " "#  U0  rrNr&rrrr(r(s4J WSWW Z(  * d;&7  !#;!M!Vc!rr(c\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r S\S \S\4S jrS rg ) QwenImageEditPlusRoPEInputsStepia Step that prepares the RoPE inputs for denoising process. This is used in QwenImage Edit Plus. Unlike Edit, Edit Plus handles lists of image_height/image_width for multiple reference images. Should be placed after prepare_latents step. Inputs: batch_size (`int`, *optional*, defaults to 1): Number of prompts, the final batch size of model inputs should be batch_size * num_images_per_prompt. Can be generated in input step. image_height (`list`): The heights of the reference images. Can be generated in input step. image_width (`list`): The widths of the reference images. Can be generated in input step. height (`int`): The height in pixels of the generated image. width (`int`): The width in pixels of the generated image. prompt_embeds_mask (`Tensor`): mask for the text embeddings. 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. Outputs: img_shapes (`list`): The shapes of the image latents, used for RoPE calculation txt_seq_lens (`list`): The sequence lengths of the prompt embeds, used for RoPE calculation negative_txt_seq_lens (`list`): The sequence lengths of the negative prompt embeds, used for RoPE calculation zqwenimage-edit-plusrBcg)NzStep that prepares the RoPE inputs for denoising process. This is used in QwenImage Edit Plus. Unlike Edit, Edit Plus handles lists of image_height/image_width for multiple reference images. Should be placed after prepare_latents step.rrDs rrF+QwenImageEditPlusRoPEInputsStep.descriptions  ; rc ,[R"S5[SS[[SS9[SS[[SS9[R"SSS 9[R"S SS 9[R"S 5[R"S 5/$) NrWr,TzDThe heights of the reference images. Can be generated in input step.rr-zCThe widths of the reference images. Can be generated in input step.rSrrTrr)r rZrzr6rDs rr[&QwenImageEditPlusRoPEInputsStep.inputss    - #s)b   "s)a     4 8   $ 7    4 5    = >#  rc [SS[[[[[[4SS9[SS[[SS9[SS[[SS9/$) Nrr:The shapes of the image latents, used for RoPE calculationr txt_seq_lensDThe sequence lengths of the prompt embeds, used for RoPE calculationnegative_txt_seq_lensMThe sequence lengths of the negative prompt embeds, used for RoPE calculationrrDs rre4QwenImageEditPlusRoPEInputsStep.intermediate_outputssk !3tE#sC-$89:X   #3s)b   ,3s)k    rrprqc RURU5nURnSURU-S-URU-S-4/[ UR UR 5VVs/sHupVSXT-S-Xd-S-4PM snnQ/UR-UlURb'URRSS9R5OSUl URb'URRSS9R5OSUlURX#5 X4$s snnf)Nrr dim)rtrlrSrTzipr,r-rWrrsumtolistr;rr=r|)rErprqr~rl img_height img_widths rr(QwenImageEditPlusRoPEInputsStep.__call__3sO**51 %66 K&&*::a?ARARVfAfjkAkl 25[5M5M{OfOf1g1g-  6!;Y=Z^_=_`1g "   " ""# CNB`B`BlK * * . .1 . 5 < < >rv  66B  3 3 7 7A 7 > E E G ) U0  %s&D#rNr&rrrr4r4s>'J  S   Z(  * d;&7  ,!#;!M!Vc!rr4c\rSrSrSrSr\S\4Sj5r\S\ \ 4Sj5r \S\ \ 4Sj5r \R"5S\S\4S j5rS rg ) QwenImageLayeredRoPEInputsStepiRa Step that prepares the RoPE inputs for the denoising process. Should be place after prepare_latents step Inputs: batch_size (`int`, *optional*, defaults to 1): Number of prompts, the final batch size of model inputs should be batch_size * num_images_per_prompt. Can be generated in input step. layers (`int`, *optional*, defaults to 4): Number of layers to extract from the image height (`int`): The height in pixels of the generated image. width (`int`): The width in pixels of the generated image. prompt_embeds_mask (`Tensor`): mask for the text embeddings. 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. Outputs: img_shapes (`list`): The shapes of the image latents, used for RoPE calculation txt_seq_lens (`list`): The sequence lengths of the prompt embeds, used for RoPE calculation negative_txt_seq_lens (`list`): The sequence lengths of the negative prompt embeds, used for RoPE calculation additional_t_cond (`Tensor`): The additional t cond, used for RoPE calculation rrBcgrrrDs rrF*QwenImageLayeredRoPEInputsStep.descriptionrrrc[R"S5[R"S5[R"SSS9[R"SSS9[R"S5[R"S5/$) NrWrrSTrrTrrrYrDs rr[%QwenImageLayeredRoPEInputsStep.inputsxsh    -    )   4 8   $ 7    4 5    = >   rc [S[[[[[[4SSS9[S[[SSS9[S[[SSS9[S [R SS S9/$) Nrrr:)r`rarrFr;r<r=r>additional_t_condz0The additional t cond, used for RoPE calculation)rrzrr6rcrdrDs rre3QwenImageLayeredRoPEInputsStep.intermediate_outputss !tE#sC-$89:3X   #s)3b   ,s)3k   (,,3N  '  rrqcURU5nURnSURUR-S-URUR-S-4nU/UR S--/UR -UlURb'URRSS9R5OSUl URb'URRSS9R5OSUl [R"S/UR -5R!U[R"S9UlUR'X#5 X4$)Nrr rArr)rtrurSrlrTrrWrrrDrEr;rr=rctensorrlongrPr|)rErprqr~r!rs rr'QwenImageLayeredRoPEInputsStep.__call__sI**51 --    *"="= = B   !rv  66B  3 3 7 7A 7 > E E G ) ). aS;;Q;Q5Q(R(U(U]ckpkuku(U(v % U0  rrN)rrrrrrrrrFrzr r[rrercrr rrrrrrJrJRs:%J  S   Z(   d;&7  8 ]]_!-!M!!rrJc\rSrSrSrSr\S\\4Sj5r \S\ 4Sj5r \S\\ 4Sj5r \S\\4Sj5r\R""5S \S \S\4S j5rS rg )%QwenImageControlNetBeforeDenoiserStepia~ step that prepare inputs for controlnet. Insert before the Denoise Step, after set_timesteps step. Components: controlnet (`QwenImageControlNetModel`) Inputs: control_guidance_start (`float`, *optional*, defaults to 0.0): When to start applying ControlNet. control_guidance_end (`float`, *optional*, defaults to 1.0): When to stop applying ControlNet. controlnet_conditioning_scale (`float`, *optional*, defaults to 1.0): Scale for ControlNet conditioning. control_image_latents (`Tensor`): The control image latents to use for the denoising process. Can be generated in controlnet vae encoder step. timesteps (`Tensor`): The timesteps to use for the denoising process. Can be generated in set_timesteps step. Outputs: controlnet_keep (`list`): The controlnet keep values rArBc$[S[5/$)N controlnet)r rrDs rrN9QwenImageControlNetBeforeDenoiserStep.expected_componentss ,(@ A  rcg)Nzbstep that prepare inputs for controlnet. Insert before the Denoise Step, after set_timesteps step.rrDs rrF1QwenImageControlNetBeforeDenoiserStep.descriptionssrc [R"S5[R"S5[R"S5[SS[RSS9[SS[RS S9/$) Ncontrol_guidance_startcontrol_guidance_endcontrolnet_conditioning_scalecontrol_image_latentsTzlThe control image latents to use for the denoising process. Can be generated in controlnet vae encoder step.rr"rrrDs rr[,QwenImageControlNetBeforeDenoiserStep.inputssp    8 9    6 7    ? @ ,,,K    ,,u    rc0[S[[SS9/$)Ncontrolnet_keepzThe controlnet keep values)rarF)rrzfloatrDs rre:QwenImageControlNetBeforeDenoiserStep.intermediate_outputss  )T%[Nj k  rrprqc URU5n[UR5n[UR[ 5(dH[UR [ 5(a)[UR 5UR/-UlO[UR [ 5(dH[UR[ 5(a)[UR5UR /-UlO[UR[ 5(dt[UR [ 5(dU[U[5(a[UR5OSnXSR/-XSR /-sUlUl[U[5(a4[UR[5(aUR/W-Ul /Ul [[UR55Hn[URUR 5VVs/sHQupxS[U[UR5- U:=(d US-[UR5- U:5- PMS n nnURR![U["5(aU SOU 5 M UR%X#5 X4$s snnf)Nrrr)rtrrYryr^rzr_r-rrar`rerdranger"rCappendrr|) rErprqr~rYmultisekeepss rr.QwenImageControlNetBeforeDenoiserStep.__call__sy**51 ":#8#89 +<>EEj  , ,dO O ;EZQn:o:oK556uv ::;;8899 QK . 0P j"? @ @Z  5 5uF F :E9b9b8cfj8jK 5') #s;0012A B BKDdDdeeDAeAK$9$9 ::Q>j1q5CP[PePeLfBfijBjkke   ' ' . .:jRj;k;kuQxqv w 3 U0  sAKrN)rrrrrrrrzr rNrrFr r[rrercrrr rrrrrrWrWs0J  T-%8  tStt Z(  & d;&7   ]]_.!#;.!M.!Vc.!.!rrW)rrrr)NNNN)*r'numpyrrcmodelsrr schedulersrutils.torch_utilsrrmodular_pipeliner r modular_pipeline_utilsr r rrrrr6rerrr!rzr2r=r?rrrrrrrr(r4rJrWrrrrvs M9<CKKee        '+(,"&!% 8*t8* %,,  %8*Cy4 8* K$  8*x 4$p!"7p!hs!)>s!ne!.Ce!RQ!%:Q!tR! 5R!lL!'<L!`c!,Ac!\B!3B!LW!"7W!vr!&;r!lk!%:k!dn!,An!r