3jK.SSKJr SSKJr SSKrSSKJr SSKJrJ r SSK J r SSK J r JrJr SS KJr SS KJrJr SS KJrJrJr SS KJr SS KJr SSKJrJr \R@"\!5r"\"SS\ 55r#"SS\\\\ 5r$"SS\5r%g)) dataclass)AnyN) ConfigMixinregister_to_config)PeftAdapterMixin) BaseOutputapply_lora_scalelogging)AttentionMixin)ControlNetConditioningEmbedding zero_module)*CombinedTimestepGuidanceTextProjEmbeddings"CombinedTimestepTextProjEmbeddings FluxPosEmbed)Transformer2DModelOutput) ModelMixin)FluxSingleTransformerBlockFluxTransformerBlockc^\rSrSr%\\R \S'\\R \S'Srg)FluxControlNetOutput'controlnet_block_samplescontrolnet_single_block_samplesN) __name__ __module__ __qualname____firstlineno__tupletorchTensor__annotations____static_attributes__rf/home/wildlama/miniconda3/lib/python3.13/site-packages/diffusers/models/controlnets/controlnet_flux.pyrr's #ELL11%*5<<%88r&rc ^\rSrSrSr\SSSSSSS S S /S QS S 4 S\S\S\S\S\S\S\S\S\S\\S\S\4U4Sjjj5r \ S+S\S\S\S\4Sjj5r \ "S5S,S\ RS\ RS\ RS \S!\ RS"\ RS#\ R S$\ RS%\ RS&\ RS\\\4S -S'\S(\ R(\-4S)jj5rS*rU=r$)-FluxControlNetModel-T@&iiF)8r2N patch_size in_channels num_layersnum_single_layersattention_head_dimnum_attention_headsjoint_attention_dimpooled_projection_dimguidance_embedsaxes_dims_ropenum_modeconditioning_embedding_channelsc H>[TU]5 X lXe-Ul[ SU S9UlU (a[ O[n U "URUS9Ul[R"XpR5Ul [RRX R5Ul [R"[U5Vs/sHn[!URUUS9PM sn5Ul[R"[U5Vs/sHn[%URUUS9PM sn5Ul[R"/5Ul[[+UR"55HPnUR(R-[/[R"URUR555 MR [R"/5Ul[[+UR&55HPnUR0R-[/[R"URUR555 MR U SLUlUR2(a%[R4"XR5UlU b>[9U SS9Ul[RRX R5UlO>SUl[/[RRX R55UlSUlgs snfs snf)Ni')thetaaxes_dim) embedding_dimr:)dimr8r7)r1r1r1r1)r>block_out_channelsF) super__init__ out_channels inner_dimr pos_embedrrtime_text_embednnLinearcontext_embedderr" x_embedder ModuleListrangertransformer_blocksrsingle_transformer_blockscontrolnet_blockslenappendrcontrolnet_single_blocksunion Embeddingcontrolnet_mode_embedderrinput_hint_blockcontrolnet_x_embeddergradient_checkpointing)selfr3r4r5r6r7r8r9r:r;r<r=r>text_time_guidance_clsi_ __class__s r'rFFluxControlNetModel.__init__0sk ',A%ENK:I 6Oq  6..@U !# *=~~ N((//+~~F"$--z*   +A %(;'9 +   # *,01   2A +(;'9 2   * &"$r!2s42234A  " " ) )+biiPTP^P^6_*` a5)+ b(9%s499:;A  ) ) 0 0RYYt~~W[WeWe=f1g h<T) ::,.LL>>,RD ) * 6$C0Odt%D !*/nn)UD &$(D !)4UXX__[R`R`5a)bD &&+#W  s LLc[UR5nX'S'X7S'XGS'XWS'URU5nU(GaJURR URR 55 UR R UR R 55 URR URR 55 URR URR 55 URR URR 5SS9 URR URR 5SS9 [UR5Ul U$)Nr5r6r7r8F)strict) dictconfig from_configrIload_state_dict state_dictrJrMrNrQrRrr[) cls transformerr5r6r7r8load_weights_from_transformerrf controlnets r'from_transformer$FluxControlNetModel.from_transformer}sFk(())|&7"#'9#$(;$%__V, (  0 01F1F1Q1Q1S T  & & 6 6{7R7R7]7]7_ `  ' ' 7 7 8T8T8_8_8a b  ! ! 1 1+2H2H2S2S2U V  ) ) 9 9+:X:X:c:c:ens 9 t  0 0 @ @55@@B5 A 0;:;[;[/\J ,r&joint_attention_kwargs hidden_statescontrolnet_condcontrolnet_modeconditioning_scaleencoder_hidden_statespooled_projectionstimestepimg_idstxt_idsguidance return_dictreturnc URU5nURbURU5nURupnnXRR-nUURR-nUR XUURRUURR5nUR SSSSSS5nUR U UU-S5nXRU5-nURUR5S -nU bU RUR5S -n OSn U cURXv5OURXzU5nURU5nU RS:Xa[RS 5 U Sn URS:Xa[RS 5 USnUR(aPUc [!S 5eUR#U5n[$R&"UU/SS 9n[$R&"U SSU /SS 9n [$R&"X4SS 9nUR)U5nSn[+UR,5HZunn[$R."5(a)UR0(aUR3UUUUU5upQO U"UUUUS9upQUU4-nM\ Sn[+UR45HZunn[$R."5(a)UR0(aUR3UUUUU5upQO U"UUUUS9upQUU4-nM\ SnUR85HunnU"U5nUU4-nM Sn[7UUR:5HunnU"U5nUU4-nM UV s/sHn U U-PM nn UV s/sHn U U-PM nn [=U5S:XaSOUn[=U5S:XaSOUnU (dUU4$[?UUS9$s sn fs sn f)a The [`FluxTransformer2DModel`] forward method. Args: hidden_states (`torch.FloatTensor` of shape `(batch size, channel, height, width)`): Input `hidden_states`. controlnet_cond (`torch.Tensor`): The conditional input tensor of shape `(batch_size, sequence_length, hidden_size)`. controlnet_mode (`torch.Tensor`): The mode tensor of shape `(batch_size, 1)`. conditioning_scale (`float`, defaults to `1.0`): The scale factor for ControlNet outputs. encoder_hidden_states (`torch.FloatTensor` of shape `(batch size, sequence_len, embed_dims)`): Conditional embeddings (embeddings computed from the input conditions such as prompts) to use. pooled_projections (`torch.FloatTensor` of shape `(batch_size, projection_dim)`): Embeddings projected from the embeddings of input conditions. timestep ( `torch.LongTensor`): Used to indicate denoising step. block_controlnet_hidden_states: (`list` of `torch.Tensor`): A list of tensors that if specified are added to the residuals of transformer blocks. joint_attention_kwargs (`dict`, *optional*): A kwargs dictionary that if specified is passed along to the `AttentionProcessor` as defined under `self.processor` in [diffusers.models.attention_processor](https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py). return_dict (`bool`, *optional*, defaults to `True`): Whether or not to return a [`~models.transformer_2d.Transformer2DModelOutput`] instead of a plain tuple. Returns: If `return_dict` is True, an [`~models.transformer_2d.Transformer2DModelOutput`] is returned, otherwise a `tuple` where the first element is the sample tensor. Nrr r+rizrPassing `txt_ids` 3d torch.Tensor is deprecated.Please remove the batch dimension and pass it as a 2d torch TensorzrPassing `img_ids` 3d torch.Tensor is deprecated.Please remove the batch dimension and pass it as a 2d torch TensorzA`controlnet_mode` cannot be `None` when applying ControlNet-Union)rCr)rqrutembimage_rotary_emb)rr) rNrZshaperfr3reshapepermuter[todtyperJrMndimloggerwarningrW ValueErrorrYr"catrI enumeraterQis_grad_enabledr\_gradient_checkpointing_funcrRziprSrVrTr)!r]rqrrrsrtrurvrwrxryrzrpr{ batch_sizechannels height_pwwidth_pwheightwidthrcontrolnet_mode_embidsr block_samples index_blockblocksingle_block_samplesr block_samplecontrolnet_blockrsingle_block_samplesamples! r'forwardFluxControlNetModel.forwardsZ` 6  ,"33ODO8G8M8M 5J)X++"8"88F 6 66E-55fdkk.D.DeT[[McMcO.55aAq!QGO-55j&5.RTUO%(B(B?(SS ;;}223d:  {{=#6#67$>HH   >%%h:LM !% 5 56K L <<1  NNU ajG <<1  NNU ajG ::& !dee"&"?"?"P $)II/BDY.Z`a$b !ii!g 6A>Gii*2>>#. "+D,C,C"D K$$&&4+F+F7;7X7X!)$ 84%}8="/*?%5 84% *],<./ )-$'.2+/%) $ $!%8< e ||e e  e " e %|| e "LLe ""e e e ,,e !%S#X 5e e    5 5e 0e r&r)cZ^\rSrSrSrU4SjrSS\RS\\RS\\RS\\ S \RS \RS \RS \RS \RS\RS\ \\4S-S\S\\-4SjjrSrU=r$)FluxMultiControlNetModeliFa `FluxMultiControlNetModel` wrapper class for Multi-FluxControlNetModel This module is a wrapper for multiple instances of the `FluxControlNetModel`. The `forward()` API is designed to be compatible with `FluxControlNetModel`. Args: controlnets (`list[FluxControlNetModel]`): Provides additional conditioning to the unet during the denoising process. You must set multiple `FluxControlNetModel` as a list. cX>[TU]5 [R"U5Ulg)N)rErFrKrOnets)r] controlnetsras r'rF!FluxMultiControlNetModel.__init__Ss MM+. r&Nrqrrrsrtrurvrwrxryrzrpr{r|c 0[UR5S:XaURSn [[X#U55HununnnU "UUUSS2S4UUU UUU UU U S9 unnUS:XaUnUnM1Ub'Wb$[UU5VVs/sH unnUU-PM nnnUcM`WcMe[UU5VVs/sH unnUU-PM nnnM WW4$[[X#X@R55Hununnnn U "UUUSS2S4UUU UUU UU U S9 unnUS:XaUnUnM2Ub'Wb$[UU5VVs/sH unnUU-PM nnnUcMaWcMf[UU5VVs/sH unnUU-PM nnnM WW4$s snnfs snnfs snnfs snnf)Nr+r) rqrrrsrtrwrzrvruryrxrpr{)rTrrr)r]rqrrrsrtrurvrwrxryrzrpr{rmr_imagemodescalerrcontrol_block_samplescontrol_single_block_samplescontrol_block_samplercontrol_single_block_samples r'r FluxMultiControlNetModel.forwardWsz" tyy>Q 1J+4S[m5n+o''E46@"/$)$(DM',%%'9*?##+A + 73 3 6,9)3G0$05J5VGJJ_anFo1Fo B 4l1<?Fo.1,7RN8N I ;\8,FN5847,pR%&BBBI8AO6H))T833E4 7A"/$)$(DM',%%'9*?##+A + 73 3 6,9)3G0$05J5VGJJ_anFo1Fo B 4l1<?Fo.1,7RN8N I ;\8,FN584;8H%&BBBg1 8D1 8sF2F5F #F)r)NNNNNNNT)rrrr __doc__rFr"rrtensorrr#rrerrrrr!rr%rrs@r'rrFs //3+/%) $ $!%8< ]C((]Cell+]Cell+ ]C !K ]C %|| ]C"LL]C""]C]C]C,,]C!%S#X 5]C]C  %]C]Cr&r)& dataclassesrtypingrr"torch.nnrKconfiguration_utilsrrloadersrutilsr r r attentionr controlnets.controlnetrr embeddingsrrrmodeling_outputsrmodeling_utilsrtransformers.transformer_fluxrr get_loggerrrrr)rrr&r'rs" B' 'Quu7'\   H % 9:9 9 V *nkCSV rnCznCr&