sketchkit.colorization.manga_ninja.manganinja package

Subpackages

• sketchkit.colorization.manga_ninja.manganinja.annotator package
  • Subpackages
    • sketchkit.colorization.manga_ninja.manganinja.annotator.lineart package
      • Module contents
  • Module contents
• sketchkit.colorization.manga_ninja.manganinja.models package
  • Submodules
  • sketchkit.colorization.manga_ninja.manganinja.models.attention module
    • BasicTransformerBlock
      • BasicTransformerBlock.forward()
      • BasicTransformerBlock.set_chunk_feed_forward()
    • TemporalBasicTransformerBlock
      • TemporalBasicTransformerBlock.forward()
  • sketchkit.colorization.manga_ninja.manganinja.models.attention_processor module
    • Attention
      • Attention.batch_to_head_dim()
      • Attention.forward()
      • Attention.fuse_projections()
      • Attention.get_attention_scores()
      • Attention.get_processor()
      • Attention.head_to_batch_dim()
      • Attention.norm_encoder_hidden_states()
      • Attention.prepare_attention_mask()
      • Attention.set_attention_slice()
      • Attention.set_processor()
      • Attention.set_use_memory_efficient_attention_xformers()
    • AttnAddedKVProcessor
    • AttnAddedKVProcessor2_0
    • AttnProcessor
    • AttnProcessor2_0
    • CustomDiffusionAttnProcessor
    • CustomDiffusionAttnProcessor2_0
    • CustomDiffusionXFormersAttnProcessor
    • FusedAttnProcessor2_0
    • IPAdapterAttnProcessor
    • IPAdapterAttnProcessor2_0
    • LoRAAttnAddedKVProcessor
    • LoRAAttnProcessor
    • LoRAAttnProcessor2_0
    • LoRAXFormersAttnProcessor
    • SlicedAttnAddedKVProcessor
    • SlicedAttnProcessor
    • SpatialNorm
      • SpatialNorm.forward()
    • XFormersAttnAddedKVProcessor
    • XFormersAttnProcessor
  • sketchkit.colorization.manga_ninja.manganinja.models.mutual_self_attention_multi_scale module
    • ReferenceAttentionControl
      • ReferenceAttentionControl.clear()
      • ReferenceAttentionControl.register_reference_hooks()
      • ReferenceAttentionControl.update()
    • filter_matrices_by_size()
    • torch_dfs()
  • sketchkit.colorization.manga_ninja.manganinja.models.refunet_2d_condition module
    • RefUNet2DConditionModel
      • RefUNet2DConditionModel._set_gradient_checkpointing()
      • RefUNet2DConditionModel._supports_gradient_checkpointing
      • RefUNet2DConditionModel.attn_processors
      • RefUNet2DConditionModel.disable_freeu()
      • RefUNet2DConditionModel.enable_freeu()
      • RefUNet2DConditionModel.forward()
      • RefUNet2DConditionModel.set_attention_slice()
      • RefUNet2DConditionModel.set_attn_processor()
      • RefUNet2DConditionModel.set_default_attn_processor()
    • UNet2DConditionOutput
      • UNet2DConditionOutput.ref_features
      • UNet2DConditionOutput.sample
  • sketchkit.colorization.manga_ninja.manganinja.models.transformer_2d module
    • Transformer2DModel
      • Transformer2DModel._set_gradient_checkpointing()
      • Transformer2DModel._supports_gradient_checkpointing
      • Transformer2DModel.forward()
    • Transformer2DModelOutput
      • Transformer2DModelOutput.ref_feature
      • Transformer2DModelOutput.sample
  • sketchkit.colorization.manga_ninja.manganinja.models.unet_2d_blocks module
    • AutoencoderTinyBlock
      • AutoencoderTinyBlock.forward()
    • CrossAttnDownBlock2D
      • CrossAttnDownBlock2D.forward()
    • CrossAttnUpBlock2D
      • CrossAttnUpBlock2D.forward()
    • DownBlock2D
      • DownBlock2D.forward()
    • UNetMidBlock2D
      • UNetMidBlock2D.forward()
    • UNetMidBlock2DCrossAttn
      • UNetMidBlock2DCrossAttn.forward()
    • UpBlock2D
      • UpBlock2D.forward()
    • get_down_block()
    • get_mid_block()
    • get_up_block()
  • sketchkit.colorization.manga_ninja.manganinja.models.unet_2d_condition module
    • UNet2DConditionModel
      • UNet2DConditionModel._check_config()
      • UNet2DConditionModel._set_add_embedding()
      • UNet2DConditionModel._set_class_embedding()
      • UNet2DConditionModel._set_encoder_hid_proj()
      • UNet2DConditionModel._set_gradient_checkpointing()
      • UNet2DConditionModel._set_pos_net_if_use_gligen()
      • UNet2DConditionModel._set_time_proj()
      • UNet2DConditionModel._supports_gradient_checkpointing
      • UNet2DConditionModel.attn_processors
      • UNet2DConditionModel.disable_freeu()
      • UNet2DConditionModel.enable_freeu()
      • UNet2DConditionModel.forward()
      • UNet2DConditionModel.fuse_qkv_projections()
      • UNet2DConditionModel.get_aug_embed()
      • UNet2DConditionModel.get_class_embed()
      • UNet2DConditionModel.get_time_embed()
      • UNet2DConditionModel.process_encoder_hidden_states()
      • UNet2DConditionModel.set_attention_slice()
      • UNet2DConditionModel.set_attn_processor()
      • UNet2DConditionModel.set_default_attn_processor()
      • UNet2DConditionModel.unfuse_qkv_projections()
      • UNet2DConditionModel.unload_lora()
    • UNet2DConditionOutput
      • UNet2DConditionOutput.sample
  • Module contents
• sketchkit.colorization.manga_ninja.manganinja.utils package
  • Submodules
  • sketchkit.colorization.manga_ninja.manganinja.utils.image_util module
    • chw2hwc()
    • resize_max_res()
  • Module contents

Submodules

sketchkit.colorization.manga_ninja.manganinja.pipeline module

class sketchkit.colorization.manga_ninja.manganinja.pipeline.MangaNinjiaPipeline(reference_unet: RefUNet2DConditionModel, controlnet: ControlNetModel, denoising_unet: UNet2DConditionModel, vae: AutoencoderKL, refnet_tokenizer: CLIPTokenizer, refnet_text_encoder: CLIPTextModel, refnet_image_encoder: CLIPVisionModelWithProjection, controlnet_tokenizer: CLIPTokenizer, controlnet_text_encoder: CLIPTextModel, controlnet_image_encoder: CLIPVisionModelWithProjection, scheduler: DDIMScheduler, point_net: PointNet)

Bases: DiffusionPipeline

__encode_empty_text()

Encode text embedding for empty prompt

decode_RGB(rgb_latent: Tensor) → Tensor

Decode depth latent into depth map.

Parameters:

rgb_latent (torch.Tensor) – Depth latent to be decoded.

Returns:

Decoded depth map.

Return type:

torch.Tensor

encode_RGB(rgb_in: Tensor, generator) → Tensor

Encode RGB image into latent.

Parameters:

rgb_in (torch.Tensor) – Input RGB image to be encoded.

Returns:

Image latent.

Return type:

torch.Tensor

get_timesteps(num_inference_steps, strength, device, denoising_start=None)

rgb_latent_scale_factor = 0.18215

single_infer(is_lineart: bool, ref1: Tensor, raw2: Tensor, edit2: Tensor, num_inference_steps: int, show_pbar: bool, guidance_scale_ref: float, guidance_scale_point: float, refnet_encoder_hidden_states: Tensor, controlnet_encoder_hidden_states: Tensor, reference_control_writer: ReferenceAttentionControl, reference_control_reader: ReferenceAttentionControl, preprocessor, generator, point_ref, point_main)

class sketchkit.colorization.manga_ninja.manganinja.pipeline.MangaNinjiaPipelineOutput

Bases: BaseOutput

img_np: ndarray

img_pil: Image

to_save_dict: dict

sketchkit.colorization.manga_ninja.manganinja.point_network module

class sketchkit.colorization.manga_ninja.manganinja.point_network.PointNet(conditioning_channels: int = 1, out_channels: Tuple[int] = (320, 640, 1280, 1280), downsamples: Tuple[int] = (6, 2, 2, 2))

Bases: ModelMixin

forward(x)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

Module contents