mmagic.models.editors.animatediff.animatediff_utils

Module Contents

Functions

shave_segments(path[, n_shave_prefix_segments])	Removes segments.
renew_resnet_paths(old_list[, n_shave_prefix_segments])	Updates paths inside resnets to the new naming scheme (local
renew_vae_resnet_paths(old_list[, n_shave_prefix_segments])	Updates paths inside resnets to the new naming scheme (local
renew_attention_paths(old_list[, n_shave_prefix_segments])	Updates paths inside attentions to the new naming scheme (local
renew_vae_attention_paths(old_list[, ...])	Updates paths inside attentions to the new naming scheme (local
assign_to_checkpoint(paths, checkpoint, old_checkpoint)	This does the final conversion step: take locally converted weights and
conv_attn_to_linear(checkpoint)
create_unet_diffusers_config(original_config, image_size)	Creates a config for the diffusers based on the config of the LDM
create_vae_diffusers_config(original_config, image_size)	Creates a config for the diffusers based on the config of the LDM
convert_ldm_unet_checkpoint(checkpoint, config[, ...])	Takes a state dict and a config, and returns a converted checkpoint.
convert_ldm_vae_checkpoint(checkpoint, config)
convert_ldm_clip_checkpoint(checkpoint)
convert_paint_by_example_checkpoint(checkpoint)
convert_open_clip_checkpoint(checkpoint)
stable_unclip_image_encoder(original_config)	Returns the image processor and clip image encoder for the img2img
stable_unclip_image_noising_components(original_config)	Returns the noising components for the img2img and txt2img unclip
save_videos_grid(videos, path[, rescale, n_rows, fps])

Attributes

textenc_conversion_lst
textenc_conversion_map
textenc_transformer_conversion_lst
protected
textenc_pattern

mmagic.models.editors.animatediff.animatediff_utils.shave_segments(path, n_shave_prefix_segments=1)

Removes segments.

Positive values shave the first segments, negative shave the last segments.

mmagic.models.editors.animatediff.animatediff_utils.renew_resnet_paths(old_list, n_shave_prefix_segments=0)

Updates paths inside resnets to the new naming scheme (local renaming)

mmagic.models.editors.animatediff.animatediff_utils.renew_vae_resnet_paths(old_list, n_shave_prefix_segments=0)

Updates paths inside resnets to the new naming scheme (local renaming)

mmagic.models.editors.animatediff.animatediff_utils.renew_attention_paths(old_list, n_shave_prefix_segments=0)

Updates paths inside attentions to the new naming scheme (local renaming)

mmagic.models.editors.animatediff.animatediff_utils.renew_vae_attention_paths(old_list, n_shave_prefix_segments=0)

Updates paths inside attentions to the new naming scheme (local renaming)

mmagic.models.editors.animatediff.animatediff_utils.assign_to_checkpoint(paths, checkpoint, old_checkpoint, attention_paths_to_split=None, additional_replacements=None, config=None)

This does the final conversion step: take locally converted weights and apply a global renaming to them. It splits attention layers, and takes into account additional replacements that may arise.

Assigns the weights to the new checkpoint.

mmagic.models.editors.animatediff.animatediff_utils.conv_attn_to_linear(checkpoint)

mmagic.models.editors.animatediff.animatediff_utils.create_unet_diffusers_config(original_config, image_size: int, controlnet=False)

Creates a config for the diffusers based on the config of the LDM model.

mmagic.models.editors.animatediff.animatediff_utils.create_vae_diffusers_config(original_config, image_size: int)

Creates a config for the diffusers based on the config of the LDM model.

mmagic.models.editors.animatediff.animatediff_utils.convert_ldm_unet_checkpoint(checkpoint, config, path=None, extract_ema=False, controlnet=False)

Takes a state dict and a config, and returns a converted checkpoint.

mmagic.models.editors.animatediff.animatediff_utils.convert_ldm_vae_checkpoint(checkpoint, config)

mmagic.models.editors.animatediff.animatediff_utils.convert_ldm_clip_checkpoint(checkpoint)

mmagic.models.editors.animatediff.animatediff_utils.textenc_conversion_lst = [('cond_stage_model.model.positional_embedding',...

mmagic.models.editors.animatediff.animatediff_utils.textenc_conversion_map

mmagic.models.editors.animatediff.animatediff_utils.textenc_transformer_conversion_lst = [('resblocks.', 'text_model.encoder.layers.'), ('ln_1', 'layer_norm1'), ('ln_2', 'layer_norm2'),...

mmagic.models.editors.animatediff.animatediff_utils.protected

mmagic.models.editors.animatediff.animatediff_utils.textenc_pattern

mmagic.models.editors.animatediff.animatediff_utils.convert_paint_by_example_checkpoint(checkpoint)

mmagic.models.editors.animatediff.animatediff_utils.convert_open_clip_checkpoint(checkpoint)

mmagic.models.editors.animatediff.animatediff_utils.stable_unclip_image_encoder(original_config)

Returns the image processor and clip image encoder for the img2img unclip pipeline.

We currently know of two types of stable unclip models which separately use the clip and the openclip image encoders.

mmagic.models.editors.animatediff.animatediff_utils.stable_unclip_image_noising_components(original_config, clip_stats_path: Optional[str] = None, device: Optional[str] = None)

Returns the noising components for the img2img and txt2img unclip pipelines.

Converts the stability noise augmentor into 1. a StableUnCLIPImageNormalizer for holding the CLIP stats 2. a DDPMScheduler for holding the noise schedule

If the noise augmentor config specifies a clip stats path, the clip_stats_path must be provided.

mmagic.models.editors.animatediff.animatediff_utils.save_videos_grid(videos: torch.Tensor, path: str, rescale=False, n_rows=6, fps=8)