Skip to content

pipelines

Classes

fastvideo.configs.pipelines.Cosmos25Config dataclass 

Cosmos25Config(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 0.0, flow_shift: float = 5.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = (lambda: Cosmos25VideoConfig(arch_config=Cosmos25ArchConfig(num_attention_heads=16, attention_head_dim=128, in_channels=16, out_channels=16, num_layers=28, patch_size=[1, 2, 2], max_size=[128, 240, 240], rope_scale=[1.0, 3.0, 3.0], text_embed_dim=1024, mlp_ratio=4.0, adaln_lora_dim=256, use_adaln_lora=True, concat_padding_mask=True, extra_pos_embed_type=None, use_crossattn_projection=True, rope_enable_fps_modulation=False, qk_norm='rms_norm')))(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Cosmos25VAEConfig(), vae_precision: str = 'bf16', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Reason1Config(arch_config=Reason1ArchConfig(embedding_concat_strategy='full_concat')),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (_identity_preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (reason1_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Bases: PipelineConfig

Configuration for Cosmos 2.5 (Predict2.5) video generation pipeline.

fastvideo.configs.pipelines.CosmosConfig dataclass 

CosmosConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: int = 6, flow_shift: float = 1.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = CosmosVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = CosmosVAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5LargeConfig(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_large_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, conditioning_strategy: str = 'frame_replace', min_num_conditional_frames: int = 1, max_num_conditional_frames: int = 2, sigma_conditional: float = 0.0001, sigma_data: float = 1.0, state_ch: int = 16, state_t: int = 24, text_encoder_class: str = 'T5')

Bases: PipelineConfig

Configuration for Cosmos2 Video2World pipeline matching diffusers.

fastvideo.configs.pipelines.FastHunyuanConfig dataclass 

FastHunyuanConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: int = 6, flow_shift: int = 17, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = HunyuanVAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (LlamaConfig(), CLIPTextConfig()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp16', 'fp16'))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (llama_preprocess_text, clip_preprocess_text))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], tensor], ...] = (lambda: (llama_postprocess_text, clip_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Bases: HunyuanConfig

Configuration specifically optimized for FastHunyuan weights.

fastvideo.configs.pipelines.HYWorldConfig dataclass 

HYWorldConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 5, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HYWorldConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Hunyuan15VAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = SiglipVisionConfig(), image_encoder_precision: str = 'fp16', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Qwen2_5_VLConfig(), T5Config()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp16', 'fp32'))(), preprocess_text_funcs: tuple[Callable[[Any], Any], ...] = (lambda: (qwen_preprocess_text, byt5_preprocess_text))(), postprocess_text_funcs: tuple[Callable[..., Any], ...] = (lambda: (qwen_postprocess_text, byt5_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, text_encoder_crop_start: int = PROMPT_TEMPLATE_TOKEN_LENGTH, text_encoder_max_lengths: tuple[int, ...] = (lambda: (1000 + PROMPT_TEMPLATE_TOKEN_LENGTH, 256))())

Bases: Hunyuan15T2V480PConfig

Base configuration for HYWorld pipeline architecture.

fastvideo.configs.pipelines.Hunyuan15T2V480PConfig dataclass 

Hunyuan15T2V480PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 5, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideo15Config(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Hunyuan15VAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Qwen2_5_VLConfig(), T5Config()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16', 'fp32'))(), preprocess_text_funcs: tuple[Callable[[Any], Any], ...] = (lambda: (qwen_preprocess_text, byt5_preprocess_text))(), postprocess_text_funcs: tuple[Callable[..., Any], ...] = (lambda: (qwen_postprocess_text, byt5_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, text_encoder_crop_start: int = PROMPT_TEMPLATE_TOKEN_LENGTH, text_encoder_max_lengths: tuple[int, ...] = (lambda: (1000 + PROMPT_TEMPLATE_TOKEN_LENGTH, 256))())

Bases: PipelineConfig

Base configuration for HunYuan pipeline architecture.

fastvideo.configs.pipelines.Hunyuan15T2V720PConfig dataclass 

Hunyuan15T2V720PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 9, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideo15Config(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Hunyuan15VAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Qwen2_5_VLConfig(), T5Config()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16', 'fp32'))(), preprocess_text_funcs: tuple[Callable[[Any], Any], ...] = (lambda: (qwen_preprocess_text, byt5_preprocess_text))(), postprocess_text_funcs: tuple[Callable[..., Any], ...] = (lambda: (qwen_postprocess_text, byt5_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, text_encoder_crop_start: int = PROMPT_TEMPLATE_TOKEN_LENGTH, text_encoder_max_lengths: tuple[int, ...] = (lambda: (1000 + PROMPT_TEMPLATE_TOKEN_LENGTH, 256))())

Bases: Hunyuan15T2V480PConfig

Base configuration for HunYuan pipeline architecture.

fastvideo.configs.pipelines.HunyuanConfig dataclass 

HunyuanConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: int = 6, flow_shift: int = 7, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = HunyuanVAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (LlamaConfig(), CLIPTextConfig()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp16', 'fp16'))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (llama_preprocess_text, clip_preprocess_text))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], tensor], ...] = (lambda: (llama_postprocess_text, clip_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Bases: PipelineConfig

Base configuration for HunYuan pipeline architecture.

fastvideo.configs.pipelines.HunyuanGameCraftPipelineConfig dataclass 

HunyuanGameCraftPipelineConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 5, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanGameCraftConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = GameCraftVAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (LlamaConfig(), CLIPTextConfig()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp16', 'fp16'))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (llama_preprocess_text, clip_preprocess_text))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (llama_postprocess_text, clip_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Bases: PipelineConfig

Configuration for HunyuanGameCraft pipeline.

Inherits text encoding from HunyuanVideo but uses: - GameCraft DiT with CameraNet - Same VAE (HunyuanVAE) - Same text encoders (LLaMA + CLIP)

fastvideo.configs.pipelines.LTX2T2VConfig dataclass 

LTX2T2VConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = None, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = LTX2VideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = LTX2VAEConfig(), vae_precision: str = 'bf16', vae_tiling: bool = True, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (LTX2GemmaConfig(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (ltx2_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, audio_decoder_config: ModelConfig = LTX2AudioDecoderConfig(), vocoder_config: ModelConfig = LTX2VocoderConfig(), audio_decoder_precision: str = 'bf16', vocoder_precision: str = 'bf16')

Bases: PipelineConfig

Configuration for LTX-2 T2V pipeline.

fastvideo.configs.pipelines.PipelineConfig dataclass 

PipelineConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = None, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = DiTConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = VAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (EncoderConfig(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], tensor], ...] = (lambda: (postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Base configuration for all pipeline architectures.

Functions

fastvideo.configs.pipelines.PipelineConfig.from_kwargs classmethod 
from_kwargs(kwargs: dict[str, Any], config_cli_prefix: str = '') -> PipelineConfig

Load PipelineConfig from kwargs Dictionary. kwargs: dictionary of kwargs config_cli_prefix: prefix of CLI arguments for this PipelineConfig instance

Source code in fastvideo/configs/pipelines/base.py 
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
@classmethod
def from_kwargs(cls, kwargs: dict[str, Any], config_cli_prefix: str = "") -> "PipelineConfig":
    """
    Load PipelineConfig from kwargs Dictionary.
    kwargs: dictionary of kwargs
    config_cli_prefix: prefix of CLI arguments for this PipelineConfig instance
    """
    from fastvideo.registry import get_pipeline_config_cls_from_name

    prefix_with_dot = f"{config_cli_prefix}." if (config_cli_prefix.strip() != "") else ""
    model_path: str | None = kwargs.get(prefix_with_dot + 'model_path', None) or kwargs.get('model_path')
    pipeline_config_or_path: str | PipelineConfig | dict[str, Any] | None = kwargs.get(
        prefix_with_dot + 'pipeline_config', None) or kwargs.get('pipeline_config')
    if model_path is None:
        raise ValueError("model_path is required in kwargs")

    # 1. Get the pipeline config class from the registry
    pipeline_config_cls = get_pipeline_config_cls_from_name(model_path)

    # 2. Instantiate PipelineConfig
    if pipeline_config_cls is None:
        logger.warning("Couldn't find pipeline config for %s. Using the default pipeline config.", model_path)
        pipeline_config = cls()
    else:
        pipeline_config = pipeline_config_cls()

    # 3. Load PipelineConfig from a json file or a PipelineConfig object if provided
    if isinstance(pipeline_config_or_path, str):
        pipeline_config.load_from_json(pipeline_config_or_path)
        kwargs[prefix_with_dot + 'pipeline_config_path'] = pipeline_config_or_path
    elif isinstance(pipeline_config_or_path, PipelineConfig):
        pipeline_config = pipeline_config_or_path
    elif isinstance(pipeline_config_or_path, dict):
        pipeline_config.update_pipeline_config(pipeline_config_or_path)

    # 4. Update PipelineConfig from CLI arguments if provided
    kwargs[prefix_with_dot + 'model_path'] = model_path
    pipeline_config.update_config_from_dict(kwargs, config_cli_prefix)

    return pipeline_config
fastvideo.configs.pipelines.PipelineConfig.from_pretrained classmethod 
from_pretrained(model_path: str) -> PipelineConfig

use the pipeline class setting from model_path to match the pipeline config

Source code in fastvideo/configs/pipelines/base.py 
188
189
190
191
192
193
194
195
196
@classmethod
def from_pretrained(cls, model_path: str) -> "PipelineConfig":
    """
    use the pipeline class setting from model_path to match the pipeline config
    """
    from fastvideo.registry import get_pipeline_config_cls_from_name
    pipeline_config_cls = get_pipeline_config_cls_from_name(model_path)

    return cast(PipelineConfig, pipeline_config_cls(model_path=model_path))

fastvideo.configs.pipelines.WanI2V480PConfig dataclass 

WanI2V480PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 3.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = CLIPVisionConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: WanT2V480PConfig

Base configuration for Wan I2V 14B 480P pipeline architecture.

fastvideo.configs.pipelines.WanI2V720PConfig dataclass 

WanI2V720PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 5.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = CLIPVisionConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: WanI2V480PConfig

Base configuration for Wan I2V 14B 720P pipeline architecture.

fastvideo.configs.pipelines.WanT2V480PConfig dataclass 

WanT2V480PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 3.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: PipelineConfig

Base configuration for Wan T2V 1.3B pipeline architecture.

fastvideo.configs.pipelines.WanT2V720PConfig dataclass 

WanT2V720PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 5.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: WanT2V480PConfig

Base configuration for Wan T2V 14B 720P pipeline architecture.

Modules

fastvideo.configs.pipelines.base

Classes

fastvideo.configs.pipelines.base.PipelineConfig dataclass 
PipelineConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = None, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = DiTConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = VAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (EncoderConfig(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], tensor], ...] = (lambda: (postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Base configuration for all pipeline architectures.

Functions
fastvideo.configs.pipelines.base.PipelineConfig.from_kwargs classmethod 
from_kwargs(kwargs: dict[str, Any], config_cli_prefix: str = '') -> PipelineConfig

Load PipelineConfig from kwargs Dictionary. kwargs: dictionary of kwargs config_cli_prefix: prefix of CLI arguments for this PipelineConfig instance

Source code in fastvideo/configs/pipelines/base.py 
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
@classmethod
def from_kwargs(cls, kwargs: dict[str, Any], config_cli_prefix: str = "") -> "PipelineConfig":
    """
    Load PipelineConfig from kwargs Dictionary.
    kwargs: dictionary of kwargs
    config_cli_prefix: prefix of CLI arguments for this PipelineConfig instance
    """
    from fastvideo.registry import get_pipeline_config_cls_from_name

    prefix_with_dot = f"{config_cli_prefix}." if (config_cli_prefix.strip() != "") else ""
    model_path: str | None = kwargs.get(prefix_with_dot + 'model_path', None) or kwargs.get('model_path')
    pipeline_config_or_path: str | PipelineConfig | dict[str, Any] | None = kwargs.get(
        prefix_with_dot + 'pipeline_config', None) or kwargs.get('pipeline_config')
    if model_path is None:
        raise ValueError("model_path is required in kwargs")

    # 1. Get the pipeline config class from the registry
    pipeline_config_cls = get_pipeline_config_cls_from_name(model_path)

    # 2. Instantiate PipelineConfig
    if pipeline_config_cls is None:
        logger.warning("Couldn't find pipeline config for %s. Using the default pipeline config.", model_path)
        pipeline_config = cls()
    else:
        pipeline_config = pipeline_config_cls()

    # 3. Load PipelineConfig from a json file or a PipelineConfig object if provided
    if isinstance(pipeline_config_or_path, str):
        pipeline_config.load_from_json(pipeline_config_or_path)
        kwargs[prefix_with_dot + 'pipeline_config_path'] = pipeline_config_or_path
    elif isinstance(pipeline_config_or_path, PipelineConfig):
        pipeline_config = pipeline_config_or_path
    elif isinstance(pipeline_config_or_path, dict):
        pipeline_config.update_pipeline_config(pipeline_config_or_path)

    # 4. Update PipelineConfig from CLI arguments if provided
    kwargs[prefix_with_dot + 'model_path'] = model_path
    pipeline_config.update_config_from_dict(kwargs, config_cli_prefix)

    return pipeline_config
fastvideo.configs.pipelines.base.PipelineConfig.from_pretrained classmethod 
from_pretrained(model_path: str) -> PipelineConfig

use the pipeline class setting from model_path to match the pipeline config

Source code in fastvideo/configs/pipelines/base.py 
188
189
190
191
192
193
194
195
196
@classmethod
def from_pretrained(cls, model_path: str) -> "PipelineConfig":
    """
    use the pipeline class setting from model_path to match the pipeline config
    """
    from fastvideo.registry import get_pipeline_config_cls_from_name
    pipeline_config_cls = get_pipeline_config_cls_from_name(model_path)

    return cast(PipelineConfig, pipeline_config_cls(model_path=model_path))

Functions

fastvideo.configs.pipelines.base.parse_int_list 
parse_int_list(value: str) -> list[int]

Parse a comma-separated string of integers into a list.

Source code in fastvideo/configs/pipelines/base.py 
312
313
314
315
316
def parse_int_list(value: str) -> list[int]:
    """Parse a comma-separated string of integers into a list."""
    if not value:
        return []
    return [int(x.strip()) for x in value.split(",")]

fastvideo.configs.pipelines.cosmos

Classes

fastvideo.configs.pipelines.cosmos.CosmosConfig dataclass 
CosmosConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: int = 6, flow_shift: float = 1.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = CosmosVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = CosmosVAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5LargeConfig(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_large_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, conditioning_strategy: str = 'frame_replace', min_num_conditional_frames: int = 1, max_num_conditional_frames: int = 2, sigma_conditional: float = 0.0001, sigma_data: float = 1.0, state_ch: int = 16, state_t: int = 24, text_encoder_class: str = 'T5')

Bases: PipelineConfig

Configuration for Cosmos2 Video2World pipeline matching diffusers.

Functions

fastvideo.configs.pipelines.cosmos.t5_large_postprocess_text 
t5_large_postprocess_text(outputs: BaseEncoderOutput) -> Tensor

Postprocess T5 Large text encoder outputs for Cosmos pipeline.

Return raw last_hidden_state without truncation/padding.

Source code in fastvideo/configs/pipelines/cosmos.py 
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
def t5_large_postprocess_text(outputs: BaseEncoderOutput) -> torch.Tensor:
    """Postprocess T5 Large text encoder outputs for Cosmos pipeline.

    Return raw last_hidden_state without truncation/padding.
    """
    hidden_state = outputs.last_hidden_state

    if hidden_state is None:
        raise ValueError("T5 Large outputs missing last_hidden_state")

    nan_count = torch.isnan(hidden_state).sum()
    if nan_count > 0:
        hidden_state = hidden_state.masked_fill(torch.isnan(hidden_state), 0.0)

    # Zero out embeddings beyond actual sequence length
    if outputs.attention_mask is not None:
        attention_mask = outputs.attention_mask
        lengths = attention_mask.sum(dim=1).cpu()
        for i, length in enumerate(lengths):
            hidden_state[i, length:] = 0

    return hidden_state

fastvideo.configs.pipelines.cosmos2_5

Classes

fastvideo.configs.pipelines.cosmos2_5.Cosmos25Config dataclass 
Cosmos25Config(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 0.0, flow_shift: float = 5.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = (lambda: Cosmos25VideoConfig(arch_config=Cosmos25ArchConfig(num_attention_heads=16, attention_head_dim=128, in_channels=16, out_channels=16, num_layers=28, patch_size=[1, 2, 2], max_size=[128, 240, 240], rope_scale=[1.0, 3.0, 3.0], text_embed_dim=1024, mlp_ratio=4.0, adaln_lora_dim=256, use_adaln_lora=True, concat_padding_mask=True, extra_pos_embed_type=None, use_crossattn_projection=True, rope_enable_fps_modulation=False, qk_norm='rms_norm')))(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Cosmos25VAEConfig(), vae_precision: str = 'bf16', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Reason1Config(arch_config=Reason1ArchConfig(embedding_concat_strategy='full_concat')),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (_identity_preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (reason1_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Bases: PipelineConfig

Configuration for Cosmos 2.5 (Predict2.5) video generation pipeline.

fastvideo.configs.pipelines.gen3c

Classes

fastvideo.configs.pipelines.gen3c.Gen3CConfig dataclass 
Gen3CConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: int = 6, flow_shift: float = 1.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = Gen3CVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Gen3CVAEConfig(), vae_precision: str = 'bf16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (_Gen3CT5LargeConfig(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_large_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, conditioning_strategy: str = 'frame_replace', min_num_conditional_frames: int = 1, max_num_conditional_frames: int = 2, sigma_conditional: float = 0.001, sigma_data: float = 0.5, state_ch: int = 16, state_t: int = 16, text_encoder_class: str = 'T5', frame_buffer_max: int = 2, noise_aug_strength: float = 0.0, filter_points_threshold: float = 0.05, use_moge_depth: bool = True, moge_model_name: str = 'Ruicheng/moge-vitl', offload_moge_after_depth: bool = True, default_trajectory_type: str = 'left', default_movement_distance: float = 0.3, default_camera_rotation: str = 'center_facing', video_resolution: tuple[int, int] = (704, 1280), num_frames: int = 121, fps: int = 24, cfg_behavior: str = 'legacy', default_negative_prompt: str = 'The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', autoregressive_chunk_frames: int = 121, autoregressive_overlap_frames: int = 1)

Bases: PipelineConfig

Configuration for GEN3C Video Generation Pipeline.

GEN3C extends Cosmos with 3D cache for camera-controlled video generation. Key parameters: - frame_buffer_max: Number of 3D cache buffers (default: 2) - noise_aug_strength: Strength of noise augmentation per buffer - filter_points_threshold: Threshold for filtering unreliable depth points

fastvideo.configs.pipelines.gen3c.Gen3CInferenceConfig dataclass 
Gen3CInferenceConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: int = 6, flow_shift: float = 1.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = Gen3CVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Gen3CVAEConfig(), vae_precision: str = 'bf16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (_Gen3CT5LargeConfig(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_large_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, conditioning_strategy: str = 'frame_replace', min_num_conditional_frames: int = 1, max_num_conditional_frames: int = 2, sigma_conditional: float = 0.001, sigma_data: float = 0.5, state_ch: int = 16, state_t: int = 16, text_encoder_class: str = 'T5', frame_buffer_max: int = 2, noise_aug_strength: float = 0.0, filter_points_threshold: float = 0.05, use_moge_depth: bool = True, moge_model_name: str = 'Ruicheng/moge-vitl', offload_moge_after_depth: bool = True, default_trajectory_type: str = 'left', default_movement_distance: float = 0.3, default_camera_rotation: str = 'center_facing', video_resolution: tuple[int, int] = (704, 1280), num_frames: int = 121, fps: int = 24, cfg_behavior: str = 'legacy', default_negative_prompt: str = 'The video captures a series of frames showing ugly scenes, static with no motion, motion blur, over-saturation, shaky footage, low resolution, grainy texture, pixelated images, poorly lit areas, underexposed and overexposed scenes, poor color balance, washed out colors, choppy sequences, jerky movements, low frame rate, artifacting, color banding, unnatural transitions, outdated special effects, fake elements, unconvincing visuals, poorly edited content, jump cuts, visual noise, and flickering. Overall, the video is of poor quality.', autoregressive_chunk_frames: int = 121, autoregressive_overlap_frames: int = 1, batch_size: int = 1, gradient_checkpointing: bool = False, guidance_scale: float = 1.0, num_inference_steps: int = 35)

Bases: Gen3CConfig

Configuration for GEN3C inference with optimized defaults.

Functions

fastvideo.configs.pipelines.gen3c.t5_large_postprocess_text 
t5_large_postprocess_text(outputs: BaseEncoderOutput) -> Tensor

Postprocess T5 Large text encoder outputs for GEN3C pipeline.

Return raw last_hidden_state without truncation/padding.

Source code in fastvideo/configs/pipelines/gen3c.py 
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
def t5_large_postprocess_text(outputs: BaseEncoderOutput) -> torch.Tensor:
    """Postprocess T5 Large text encoder outputs for GEN3C pipeline.

    Return raw last_hidden_state without truncation/padding.
    """
    hidden_state = outputs.last_hidden_state

    if hidden_state is None:
        raise ValueError("T5 Large outputs missing last_hidden_state")

    nan_count = torch.isnan(hidden_state).sum()
    if nan_count > 0:
        hidden_state = hidden_state.masked_fill(torch.isnan(hidden_state), 0.0)

    # Zero out embeddings beyond actual sequence length (vectorized)
    if outputs.attention_mask is not None:
        attention_mask = outputs.attention_mask
        lengths = attention_mask.sum(dim=1)
        max_len = hidden_state.shape[1]
        mask = torch.arange(max_len, device=hidden_state.device)[None, :] >= lengths[:, None]
        hidden_state[mask] = 0.0

    return hidden_state

fastvideo.configs.pipelines.hunyuan

Classes

fastvideo.configs.pipelines.hunyuan.FastHunyuanConfig dataclass 
FastHunyuanConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: int = 6, flow_shift: int = 17, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = HunyuanVAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (LlamaConfig(), CLIPTextConfig()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp16', 'fp16'))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (llama_preprocess_text, clip_preprocess_text))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], tensor], ...] = (lambda: (llama_postprocess_text, clip_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Bases: HunyuanConfig

Configuration specifically optimized for FastHunyuan weights.

fastvideo.configs.pipelines.hunyuan.HunyuanConfig dataclass 
HunyuanConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: int = 6, flow_shift: int = 7, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = HunyuanVAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (LlamaConfig(), CLIPTextConfig()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp16', 'fp16'))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (llama_preprocess_text, clip_preprocess_text))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], tensor], ...] = (lambda: (llama_postprocess_text, clip_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Bases: PipelineConfig

Base configuration for HunYuan pipeline architecture.

fastvideo.configs.pipelines.hunyuan15

Classes

fastvideo.configs.pipelines.hunyuan15.Hunyuan15I2V720PConfig dataclass 
Hunyuan15I2V720PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 7, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideo15Config(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Hunyuan15VAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Qwen2_5_VLConfig(), T5Config()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16', 'fp32'))(), preprocess_text_funcs: tuple[Callable[[Any], Any], ...] = (lambda: (qwen_preprocess_text, byt5_preprocess_text))(), postprocess_text_funcs: tuple[Callable[..., Any], ...] = (lambda: (qwen_postprocess_text, byt5_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, text_encoder_crop_start: int = PROMPT_TEMPLATE_TOKEN_LENGTH, text_encoder_max_lengths: tuple[int, ...] = (lambda: (1000 + PROMPT_TEMPLATE_TOKEN_LENGTH, 256))())

Bases: Hunyuan15T2V720PConfig

Base configuration for HunYuan pipeline architecture.

fastvideo.configs.pipelines.hunyuan15.Hunyuan15SR1080PConfig dataclass 
Hunyuan15SR1080PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 7, flow_shift_sr: int = 2, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideo15Config(), dit_precision: str = 'bf16', upsampler_config: tuple[UpsamplerConfig, ...] = (lambda: (SRTo720pUpsamplerConfig(), SRTo1080pUpsamplerConfig()))(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Hunyuan15VAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Qwen2_5_VLConfig(), T5Config()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16', 'fp32'))(), preprocess_text_funcs: tuple[Callable[[Any], Any], ...] = (lambda: (qwen_preprocess_text, byt5_preprocess_text))(), postprocess_text_funcs: tuple[Callable[..., Any], ...] = (lambda: (qwen_postprocess_text, byt5_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, text_encoder_crop_start: int = PROMPT_TEMPLATE_TOKEN_LENGTH, text_encoder_max_lengths: tuple[int, ...] = (lambda: (1000 + PROMPT_TEMPLATE_TOKEN_LENGTH, 256))())

Bases: Hunyuan15T2V720PConfig

Base configuration for HunYuan pipeline architecture.

fastvideo.configs.pipelines.hunyuan15.Hunyuan15T2V480PConfig dataclass 
Hunyuan15T2V480PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 5, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideo15Config(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Hunyuan15VAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Qwen2_5_VLConfig(), T5Config()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16', 'fp32'))(), preprocess_text_funcs: tuple[Callable[[Any], Any], ...] = (lambda: (qwen_preprocess_text, byt5_preprocess_text))(), postprocess_text_funcs: tuple[Callable[..., Any], ...] = (lambda: (qwen_postprocess_text, byt5_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, text_encoder_crop_start: int = PROMPT_TEMPLATE_TOKEN_LENGTH, text_encoder_max_lengths: tuple[int, ...] = (lambda: (1000 + PROMPT_TEMPLATE_TOKEN_LENGTH, 256))())

Bases: PipelineConfig

Base configuration for HunYuan pipeline architecture.

fastvideo.configs.pipelines.hunyuan15.Hunyuan15T2V720PConfig dataclass 
Hunyuan15T2V720PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 9, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanVideo15Config(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Hunyuan15VAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Qwen2_5_VLConfig(), T5Config()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16', 'fp32'))(), preprocess_text_funcs: tuple[Callable[[Any], Any], ...] = (lambda: (qwen_preprocess_text, byt5_preprocess_text))(), postprocess_text_funcs: tuple[Callable[..., Any], ...] = (lambda: (qwen_postprocess_text, byt5_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, text_encoder_crop_start: int = PROMPT_TEMPLATE_TOKEN_LENGTH, text_encoder_max_lengths: tuple[int, ...] = (lambda: (1000 + PROMPT_TEMPLATE_TOKEN_LENGTH, 256))())

Bases: Hunyuan15T2V480PConfig

Base configuration for HunYuan pipeline architecture.

Functions

fastvideo.configs.pipelines.hunyuan15.extract_glyph_texts 
extract_glyph_texts(prompt: str) -> str | None

Extract glyph texts from prompt using regex pattern.

Parameters:

Name Type Description Default
prompt str

Input prompt string

 required

Returns:

Type Description
str | None

List of extracted glyph texts
Source code in fastvideo/configs/pipelines/hunyuan15.py 
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
def extract_glyph_texts(prompt: str) -> str | None:
    """
    Extract glyph texts from prompt using regex pattern.

    Args:
        prompt: Input prompt string

    Returns:
        List of extracted glyph texts
    """
    pattern = r"\"(.*?)\"|“(.*?)”"
    matches = re.findall(pattern, prompt)
    result = [match[0] or match[1] for match in matches]
    result = list(dict.fromkeys(result)) if len(result) > 1 else result

    formatted_result = ". ".join([f'Text "{text}"' for text in result]) + ". " if result else None

    return formatted_result
fastvideo.configs.pipelines.hunyuan15.format_text_input 
format_text_input(prompt: str, system_message: str) -> list[dict[str, Any]]

Apply text to template.

Parameters:

Name Type Description Default
prompt List[str]

Input text.

 required
system_message str

System message.

 required

Returns:

Type Description
list[dict[str, Any]]

List[Dict[str, Any]]: List of chat conversation.
Source code in fastvideo/configs/pipelines/hunyuan15.py 
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
def format_text_input(prompt: str, system_message: str) -> list[dict[str, Any]]:
    """
    Apply text to template.

    Args:
        prompt (List[str]): Input text.
        system_message (str): System message.

    Returns:
        List[Dict[str, Any]]: List of chat conversation.
    """

    template = [{"role": "system", "content": system_message}, {"role": "user", "content": prompt if prompt else " "}]

    return template

fastvideo.configs.pipelines.hunyuangamecraft

Pipeline configuration for HunyuanGameCraft.

HunyuanGameCraft extends HunyuanVideo with: 1. CameraNet for camera/action conditioning (Plücker coordinates) 2. Mask-based conditioning for autoregressive generation 3. 33 input channels (16 latent + 16 gt_latent + 1 mask)

Text encoders are the same as HunyuanVideo: - LLaVA-LLaMA-3-8B for primary text encoding (4096 dim) - CLIP ViT-L/14 for secondary pooled embeddings (768 dim)

Classes

fastvideo.configs.pipelines.hunyuangamecraft.HunyuanGameCraftPipelineConfig dataclass 
HunyuanGameCraftPipelineConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 5, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HunyuanGameCraftConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = GameCraftVAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (LlamaConfig(), CLIPTextConfig()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp16', 'fp16'))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (llama_preprocess_text, clip_preprocess_text))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (llama_postprocess_text, clip_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None)

Bases: PipelineConfig

Configuration for HunyuanGameCraft pipeline.

Inherits text encoding from HunyuanVideo but uses: - GameCraft DiT with CameraNet - Same VAE (HunyuanVAE) - Same text encoders (LLaMA + CLIP)

fastvideo.configs.pipelines.hyworld

Classes

fastvideo.configs.pipelines.hyworld.HYWorldConfig dataclass 
HYWorldConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: int = 5, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = HYWorldConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = Hunyuan15VAEConfig(), vae_precision: str = 'fp16', vae_tiling: bool = True, vae_sp: bool = True, image_encoder_config: EncoderConfig = SiglipVisionConfig(), image_encoder_precision: str = 'fp16', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (Qwen2_5_VLConfig(), T5Config()))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp16', 'fp32'))(), preprocess_text_funcs: tuple[Callable[[Any], Any], ...] = (lambda: (qwen_preprocess_text, byt5_preprocess_text))(), postprocess_text_funcs: tuple[Callable[..., Any], ...] = (lambda: (qwen_postprocess_text, byt5_postprocess_text))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, text_encoder_crop_start: int = PROMPT_TEMPLATE_TOKEN_LENGTH, text_encoder_max_lengths: tuple[int, ...] = (lambda: (1000 + PROMPT_TEMPLATE_TOKEN_LENGTH, 256))())

Bases: Hunyuan15T2V480PConfig

Base configuration for HYWorld pipeline architecture.

fastvideo.configs.pipelines.longcat

Classes

fastvideo.configs.pipelines.longcat.LongCatDiTArchConfig dataclass 
LongCatDiTArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = list(), _fsdp_shard_conditions: list = list(), _compile_conditions: list = list(), param_names_mapping: dict = dict(), reverse_param_names_mapping: dict = dict(), lora_param_names_mapping: dict = dict(), _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (SAGE_ATTN, FLASH_ATTN, TORCH_SDPA, VIDEO_SPARSE_ATTN, VMOBA_ATTN, SAGE_ATTN_THREE, SLA_ATTN, SAGE_SLA_ATTN), hidden_size: int = 0, num_attention_heads: int = 0, num_channels_latents: int = 0, in_channels: int = 16, out_channels: int = 16, exclude_lora_layers: list[str] = list(), boundary_ratio: float | None = None, adaln_tembed_dim: int = 512, caption_channels: int = 4096, depth: int = 48, enable_bsa: bool = False, enable_flashattn3: bool = False, enable_flashattn2: bool = True, enable_xformers: bool = False, frequency_embedding_size: int = 256, mlp_ratio: int = 4, num_heads: int = 32, text_tokens_zero_pad: bool = True, patch_size: list[int] = (lambda: [1, 2, 2])(), cp_split_hw: list[int] | None = None, bsa_params: dict | None = None)

Bases: DiTArchConfig

Extended DiTArchConfig with LongCat-specific fields.

fastvideo.configs.pipelines.longcat.LongCatT2V480PConfig dataclass 
LongCatT2V480PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = None, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = (lambda: DiTConfig(arch_config=LongCatDiTArchConfig()))(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'bf16', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[T5Config, ...] = (lambda: (LongCatT5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (longcat_preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (umt5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, enable_kv_cache: bool = True, offload_kv_cache: bool = False, enable_bsa: bool = False, use_distill: bool = False, enhance_hf: bool = False, bsa_params: dict | None = None, bsa_sparsity: float | None = None, bsa_cdf_threshold: float | None = None, bsa_chunk_q: list[int] | None = None, bsa_chunk_k: list[int] | None = None, t_thresh: float | None = None)

Bases: PipelineConfig

Configuration for LongCat pipeline (480p).

Components expected by loaders
  • tokenizer: AutoTokenizer
  • text_encoder: UMT5EncoderModel
  • transformer: LongCatTransformer3DModel
  • vae: AutoencoderKLWan (Wan VAE, 4x8 compression)
  • scheduler: FlowMatchEulerDiscreteScheduler
fastvideo.configs.pipelines.longcat.LongCatT2V704PConfig dataclass 
LongCatT2V704PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = None, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = (lambda: DiTConfig(arch_config=LongCatDiTArchConfig()))(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'bf16', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[T5Config, ...] = (lambda: (LongCatT5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('bf16',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (longcat_preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (umt5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, enable_kv_cache: bool = True, offload_kv_cache: bool = False, enable_bsa: bool = True, use_distill: bool = False, enhance_hf: bool = False, bsa_params: dict | None = None, bsa_sparsity: float | None = None, bsa_cdf_threshold: float | None = None, bsa_chunk_q: list[int] | None = None, bsa_chunk_k: list[int] | None = None, t_thresh: float | None = None)

Bases: LongCatT2V480PConfig

Configuration for LongCat pipeline (704p) with BSA enabled by default.

Uses the same resolution and BSA parameters as original LongCat refinement stage. BSA parameters configured in transformer config.json with chunk_3d_shape=[4,4,4]: - Input: 704×1280×96 - VAE (8x): 88×160×96
- Patch [1,2,2]: 44×80×96 - chunk [4,4,4]: 96%4=0, 44%4=0, 80%4=0 ✅

This configuration matches the original LongCat refinement stage parameters.

fastvideo.configs.pipelines.longcat.LongCatT5ArchConfig dataclass 
LongCatT5ArchConfig(stacked_params_mapping: list[tuple[str, str, str]] = (lambda: [('.qkv_proj', '.q', 'q'), ('.qkv_proj', '.k', 'k'), ('.qkv_proj', '.v', 'v')])(), architectures: list[str] = (lambda: [])(), _supported_attention_backends: tuple[AttentionBackendEnum, ...] = (FLASH_ATTN, TORCH_SDPA), output_hidden_states: bool = False, use_return_dict: bool = True, vocab_size: int = 32128, hidden_size: int = 0, num_hidden_layers: int = 0, num_attention_heads: int = 0, pad_token_id: int = 0, eos_token_id: int = 1, text_len: int = 512, hidden_state_skip_layer: int = 0, decoder_start_token_id: int = 0, output_past: bool = True, scalable_attention: bool = True, tie_word_embeddings: bool = False, tokenizer_kwargs: dict[str, Any] = dict(), _fsdp_shard_conditions: list = (lambda: [_is_transformer_layer, _is_embeddings, _is_final_layernorm])(), d_model: int = 512, d_kv: int = 64, d_ff: int = 2048, num_layers: int = 6, num_decoder_layers: int | None = None, num_heads: int = 8, relative_attention_num_buckets: int = 32, relative_attention_max_distance: int = 128, dropout_rate: float = 0.1, layer_norm_epsilon: float = 1e-06, initializer_factor: float = 1.0, feed_forward_proj: str = 'relu', dense_act_fn: str = '', is_gated_act: bool = False, is_encoder_decoder: bool = True, use_cache: bool = True, classifier_dropout: float = 0.0, dtype: str | None = None, gradient_checkpointing: bool = False)

Bases: T5ArchConfig

T5 arch that pads tokenizer output to max_length.

LongCat's denoising stage concatenates positive and negative attention masks along the batch dimension for CFG, which requires uniform seq length. The shared :class:T5ArchConfig dropped the "padding": "max_length" tokenizer kwarg so other DiTs could run with variable-length masks; LongCat still needs the uniform contract.

Functions

fastvideo.configs.pipelines.longcat.longcat_preprocess_text 
longcat_preprocess_text(prompt: str) -> str

Clean and preprocess text like original LongCat implementation.

This function applies the same text cleaning pipeline as the original LongCat-Video implementation to ensure identical tokenization results.

Steps: 1. basic_clean: Fix unicode issues and unescape HTML entities 2. whitespace_clean: Normalize whitespace to single spaces

Parameters:

Name Type Description Default
prompt str

Raw input text prompt

 required

Returns:

Type Description
str

Cleaned and normalized text prompt
Source code in fastvideo/configs/pipelines/longcat.py 
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
def longcat_preprocess_text(prompt: str) -> str:
    """Clean and preprocess text like original LongCat implementation.

    This function applies the same text cleaning pipeline as the original
    LongCat-Video implementation to ensure identical tokenization results.

    Steps:
    1. basic_clean: Fix unicode issues and unescape HTML entities
    2. whitespace_clean: Normalize whitespace to single spaces

    Args:
        prompt: Raw input text prompt

    Returns:
        Cleaned and normalized text prompt
    """
    # basic_clean: fix unicode and HTML entities
    text = ftfy.fix_text(prompt)
    text = html.unescape(html.unescape(text))
    text = text.strip()

    # whitespace_clean: normalize whitespace
    text = re.sub(r"\s+", " ", text)
    text = text.strip()

    return text
fastvideo.configs.pipelines.longcat.umt5_postprocess_text 
umt5_postprocess_text(outputs: BaseEncoderOutput) -> Tensor

Postprocess UMT5/T5 encoder outputs to fixed length 512 embeddings.

Source code in fastvideo/configs/pipelines/longcat.py 
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
def umt5_postprocess_text(outputs: BaseEncoderOutput) -> torch.Tensor:
    """
    Postprocess UMT5/T5 encoder outputs to fixed length 512 embeddings.
    """
    mask: torch.Tensor = outputs.attention_mask
    hidden_state: torch.Tensor = outputs.last_hidden_state
    seq_lens = mask.gt(0).sum(dim=1).long()
    assert torch.isnan(hidden_state).sum() == 0
    prompt_embeds = [u[:v] for u, v in zip(hidden_state, seq_lens, strict=True)]
    prompt_embeds_tensor: torch.Tensor = torch.stack(
        [torch.cat([u, u.new_zeros(512 - u.size(0), u.size(1))]) for u in prompt_embeds], dim=0)
    return prompt_embeds_tensor

fastvideo.configs.pipelines.turbodiffusion

TurboDiffusion pipeline configurations.

TurboDiffusion uses RCM (recurrent Consistency Model) scheduler with SLA (Sparse-Linear Attention) for fast 1-4 step video generation.

Classes

fastvideo.configs.pipelines.turbodiffusion.TurboDiffusionI2VConfig dataclass 
TurboDiffusionI2VConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 5.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = CLIPVisionConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = 0.9, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: PipelineConfig

Base configuration for TurboDiffusion I2V pipeline.

Uses RCM scheduler with sigma_max=200 for 1-4 step generation. Uses boundary_ratio=0.9 for high-noise to low-noise model switching.

fastvideo.configs.pipelines.turbodiffusion.TurboDiffusionI2V_A14B_Config dataclass 
TurboDiffusionI2V_A14B_Config(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 5.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = CLIPVisionConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = 0.9, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: TurboDiffusionI2VConfig

Configuration for TurboDiffusion I2V A14B model.

fastvideo.configs.pipelines.turbodiffusion.TurboDiffusionT2VConfig dataclass 
TurboDiffusionT2VConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 3.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: PipelineConfig

Base configuration for TurboDiffusion T2V pipeline.

Uses RCM scheduler with sigma_max=80 for 1-4 step generation. No boundary_ratio (single model, no switching).

fastvideo.configs.pipelines.turbodiffusion.TurboDiffusionT2V_14B_Config dataclass 
TurboDiffusionT2V_14B_Config(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 5.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: TurboDiffusionT2VConfig

Configuration for TurboDiffusion T2V 14B model.

Uses same config as 1.3B but with higher flow_shift for 14B model.

fastvideo.configs.pipelines.turbodiffusion.TurboDiffusionT2V_1_3B_Config dataclass 
TurboDiffusionT2V_1_3B_Config(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 3.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: TurboDiffusionT2VConfig

Configuration for TurboDiffusion T2V 1.3B model.

fastvideo.configs.pipelines.wan

Classes

fastvideo.configs.pipelines.wan.FastWan2_1_T2V_480P_Config dataclass 
FastWan2_1_T2V_480P_Config(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 8.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = (lambda: [1000, 757, 522])(), ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: WanT2V480PConfig

Base configuration for FastWan T2V 1.3B 480P pipeline architecture with DMD

fastvideo.configs.pipelines.wan.WANV2VConfig dataclass 
WANV2VConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 3.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = WAN2_1ControlCLIPVisionConfig(), image_encoder_precision: str = 'bf16', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: WanI2V480PConfig

Configuration for WAN2.1 1.3B Control pipeline.

fastvideo.configs.pipelines.wan.WanI2V480PConfig dataclass 
WanI2V480PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 3.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = CLIPVisionConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: WanT2V480PConfig

Base configuration for Wan I2V 14B 480P pipeline architecture.

fastvideo.configs.pipelines.wan.WanI2V720PConfig dataclass 
WanI2V720PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 5.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = CLIPVisionConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: WanI2V480PConfig

Base configuration for Wan I2V 14B 720P pipeline architecture.

fastvideo.configs.pipelines.wan.WanT2V480PConfig dataclass 
WanT2V480PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 3.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: PipelineConfig

Base configuration for Wan T2V 1.3B pipeline architecture.

fastvideo.configs.pipelines.wan.WanT2V720PConfig dataclass 
WanT2V720PConfig(model_path: str = '', pipeline_config_path: str | None = None, embedded_cfg_scale: float = 6.0, flow_shift: float | None = 5.0, flow_shift_sr: float | None = None, disable_autocast: bool = False, is_causal: bool = False, dit_config: DiTConfig = WanVideoConfig(), dit_precision: str = 'bf16', upsampler_config: UpsamplerConfig = UpsamplerConfig(), upsampler_precision: str = 'fp32', vae_config: VAEConfig = WanVAEConfig(), vae_precision: str = 'fp32', vae_tiling: bool = False, vae_sp: bool = False, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', text_encoder_configs: tuple[EncoderConfig, ...] = (lambda: (T5Config(),))(), text_encoder_precisions: tuple[str, ...] = (lambda: ('fp32',))(), preprocess_text_funcs: tuple[Callable[[str], str], ...] = (lambda: (preprocess_text,))(), postprocess_text_funcs: tuple[Callable[[BaseEncoderOutput], Tensor], ...] = (lambda: (t5_postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, boundary_ratio: float | None = None, precision: str = 'bf16', warp_denoising_step: bool = True)

Bases: WanT2V480PConfig

Base configuration for Wan T2V 14B 720P pipeline architecture.