V1 API

FastVideo's V1 API provides a streamlined interface for video generation tasks with powerful customization options. This page documents the primary components of the API.

Video Generator

This class will be the primary Python API for generating videos and images.

`fastvideo.entrypoints.video_generator.VideoGenerator`

VideoGenerator(fastvideo_args: FastVideoArgs, executor_class: type[Executor], log_stats: bool, *, log_queue = None)

A unified class for generating videos using diffusion models.

This class provides a simple interface for video generation with rich customization options, similar to popular frameworks like HF Diffusers.

Initialize the video generator.

Parameters

fastvideo_args: FastVideoArgs: The inference arguments
executor_class: type[Executor]: The executor class to use for inference
log_stats: bool: Whether to log statistics
log_queue: Optional multiprocessing.Queue to forward worker logs to

`fastvideo.entrypoints.video_generator.VideoGenerator.from_pretrained`

from_pretrained(model_path: str | GeneratorConfig | Mapping[str, Any] | None = None, **kwargs) -> VideoGenerator

Create a video generator from a pretrained model.

Parameters

model_path: str | GeneratorConfig | Mapping[str, Any] | None: Path or identifier for the pretrained model
pipeline_config: Pipeline config to use for inference
**kwargs: Additional arguments to customize model loading, set any FastVideoArgs or PipelineConfig attributes here.

Returns

VideoGenerator: The created video generator

Priority level: Default pipeline config < User's pipeline config < User's kwargs

Stable convenience kwargs remain supported here for common engine and offload settings. Advanced model- or pipeline-specific options should move to VideoGenerator.from_config(...).

VideoGenerator.from_pretrained() should be the primary way of creating a new video generator.

Configuring FastVideo

The following two classes PipelineConfig and SamplingParam are used to configure initialization and sampling parameters, respectively.

PipelineConfig

`fastvideo.configs.pipelines.base.PipelineConfig`

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, scheduler_step_in_fp32: 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_decode_precision: str | None = None, vae_tiling: bool = True, vae_sp: bool = True, output_channels: int = 3, image_encoder_config: EncoderConfig = EncoderConfig(), image_encoder_precision: str = 'fp32', image_encoder_configs: tuple[EncoderConfig, ...] | None = None, image_encoder_precisions: tuple[str, ...] | None = None, 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], torch.tensor], ...] = (lambda: (postprocess_text,))(), dmd_denoising_steps: list[int] | None = None, ti2v_task: bool = False, lucy_edit_task: bool = False, boundary_ratio: float | None = None)

Base configuration for all pipeline architectures.

`fastvideo.configs.pipelines.base.PipelineConfig.from_pretrained`

from_pretrained(model_path: str) -> PipelineConfig

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

`fastvideo.configs.pipelines.base.PipelineConfig.dump_to_json`

dump_to_json(file_path: str)

SamplingParam

`fastvideo.api.sampling_param.SamplingParam`

SamplingParam(data_type: str = 'video', image_path: str | None = None, pil_image: Any | None = None, last_image: Any | None = None, references: list[Any] | None = None, video_path: str | None = None, pose_video_path: str | None = None, face_video_path: str | None = None, background_video_path: str | None = None, mask_video_path: str | None = None, animate_mode: str | None = None, audio_path: str | None = None, latents: Any | None = None, audio_latents: Any | None = None, mouse_cond: Any | None = None, keyboard_cond: Any | None = None, grid_sizes: Any | None = None, pose: str | None = None, prompt_attention_mask: list = list(), negative_attention_mask: list = list(), c2ws_plucker_emb: Any | None = None, action_path: str | None = None, t_thresh: float = 0.5, num_cond_frames: int = 0, prompt: str | list[str] | None = None, negative_prompt: str = 'Bright tones, overexposed, static, blurred details, subtitles, style, works, paintings, images, static, overall gray, worst quality, low quality, JPEG compression residue, ugly, incomplete, extra fingers, poorly drawn hands, poorly drawn faces, deformed, disfigured, misshapen limbs, fused fingers, still picture, messy background, three legs, many people in the background, walking backwards', max_sequence_length: int | None = None, prompt_path: str | None = None, output_path: str = 'outputs/', output_video_name: str | None = None, num_videos_per_prompt: int = 1, seed: int = 1024, num_frames: int = 125, height: int = 720, width: int = 1280, height_sr: int = 1072, width_sr: int = 1920, fps: int = 24, num_inference_steps: int = 50, num_inference_steps_sr: int = 50, guidance_scale: float = 1.0, batch_cfg: bool = False, guidance_scale_2: float | None = None, use_embedded_guidance: bool = False, true_cfg_scale: float = 1.0, guidance_rescale: float = 0.0, boundary_ratio: float | None = None, sigmas: list[float] | None = None, pyramid_num_inference_steps_list: list[int] | None = None, history_sizes: list[int] | None = None, num_latent_frames_per_chunk: int = 9, keep_first_frame: bool = True, is_skip_first_chunk: bool = False, use_zero_init: bool = True, zero_steps: int = 1, is_amplify_first_chunk: bool = False, enable_teacache: bool = False, trajectory_type: str | None = None, movement_distance: float | None = None, camera_rotation: str | None = None, ltx2_cfg_scale_video: float = 1.0, ltx2_cfg_scale_audio: float = 1.0, ltx2_modality_scale_video: float = 1.0, ltx2_modality_scale_audio: float = 1.0, ltx2_rescale_scale: float = 0.0, ltx2_stg_scale_video: float = 0.0, ltx2_stg_scale_audio: float = 0.0, ltx2_stg_blocks_video: list[int] = (lambda: [29])(), ltx2_stg_blocks_audio: list[int] = (lambda: [29])(), ltx2_use_ancestral_sampler: bool = False, ltx2_images: list[tuple[str, int, float]] | None = None, ltx2_image_crf: float = 33.0, ltx2_conditioning_latent_stage1: Any | None = None, ltx2_conditioning_latent_stage2: Any | None = None, ltx2_video_conditions: list[tuple[list[str], int, float]] | None = None, audio_start_in_s: float | None = None, audio_end_in_s: float | None = None, init_audio: Any = None, init_audio_strength: float | None = None, init_noise_level: float | None = None, inpaint_audio: Any = None, inpaint_mask: Any = None, continuation_state: ContinuationState | None = None, return_continuation_state: bool = False, save_video: bool = True, return_frames: bool = True, return_trajectory_latents: bool = False, return_trajectory_decoded: bool = False, *, return_samples: bool = False)

Sampling parameters for video generation.

`fastvideo.api.sampling_param.SamplingParam.from_pretrained`

from_pretrained(model_path: str) -> SamplingParam

Returning samples without previews

Set OutputConfig(return_samples=True, return_frames=False, save_video=False) to receive the decoded CPU tensor in GenerationResult.samples without building preview grids or writing media. Pixel samples retain the existing float [0, 1] contract. return_frames=True continues to return both samples and preview frames. The legacy sampling/keyword interface accepts the same return_samples option.