Method predict

predict.py:159–340

Run a single prediction on the model

(
        self,
        prompt: str = Input(
            description="Prompt for video generation.",
            default="a Corgi walking in the park at sunrise, oil painting style",
        ),
        sample_method: str = Input(
            description="Choose a scheduler for sampling base video output.",
            default="ddpm",
            choices=["ddim", "eulerdiscrete", "ddpm"],
        ),
        width: int = Input(description="Width of output video.", default=512),
        height: int = Input(description="Height of output video", default=320),
        num_inference_steps: int = Input(
            description="Number of denoising steps", default=50
        ),
        guidance_scale: float = Input(
            description="Scale for classifier-free guidance", default=7.0
        ),
        quality: int = Input(
            description="Quality of the output vide0", le=10, ge=0, default=9
        ),
        seed: int = Input(
            description="Random seed. Leave blank to randomize the seed", default=None
        ),
        interpolation: bool = Input(
            description="Default output has 16 frames. Set interpolation to True to get 61 frames output.",
            default=False,
        ),
        super_resolution: bool = Input(
            description="Super resolution 4x when set to True.", default=False
        ),
        video_fps: int = Input(
            description="Number of frames per second in the output video", default=8
        ),
    )

        self.sr_pipeline = self.sr_pipeline.to(self.device)

    def predict(
        self,
        prompt: str = Input(
            description="Prompt for video generation.",
            default="a Corgi walking in the park at sunrise, oil painting style",
        ),
        sample_method: str = Input(
            description="Choose a scheduler for sampling base video output.",
            default="ddpm",
            choices=["ddim", "eulerdiscrete", "ddpm"],
        ),
        width: int = Input(description="Width of output video.", default=512),
        height: int = Input(description="Height of output video", default=320),
        num_inference_steps: int = Input(
            description="Number of denoising steps", default=50
        ),
        guidance_scale: float = Input(
            description="Scale for classifier-free guidance", default=7.0
        ),
        quality: int = Input(
            description="Quality of the output vide0", le=10, ge=0, default=9
        ),
        seed: int = Input(
            description="Random seed. Leave blank to randomize the seed", default=None
        ),
        interpolation: bool = Input(
            description="Default output has 16 frames. Set interpolation to True to get 61 frames output.",
            default=False,
        ),
        super_resolution: bool = Input(
            description="Super resolution 4x when set to True.", default=False
        ),
        video_fps: int = Input(
            description="Number of frames per second in the output video", default=8
        ),
    ) -> Path:
        """Run a single prediction on the model"""
        if seed is None:
            seed = int.from_bytes(os.urandom(2), "big")
        print(f"Using seed: {seed}")

        temp_output_dir = "temp_output"
        if os.path.exists(temp_output_dir):
            shutil.rmtree(temp_output_dir)
        os.makedirs(temp_output_dir)

        torch.manual_seed(seed)
        torch.cuda.manual_seed(seed)
        scheduler = self.schedulers[sample_method]

        videogen_pipeline = VideoGenPipeline(
            vae=self.vae,
            text_encoder=self.text_encoder_one,
            tokenizer=self.tokenizer_one,
            scheduler=scheduler,
            unet=self.unet,
        ).to(self.device)
        videogen_pipeline.enable_xformers_memory_efficient_attention()