Method __call__

FixModel.py:217–306

(
        self,
        prompt,
        negative_prompt="",
        input_image=None,
        input_video=None,
        denoising_strength=1.0,
        seed=None,
        rand_device="cpu",
        height=1024,
        width=1024,
        cam_id=0,
        num_frames=81,
        cfg_scale=5.0,
        num_inference_steps=50,
        sigma_shift=5.0,
        tiled=True,
        tile_size=(30, 52),
        tile_stride=(15, 26),
        tea_cache_l1_thresh=None,
        tea_cache_model_id="",
        progress_bar_cmd=tqdm,
        progress_bar_st=None,
    )

    @torch.no_grad()
    def __call__(
        self,
        prompt,
        negative_prompt="",
        input_image=None,
        input_video=None,
        denoising_strength=1.0,
        seed=None,
        rand_device="cpu",
        height=1024,
        width=1024,
        cam_id=0,
        num_frames=81,
        cfg_scale=5.0,
        num_inference_steps=50,
        sigma_shift=5.0,
        tiled=True,
        tile_size=(30, 52),
        tile_stride=(15, 26),
        tea_cache_l1_thresh=None,
        tea_cache_model_id="",
        progress_bar_cmd=tqdm,
        progress_bar_st=None,
    ):
        # Parameter check
        height, width = self.check_resize_height_width(height, width)
        if num_frames % 4 != 1:
            num_frames = (num_frames + 2) // 4 * 4 + 1
            print(f"Only `num_frames % 4 != 1` is acceptable. We round it up to {num_frames}.")
        
        # Tiler parameters
        tiler_kwargs = {"tiled": tiled, "tile_size": tile_size, "tile_stride": tile_stride}

        # Scheduler
        self.scheduler.set_timesteps(num_inference_steps, denoising_strength=denoising_strength, shift=sigma_shift)

        # Initialize noise
        noise = self.generate_noise((1, 16, (num_frames - 1) // 4 + 1, height//8, width//8), seed=seed, device=rand_device, dtype=torch.float32)
        noise = noise.to(dtype=self.torch_dtype, device=self.device)
        latents = noise

        # Encode prompts
        self.load_models_to_device(["text_encoder"])
        prompt_emb_posi = self.encode_prompt(prompt, positive=True)
        if cfg_scale != 1.0:
            prompt_emb_nega = self.encode_prompt(negative_prompt, positive=False)
            
        # Encode image
        # import pdb; pdb.set_trace()
        if input_image is not None and self.image_encoder is not None:
            self.load_models_to_device(["image_encoder", "vae"])
            # inter_frame = []
            # for frame_id in [2,4,6]:
            #     inter_frame.append(input_video[frame_id])

            image_emb = self.encode_image_condition(input_image, input_video, num_frames, height, width, cam_id)
        else:
            image_emb = {}