MCPcopy Create free account
hub / github.com/OpenImagingLab/FlashVSR / __init__

Method __init__

diffsynth/pipelines/step_video.py:22–30  ·  view source on GitHub ↗
(self, device="cuda", torch_dtype=torch.float16)

Source from the content-addressed store, hash-verified

20class StepVideoPipeline(BasePipeline):
21
22 def __init__(self, device="cuda", torch_dtype=torch.float16):
23 super().__init__(device=device, torch_dtype=torch_dtype)
24 self.scheduler = FlowMatchScheduler(sigma_min=0.0, extra_one_step=True, shift=13.0, reverse_sigmas=True, num_train_timesteps=1)
25 self.prompter = StepVideoPrompter()
26 self.text_encoder_1: HunyuanDiTCLIPTextEncoder = None
27 self.text_encoder_2: STEP1TextEncoder = None
28 self.dit: StepVideoModel = None
29 self.vae: StepVideoVAE = None
30 self.model_names = ['text_encoder_1', 'text_encoder_2', 'dit', 'vae']
31
32
33 def enable_vram_management(self, num_persistent_param_in_dit=None):

Callers

nothing calls this directly

Calls 2

FlowMatchSchedulerClass · 0.85
StepVideoPrompterClass · 0.85

Tested by

no test coverage detected