Diffusers How to use FastVideo/Pika-H3-Flex-NVFP4-QAD with Diffusers:
pip install -U diffusers transformers accelerate
import torch
from diffusers import DiffusionPipeline
from diffusers.utils import load_image, export_to_video
# switch to "mps" for apple devices
pipe = DiffusionPipeline.from_pretrained("FastVideo/Pika-H3-Flex-NVFP4-QAD", dtype=torch.bfloat16, device_map="cuda")
pipe.to("cuda")
prompt = "A man with short gray hair plays a red electric guitar."
image = load_image(
"https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/diffusers/guitar-man.png"
)
output = pipe(image=image, prompt=prompt).frames[0]
export_to_video(output, "output.mp4") Pika H3 Flex How to use FastVideo/Pika-H3-Flex-NVFP4-QAD with Pika H3 Flex:
# pip install -r examples/requirements-diffusers.txt (see the examples/ folder of this repo)
import torch
from diffusers import ComponentsManager, ModularPipeline
from diffusers.utils.export_utils import encode_video
from PIL import Image
workflow = "fl2va" # "fl2va" (first-frame to video+audio) or "ref2va" (reference images/video to video+audio)
manager = ComponentsManager()
manager.enable_auto_cpu_offload(device="cuda:0")
pipe = ModularPipeline.from_pretrained("FastVideo/Pika-H3-Flex-NVFP4-QAD", workflow=workflow, components_manager=manager)
pipe.load_components(
pretrained_model_name_or_path="FastVideo/Pika-H3-Flex-NVFP4-QAD",
dtype={"vae": torch.float32, "audio_vae": torch.float32, "default": torch.bfloat16},
)
# The bundled base components are not Pika-adapted weights: load the matching adapter.
pipe.load_lora_weights("FastVideo/Pika-H3-Flex-NVFP4-QAD", weight_name=f"{workflow}/adapter.safetensors")
result = pipe(
prompt="A cat walks through a sunlit cafe and jumps onto the counter",
image=Image.open("first_frame.png").convert("RGB"),
height=480,
width=832,
num_frames=121,
generator=torch.Generator("cuda:0").manual_seed(0),
output=["videos", "audio", "sampling_rate"],
)
encode_video(
result["videos"][0], fps=24, output_path="output.mp4",
audio=result["audio"][0].float().cpu(), audio_sample_rate=result["sampling_rate"],
) 
