ComeçarComece de graça

Build a video!

Time for you to have a go at creating a video entirely from a text prompt! You'll use a CogVideoXPipeline pipeline and the following prompt to guide the generation:

A robot doing the robot dance. The dance floor has colorful squares and a glitterball.

Note: Inference on video generation models can take a long time, so we've pre-loaded the generated video for you. Running different prompts will not generated new videos.

The CogVideoXPipeline class has already been imported for you.

Este exercício faz parte do curso

Multi-Modal Models with Hugging Face

Ver curso

Instruções do exercício

  • Create a CogVideoXPipeline from the THUDM/CogVideoX-2b checkpoint.
  • Run the pipeline with the provided prompt, setting the number of inference steps to 20, the number of frames to generate to 20, and the guidance scale to 6.

Exercício interativo prático

Experimente este exercício completando este código de exemplo.

prompt = "A robot doing the robot dance. The dance floor has colorful squares and a glitterball."

# Create a CogVideoXPipeline
pipe = ____(
    "____",
    torch_dtype=torch.float16
)

# Run the pipeline with the provided prompt
video = pipe(
    prompt=____,
    num_inference_steps=____,
    num_frames=____,
    guidance_scale=____
)
video = video.frames[0]

video_path = export_to_video(video, "output.mp4", fps=8)
video = VideoFileClip(video_path)
video.write_gif("video_ex.gif")
Editar e executar o código