智谱CogVideoX系列新开源CogVideoX-5b，视频生成质量更高，视觉效果更好，此前开源的版本为CogVideoX-2B。
GIF有点卡 ...
提示词：Picture this: a sleek, confident cat lounging casually in a sun-drenched room, its fur glistening under the warm rays. But what sets this feline apart is not just its glossy coat or the graceful poise it exudes; it's the pair of stylish sunglasses perched on its nose, adding an air of mystery and coolness to its demeanor. The sunglasses, with their reflective lenses, hide the cat's enigmatic eyes, making it seem as if it's pondering life's mysteries or perhaps just planning its next mischievous adventure. As sunlight filters through the window, casting patterns on the floor, the cat, utterly unfazed by its unusual accessory, gives off a vibe of effortless chic. It sits there, a picture of serenity and detachment, occasionally flicking its tail or letting out a soft purr, completely embodying the essence of cool. This cat doesn't just wear sunglasses; it owns the look, making anyone who glances its way do a double-take, charmed by the sight of such an unexpected yet striking fashion statement.
提示词：In a heartwarming scene, a delightful panda bear finds itself in the gentle embrace of a human, engaging in what can only be described as a whimsical dance. The panda, with its striking black and white fur, looks up with trusting, curious eyes, its round face framed by fuzzy ears. The human, filled with joy and awe, carefully supports the panda's soft, plump body, guiding it in a series of gentle, swaying movements. As they move together, the panda's clumsy yet endearing attempts to mimic the rhythm create a moment of pure magic. Its tiny paws occasionally reach out, touching the human's hands, as if trying to understand this novel form of interaction. Around them, the air is filled with laughter and soft music, enhancing the enchantment of their dance. This unique encounter, a blend of nature's innocence and human affection, unfolds like a tender dance of friendship, leaving an indelible mark of joy and connection on all who witness it.
推理的硬件需求如下：
CogVideoX-2B 模型：
• FP16 精度：
• 使用 diffusers：需要 12.5GB 显存
• INT8 精度：
• 使用 diffusers with torchaudio：需要 7.8GB 显存
CogVideoX-5B 模型：
• BF16 精度：
• 使用 diffusers：需要 20.7GB 显存
• INT8 精度：
• 使用 diffusers with torchaudio：需要 11.4GB 显存
体验界面如下：
快速上手 ?
本模型已经支持使用 Huggingface 的 diffusers 库进行部署，你可以按照以下步骤进行部署。
安装对应的依赖
# diffusers>=0.30.1
# transformers>=0.44.0
# accelerate>=0.33.0 (建议从源代码安装)
# imageio-ffmpeg>=0.5.1
pip install --upgrade transformers accelerate diffusers imageio-ffmpeg运行代码 (BF16 / FP16)
import torch
from diffusers import CogVideoXPipeline
from diffusers.utils import export_to_video
prompt = (
"A panda, dressed in a small, red jacket and a tiny hat, sits on a wooden stool "
"in a serene bamboo forest. The panda's fluffy paws strum a miniature acoustic "
"guitar, producing soft, melodic tunes. Nearby, a few other pandas gather, "
"watching curiously and some clapping in rhythm. Sunlight filters through the tall "
"bamboo, casting a gentle glow on the scene. The panda's face is expressive, showing "
"concentration and joy as it plays. The background includes a small, flowing stream "
"and vibrant green foliage, enhancing the peaceful and magical atmosphere of this "
"unique musical performance."
)
pipe = CogVideoXPipeline.from_pretrained(
"THUDM/CogVideoX-5b",
torch_dtype=torch.bfloat16
)
pipe.enable_model_cpu_offload()
pipe.vae.enable_tiling()
video = pipe(
prompt=prompt,
num_videos_per_prompt=1,
num_inference_steps=50,
num_frames=49,
guidance_scale=6,
generator=torch.Generator(device="cuda").manual_seed(42),
).frames[0]
export_to_video(video, "output.mp4", fps=8)