FEAT: support HunyuanVideo

[qinxuye]

I ran on a 48GB GPU. Loading model with CPU offload, or OOM will occur even for 320x512 resolusion.

Loading model:

xinference launch --model-name HunyuanVideo --model-type video --cpu_offload True

Inference:

from xinference.client import Client

client = Client("http://127.0.0.1:9997")
model = client.get_model("HunyuanVideo")

input_text = "A cat walks on the grass, realistic"

resp = model.text_to_video(input_text, height=320, width=512, num_frames=61, num_inference_steps=30)

b64_video = resp['data'][0]['b64_json']
video_bytes = base64.b64decode(b64_video)
with open('1.mp4', 'wb') as f:
    f.write(video_bytes)

Result:

1.mp4

[codingl2k1]

LGTM