ai-forever / kandinskyvideo Goto Github PK

View Code? Open in Web Editor NEW

152.0 14.0 20.0 396 MB

KandinskyVideo — multilingual end-to-end text2video latent diffusion model

License: Apache License 2.0

Python 100.00%

kandinsky latent-diffusion text-to-video video-generation

kandinskyvideo's Issues

open source

Thank you for your outstanding work. Will the training code be open sourced?

What are the recommended hardware specifications for inference?

How much VRAM need for inference?
And can you recommend minimum specific GPU for generating videos?
Thanks for open-sourcing this!

Any plan to release the training code?

RuntimeError: size of tensors must match except in dimsion 1

from video_kandinsky3 import get_T2V_pipeline
import os
os.environ["PYTORCH_CUDA_ALLOC_CONF"] = "max_split_size_mb:8024"

t2v_pipe = get_T2V_pipeline('cuda', fp16=True)

pfps = 'medium' # ['low', 'medium', 'high']
video = t2v_pipe(
'a red car is drifting on the mountain road, close view, fast movement',
width=640, height=384, fps=fps
)

RuntimeError: size of tensors must match except in dimsion 1. Expected size 2 but got size 3 for tensor number 1 in the list

How to save result to an mp4 file?

After your example code...

from video_kandinsky3 import get_T2V_pipeline

t2v_pipe = get_T2V_pipeline('cuda', fp16=True)

pfps = 'medium' # ['low', 'medium', 'high']
video = t2v_pipe(
    'a red car is drifting on the mountain road, close view, fast movement',
    width=640, height=384, fps=fps
)

how do I save the video to an mp4 file?

Also what is the recommended fps to use?

Recommend Projects