How to choose between T2V and I2V?

[born-to-change]

Thank you for your excellent work! It appears that I2V and I2V are two different works, T2V originates from Modelscope-T2V, while I2V is from I2VGen-XL. We would like to know which one, T2V or I2V, is more suitable for our video fine-tuning training. If we only have a few tens of thousands of 2K text-video pairs, which model has stronger generalization capabilities? We aim to achieve better results in generating video content from textual descriptions. We look forward to your response and greatly appreciate it!

[Steven-SWZhang]

Thank you for your interest in our work. Depending on your task goals, both T2V and I2V can be used if the purpose is for research. If you want to solve practical problems, I recommend using I2V because the video quality will be better, but this requires you to first generate an image and then generate the video. I hope this helps you.

[HyoKong]

@Steven-SWZhang May I ask if is there any convenient way to obtain the starting image with the resolution of 1280x720? Now most of the t2i model can only generate images with resolution 512x512.

Thanks!