The quality has significantly improved in comparison to the previous generation.
The Runway company, which conceived and developed the neural network of Stable Diffusion, demonstrated a second version of artificial intelligence that can generate video by description. The first version of AI, Gen-1, was demonstrated in early February.
The developers spent a month and a half achieving significant progress. Gen-2 (the second generation of artificial intelligence has been loosely named) can produce better videos with less visual glitches.
The neural network can receive both a text description and an image, on the basis of which you need to create a video. In addition, the neural network can style ready-made videos, focusing on the desired frame.
The technology is still being refined. The generated videos retain the characteristic ripples (however, it was worse before), the generation is at a relatively low resolution, and the artifacts can be quite impressive.
Gen-2 was demonstrated the next day after the release of the Chinese neural network Modelscope, which also generates video by description. The quality of Modelscope's generation is clearly superior to Runway's Gen-1, although it's roughly on the same level as Gen-2.
This is an interesting piece of news.
