Can we use the released EMU3-Gen (image model) to do video generation?

[sbyebss]

Hi! Authors, thank you for your great work! I wonder is it possible to use this released emu3-gen model to do image2video generation? Can we twist the existing inference code to do image2video? It sounds doable because emu3-gen also involves video data during pre-training, and you already released the tokenizer. Do you mind commenting on the feasibility of doing so and the quality of the generated video? Thank you for your help!