With 
DragNUWA 1.5, a team with links to Microsoft has presented a new AI model that can be used to create videos from still images. In contrast to previous approaches, DragNUWA "simply" allows arrows to be drawn in the image, which should provide the AI with information on how and where an object should move in the video.
In DragNUWA 1.0, an additional text prompt could still be formulated, which could even ensure that objects not previously in the image subsequently moved into the image. This prompt does not seem to be available in the interface of the current version 1.5. At least on Hugging Face, where a demo version available is provided.
DragNUWA 1.5 Turns motion paths in a photo into a video
Interestingly, DragNUWA 1.5 does not use its own model, but is based on the weights of Stable Diffusion, which was tuned to the new motion paths by the programmers. If you are interested in the details, you can find the corresponding paper here as a PDF at arxiv.org.
DragNUWA 1.5 Turns motion paths in a photo into a video
If you click through RAW GIFs, you will quickly see that DragNUWA 1.5 does not bring any new solutions for the typical AI movement artifacts into play. However, it does provide a nice impression of how the interface to an AI animation tool could work in the near future.
DragNUWA 1.5 Turns motion paths in a photo into a video
;)
;)
;)