Generative films

A computer-implemented method comprising:

• obtaining input video data depicting a scene comprising an object;
• obtaining current values of a set of adjustable parameters of an object representation model comprising a first neural network and arranged to generate, using the first neural network, animated representations of the object in which a geometry of the object is controllable by the set of adjustable parameters;
• and
• for a plurality of iterations:
• generating, using the object representation model, a video layer comprising an animated representation of the object in which the geometry of the object corresponds to the current values of the set of adjustable parameters;
• presenting, via a user interface, composite video data comprising the video layer overlaid on the object within the input video data;
• and
• updating, in response to user input via the user interface, the current values of the set of adjustable parameters;
• and
• following the plurality of iterations, generating, using a second neural network, a finalized video layer comprising a finalized animated representation of the object in which a geometry of the object corresponds to the current values of the set of adjustable parameters,
• wherein the second neural network generates the finalized video layer at a resolution which is higher than a resolution of the video layers generated during the plurality of iterations.

One or more non-transitory storage media comprising instructions which, when executed by one or more processors, cause the one or more processors to carry out operations comprising:

• obtaining input video data depicting a scene comprising an object;
• obtaining current values of a set of adjustable parameters of an object representation model comprising a first neural network and arranged to generate, using the first neural network, animated representations of the object in which a geometry of the object is controllable by the set of adjustable parameters;
• and
• for a plurality of iterations:
• generating, using the object representation model, a video layer comprising an animated representation of the object in which the geometry of the object corresponds to the current values of the set of adjustable parameters;
• presenting, via a user interface, composite video data comprising the video layer overlaid on the object in the input video data;
• and
• updating, in response to user input via the user interface, the current values of the set of adjustable parameters;
• and
• following the plurality of iterations, generating, using a second neural network, a finalized video layer comprising a finalized animated representation of the object in which a geometry of the object corresponds to the current values of the set of adjustable parameters,
• wherein the second neural network is arranged to generate the finalized video layer at a resolution which is higher than a resolution of the video layers generated during the plurality of iterations.