Generation of candidate video elements

1 . A system, comprising:

a processor configured to:

receive a requested video element type for a video creator user;

derive representative data associated with a set of seed videos;

generate a prompt based at least in part on the requested video element type, the representative data associated with the set of seed videos, and a random input, wherein to generate the prompt comprises to:

determine a prompt template corresponding to a large language model that has been customized for the video creator user;

generate the random input corresponding to a random variable included in the prompt template; and

update at least a variable included in the prompt template based at least in part on the representative data associated with the set of seed videos;

input the prompt into the large language model that has been customized for the video creator user; and

present a set of candidate video elements that has been output by the large language model; and

a memory coupled to the processor and configured to provide the processor with instructions.

2 . The system of claim 1 , wherein the processor is further configured to:

receive identifying information associated with a set of videos;

determine respective view counts associated with the set of videos;

determine an average view count based at least in part on the respective view counts; and

determine a subset of overperforming videos by comparing the respective view counts to the average view count, wherein the set of seed videos comprises the subset of overperforming videos.

3 . The system of claim 1 , wherein the processor is further to:

obtain profile information associated with the video creator user;

determine a set of reference videos associated with the video creator user;

derive loglines from the set of reference videos; and

determine a set of cast members from the set of reference videos, wherein the large language model has been trained based at least in part on training data comprising the profile information, the loglines, and the set of cast members.

4 . The system of claim 1 , wherein the representative data associated with the set of seed videos comprises one or more of the following: video titles, beat sheets, loglines, thumbnail images, video themes, and video concepts.

5 . The system of claim 1 , wherein the random input is generated using a cryptographically secure pseudorandom number generator.

6 . The system of claim 1 , wherein the random input is generated using a current location of a cursor.

7 . The system of claim 1 , wherein the processor is further configured to determine a respective creativity score associated with a candidate video element.

8 . The system of claim 7 , wherein the processor is further configured to compare the candidate video element to stored candidate video elements associated with the video creator user.

9 . The system of claim 8 , wherein the processor is configured to determine whether to present the candidate video element based at least in part on the respective creativity score and the comparison.

10 . The system of claim 1 , wherein the processor is further configured to:

receive a user feedback to a candidate video element;

generate a new prompt based at least in part on the candidate video element and the user feedback;

input the new prompt into the large language model that has been customized for the video creator user; and

present a new set of candidate video elements that has been output by the large language model.

11 . The system of claim 10 , wherein the processor is further configured to update the large language model that has been customized for the video creator user based at least in part on the user feedback.

12 . The system of claim 1 , wherein the requested video element type comprises one or more of the following: a video title, a thumbnail image, and a beat sheet.

13 . A method, comprising:

receiving a requested video element type for a video creator user;

deriving representative data associated with a set of seed videos;

generating a prompt based at least in part on the requested video element type, the representative data associated with the set of seed videos, and a random input, wherein generating the prompt comprises:

determining a prompt template corresponding to a large language model that has been customized for the video creator user;

generating the random input corresponding to a random variable included in the prompt template; and

updating at least a variable included in the prompt template based at least in part on the representative data associated with the set of seed videos;

inputting the prompt into the large language model that has been customized for the video creator user; and

presenting a set of candidate video elements that has been output by the large language model.

14 . The method of claim 13 , further comprising:

receiving identifying information associated with a set of videos;

determining respective view counts associated with the set of videos;

determining an average view count based at least in part on the respective view counts; and

determining a subset of overperforming videos by comparing the respective view counts to the average view count, wherein the set of seed videos comprises the subset of overperforming videos.

15 . The method of claim 13 , further comprising:

obtaining profile information associated with the video creator user;

determining a set of reference videos associated with the video creator user;

deriving loglines from the set of reference videos; and

determining a set of cast members from the set of reference videos, wherein the large language model has been trained based at least in part on training data comprising the profile information, the loglines, and the set of cast members.

16 . The method of claim 13 , wherein the representative data associated with the set of seed videos comprises one or more of the following: video titles, beat sheets, loglines, thumbnail images, video themes, and video concepts.

17 . The method of claim 13 , wherein the random input is generated using a cryptographically secure pseudorandom number generator.

18 . A computer program product embodied in a non-transitory computer readable medium and comprising computer instructions for:

receiving a requested video element type for a video creator user;

deriving representative data associated with a set of seed videos;

generating a prompt based at least in part on the requested video element type, the representative data associated with the set of seed videos, and a random input, wherein generating the prompt comprises:

determining a prompt template corresponding to a large language model that has been customized for the video creator user;

generating the random input corresponding to a random variable included in the prompt template; and

updating at least a variable included in the prompt template based at least in part on the representative data associated with the set of seed videos;

inputting the prompt into the large language model that has been customized for the video creator user; and

presenting a set of candidate video elements that has been output by the large language model.