We had to create a service where manager could upload a lot of text stories for videos and the system itself would generate and publish videos on Youtube.
The client provided premium access to video and photo stocks.
We created a service, that consists of several parts:
- Administrative panel interface for managers;
- Backend service for collecting data from video and photo stocks and for synchronization with the server where the After Effect works;
- Backend service of publishing videos on Youtube
- Backend service for working with openai.com for word processing.
Below you can watch a video of one of the early versions of the project with instructions on how to create tasks for video generation:
We worked out a technical solution on how to work with text in order to more accurately determine the meaning of the plot.
How does it all work? In short, the manager loads the text of the story, the system identifies short semantic sections of the text using machine learning, and based on them, it searches for video and photo materials on stocks.
Then it forms a timeline and distributes materials. After the finished timeline and files are sent for rendering.
At each step, the manager can intervene in the process to manually edit the result of automatic work.
After the release of the main version, an update was released, where it was integrated into the Google Text to Speech service (https://cloud.google.com/text-to-speech).
🔉 So that the videos have voice acting.
While working on the project, we came up with 3 types of task interface for the administrative panel, one of the options went into work without editing.
For text analysis using machine learning, we took a solution from OpenAI, the service is based on the GPT-3 algorithm, so for text it was a great option for this project.
We created and published the created services on the client’s server. The execution took several months.
We wrote a user manual for the manager how to use the tool. We continue to work in support mode.