
Posted by Caren Chang – Developer Relations Engineer
The Jetpack Media3 library permits Android apps to construct prime quality media apps. As a part of the Media3 library, the Transformer module goals to supply simple to make use of, dependable, and performant APIs for transcoding and enhancing media.
For instance, apps can use Transformer to use enhancing operations corresponding to trimming an extended piece of media file, or making use of results to video tracks. Transformer will also be used to transform media recordsdata from one format to a different, corresponding to adjusting the decision or encoding of the media file.
Creating Transformer APIs
As a part of the method to introduce new APIs, our engineering group works carefully with Google apps corresponding to Google Pictures to check and experiment the brand new APIs. Experimental flags are first launched to allow efficiency enhancements. As soon as the outcomes are profitable and conclusive, these experimental options are then constructed into the default API implementations or promoted to public APIs for all apps to make use of. This strategy permits Transformer APIs to be examined on all kinds of units.
Transformer Adoption in apps
Apps which were utilizing Transformer in manufacturing noticed in-app efficiency enhancements, much less code to keep up, and higher developer expertise. Let’s take a better have a look at how Transformer has helped apps for his or her media-editing use instances.
One among customers’ favourite options in Google Pictures is reminiscence sharing, the place snippets of your life story which are curated and introduced as Google Pictures recollections can now be shared as movies to social media and chat apps. Nevertheless, the method of mixing media objects to create a video on machine is useful resource intensive and topic to important latency, particularly on low-end units. To cut back this latency and allow the function on a wider vary of units, Pictures adopted Transformer of their media creation pipeline. Together with different enhancements made, the group discovered that Transformer performed a component in decreasing the median person latency for creating reminiscence movies by 41% on high-end units and 27% on mid-range units.
The Pictures app additionally permits customers to carry out media edits corresponding to trimming or rotating a video. By adopting Transformer APIs for rotating movies, median save latency was decreased by 79% for relevant movies. The app additionally adopted Transformer’s API for optimizing video trimming, and noticed video save latency lower by 64%.
1 Second On a regular basis is a private video journal that helps you create charming montages and timelapses. One of many app’s foremost person journeys is sequentially combining brief movies to create a significant film. After adopting Transformer for this use case, the app noticed that video encoding efficiency was as much as 5x sooner, permitting them to discover enabling 4k and HDR assist. The Transformer adoption additionally helped lower related code by 30%, making it simpler for the builders to keep up the code base.
BandLab is the next-generation music creation platform utilized by tens of millions world wide to make and share their music. The app initially used MediaCodecs for his or her video creation use instances, however discovered that the low degree implementation resulted in native crashes that had been tough to debug. After researching extra on Transformer, the group made the choice emigrate from MediaCodecs to Transformer. Total, it solely took the group 12 working days for the migration, and this resulted in a less complicated codebase and extra maintainable pipeline for his or her media creation use instances. As well as, the app noticed that each one beforehand noticed native crashes had been not occurring anymore.
What’s subsequent for Transformers?
We’re excited to see Transformer’s adoption within the developer group, and can proceed including new options to assist extra media-editing use instances for the Android ecosystem together with:
- Higher assist for previewing media edits
- Bettering the efficiency and developer expertise for video body extraction
- Simpler integration with AI results
- and far more
Hold an eye fixed out on what we’re engaged on within the Media3 Github, and file function requests to assist form the way forward for Transformer!