Changelog
Bring Your Own Audio (BYOA) | Replica & Video API
May 2024
We are excited to release this highly requested feature — Bring Your Own Audio (BYOA). Developers on all plans can now use any audio source, including natural audio recordings, to generate videos – offering more control over the video creation process. Previously, generating a video required using our standard text-to-speech service to convert text scripts into audio. With BYOA, you can use audio that is in-house generated, from a preferred vendor, or directly-recorded natural audio.
How to Use BYOA:
BYOA is integrated with our /videos
endpoint. Instead of submitting a script
field, you can now provide an audio_url
field with a link to an accessible S3 URL or other hosted audio file.
The video is then generated based on the provided audio.
Benefits of BYOA:
- Customization and Experimentation: Developers can experiment with different audio sources—from in-house audio to preferred vendors. This flexibility helps in understanding how variations in audio affect video generation
- Enhanced User Experience: Offer multiple audio options for your end-users, allowing them to select the preferred audio for their final video.
- Authenticity and Control: End-users can record natural audio to personalize their replicas, controlling tone and expression while still benefiting from eliminating concerns about physical appearance or camera setup.
See how it works in this demo
See how to generate videos from an audio file in our documentation.
Enhanced Error Messaging for Training Videos | Replica API
April 2024
We’ve updated the API to include detailed error messages in the response when a replica training fails. This allows you to directly surface these messages to your end users. They can now understand the reasons for a training failure and resubmit a corrected video without needing to contact support for troubleshooting.
Examples of training video errors, include:
- More than one face detected or obstructions present
- Video does not meet minimum duration requirement
- The video is not encoded using h.264, ensure it’s an mp4 file encoded using h.264
Auto QA for Training Videos | Replica API
April 2024
Improved Success Rates and Training Efficiency: We’ve optimized the replica training process by automatically selecting the longest stable segment from a user’s training video. This segment is used to train the replica, focusing on a minimum viable length rather than the entire video.
Benefits include:
- Higher success rates in replica training
- Reduced training duration
- Enhanced quality of replicas by excluding awkward movements in generated videos