Devices and local model setup

Different stages need different models

• Transcription recognizes speech and prepares text and speaker information.
• Text and visual indexing make footage searchable; indexing is not the same as translating it.
• Translation prepares target-language dialogue. Voice generation produces the speech track.
• Lip sync changes visible mouth movement to match the generated audio. Standard and HD options have different model and license requirements.

Prepare only what the task needs

1. Open model settings from the workbench. Identify the required model for the stage you intend to run.
2. Check its device support, storage needs and readiness in the app. An environment may need to be prepared before a model can run.
3. Download and prepare the required components. Keep the destination drive available.
4. Run a short sample on the device you will use for production before queuing a full recording.

Mac and Windows checks

On Mac, check Apple Silicon support, available unified memory and folder permissions. On Windows, check the supported NVIDIA device and driver, available GPU memory and access to the media folder. A model that fits on one device may not fit on another.

Avoid loading several large tasks simply because they use different features. Local speech, voice and lip-sync jobs still compete for memory, storage and compute.

Understand performance without a fixed speed promise

Processing time depends on the model build, video resolution, selected quality, number and visibility of faces, recording length, device load and available memory. This guide does not publish estimated seconds or tokens per second as Bonframe benchmarks.

• Measure the same representative clip with the same settings when comparing devices.
• Separate model downloads and cold loading from actual generation time.
• For memory errors, reduce other active workloads and review the selected model or quality before retrying.

Related guides

• Install Bonframe and start your first project
• Transcribe and index a folder of videos
• Multi-speaker and multi-angle lip sync