The Vionlabs processing system transforms your video assets into ready-to-use product outputs such as Previews, Thumbnails, Fingerprint+, or BingeMarkers.

Each product consists of several processing steps. Some steps read and analyze the video directly — these are called extractors — while others combine previously generated data — these are processors.

Customers interact with this system through different interfaces: the Catalog API for registering and ingesting assets, the Task API for triggering specific processing runs, the Results APIs for retrieving outputs and versions, and the WebApp, which provides visibility into products, users, and permissions.

How to start processing

There are three main ways to initiate processing:
 

1. When you ingest an asset via the Catalog API, the system automatically creates jobs for all products currently enabled for that catalog. If no products are active yet, nothing runs until a product is enabled or a task is created.
    
2. When you enable a product for a catalog, the platform generates jobs for all assets in that catalog, ensuring that the new product is produced across your full library. This is often used to roll out new capabilities or product versions and can be throttled through priorities and autoscaling limits.
    
3. Finally, you can create a task for an individual asset and product via the Task API. This is the most direct way to request processing — ideal for urgent or single-asset scenarios.

From creation to completion

Once a job is created, it moves through several predictable stages. It first appears in a Created state and waits for its prerequisites — for example, other steps that must finish first or a video file that must become accessible. When all requirements are met, it becomes Ready and eligible for execution.

The job then enters the Executing phase, where the system runs the appropriate container — on CPU or GPU, sometimes in batches for efficiency. When processing finishes successfully, the job becomes Completed and its results are available through the APIs or the Portal. If an error occurs, it moves to Failed. Crashed or stalled jobs are automatically unlocked and retried within reasonable limits, without manual intervention.

All jobs are dependency-aware: extractors run before processors, and downstream steps automatically start as soon as their inputs exist.

Cloud and edge processing

Vionlabs supports both cloud and edge execution models.

In the cloud setup, your content is processed in the Vionlabs environment, either by granting access to your storage or by delivering files to a designated bucket. In the edge setup, extraction takes place within your infrastructure, ensuring that raw content never leaves your controlled environment. Only derived data or features are sent back according to configuration.

You can decide per catalog whether to operate in cloud or edge mode, depending on compliance, bandwidth, and latency requirements. Both AWS and GCP deployments are supported.

Scaling and performance

The platform is designed to scale automatically. It increases or reduces concurrency to maintain consistent throughput within your configured ceilings, which helps control costs and resource usage.

Similar jobs can be grouped and processed in batches for better efficiency, while GPU acceleration can be enabled for products that benefit from it. These options are managed at the product, version, or catalog level and can be adjusted to balance performance and cost.

Priority and scheduling

Each processing job has a priority value that determines when it will be executed. This makes it possible to ensure that time-critical assets — for example, a title scheduled to go live soon — are processed first.
When multiple products depend on the same intermediate step, the job automatically inherits the highest priority among them, ensuring that shared processing happens only once and as early as possible.

Reprocessing and updates

Whenever a source file is replaced or a new product version becomes available, the system automatically schedules new jobs. This ensures that the latest request always takes precedence and that outdated results are replaced. Reprocessing can also be triggered manually, for instance to update metadata or to refresh outputs for higher-quality source material.

Obtaining your results

All processed outputs can be retrieved either through the Results APIs or via the Customer Portal, where you can track job status, review logs, and manage product activations.
If you prefer automated updates, you can configure webhooks to receive notifications as soon as a product has finished processing. These notifications allow your downstream systems — for example, publishing workflows or analytics pipelines — to react immediately.

Configuration options

Customers can adjust a number of parameters to align processing behavior with their operational needs.
Products and their versions can be activated per catalog, with newer versions automatically deactivating older ones to maintain consistency. Priorities can be defined per task or product so that urgent workloads receive preference. Autoscaling ceilings help cap concurrency and manage spend, while GPU usage can be toggled for products that require acceleration. You can also decide, per catalog, whether processing should occur in the Vionlabs cloud or in your own edge environment. Finally, webhook notifications can be activated to provide external systems with completion events.