Use Case · Media

Media processing on the Software Lifecycle Cloud

Media processing on KERNSTACK keeps user media orderly from the first byte: originals upload directly into a bucket on presigned URLs, workers produce the renditions the product serves (thumbnails, transcodes, waveform stills), moderation scores attach to the media record, and only processed, cleared renditions get public URLs. The original never changes; every derivative is reproducible from it.

Container Workers Object Buckets Model Inference NoSQL Tables

Talk to us Read Technical Documentation → Software Lifecycle Cloud →

The engineering bottleneck with assembled stacks

Marketplace operators resize and moderate uploads inside the upload request until the first 40-megabyte phone video teaches them otherwise. Then come the queue service, the transcoding vendor, and the moderation dashboard — three more systems, while the product still cannot answer the simple question “show me everything we have of this upload and what was decided about it”.

Architecture on the Software Lifecycle Cloud

Upload completes into a quarantine prefix; a table change or workflow step starts the processing worker, which renders derivatives into a renditions prefix and writes the media record: dimensions, durations, rendition keys, moderation scores. Publication is a state on that record — public URLs exist only for cleared renditions — and reprocessing a new size is a worker run over the untouched original, not a negotiation with users to re-upload.

Renditions are cheap; originals are sacred

Keeping originals immutable and derivatives disposable inverts the usual regret: a new card layout needs a new thumbnail ratio, and the answer is a batch job over originals, run tonight. Teams that overwrote or discarded originals learn this once, expensively. The record tying original to renditions to verdicts is one table — the question from the pain paragraph answers itself.

Uploads bypass your servers entirely on presigned URLs; workers meet the file in the bucket
Vision moderation at 3 credits per image slots into the same pipeline step as resizing
A new rendition size is a re-run over originals — users never re-upload for your redesign

Where dedicated media clouds still win

Adaptive streaming at broadcast scale and exotic codec farms are specialised businesses. The product-scale majority — marketplace photos, profile media, content thumbnails, short clips — is a bucket, workers, and discipline, metered on the same balance as the product that serves them.

Media processing Data & Execution Flow

IngressOriginal lands in quarantinemedia-processing

StateWorkers render + scoreproject resource

ReviewCleared renditions go publicdeclared in stack.json

stack.json — Media processing plan · deploy

{
  "resources": {
    "buckets": [{ "name": "media" }],
    "workers": [{ "name": "process-media", "runtime": "python" }, { "name": "moderate", "runtime": "python" }],
    "tables": [{ "name": "media-records", "key_name": "media_id", "key_type": "S", "indexes": [] }],
    "workflows": [{ "name": "process-upload", "steps": [
      { "id": "render", "worker": "process-media" },
      { "id": "moderate", "worker": "moderate" }
    ] }]
  }
}

Estimated Footprint · Team Plan ~2.600 credits / mo

Assumes 30k uploads a month with a handful of renditions each; worker transcoding dominates, vision checks at 3 credits per image where enabled.

Metered from one shared balance across compute (0.01 cr/GiB-s), storage (0.04 cr/GB-hr), network (9.00 cr/GiB), and catalog models. See full credit reference →

FAQ

Frequently asked questions

How do large uploads avoid timing out?

They never pass through your functions: the browser uploads straight to the bucket on a presigned URL, and processing starts from the completed object. File size stops being an API design constraint.

Can we add a rendition size later?

Yes — that is the design’s payoff. Workers re-run over the stored originals and write the new ratio beside the old ones. No user action, no migration of masters, because the masters never moved.

How does moderation fit the pipeline?

As a scored step on the media record before publication: vision models at 3 credits per image for the automatic band, and a human workflow step for the uncertain middle. Nothing is publicly addressable before its state says so.

Connected in the Project Graph

Related capabilities, architectures, and comparisons

Ship your stack on the Software Lifecycle Cloud.

Closed preview on dedicated European infrastructure. Bring a repository or start from a blueprint.

Talk to us See pricing & credits →