How generation works

The shared async pipeline behind every media type, request, pending, live progress, then a finished file in your Media Drive, with safe retries that never double-charge.

Every kind of media in AnyBot (images, video, and presenter clips) runs through one shared async pipeline. You send a request, get an immediate "working on it" acknowledgement, watch live progress, and end up with a finished file saved to your Media Drive. This page is the reference for that flow, so the individual generation pages can stay focused on what they make rather than how the machinery runs.

The four states of a request

State	What it means	Where you see it
Requested	Your brief is validated, priced against your plan, and queued.	Request returns right away, nothing to wait on.
Pending	The job is waiting for or holding a worker slot.	Tasks Center, marked in-flight.
Progress	The worker is actively refining, generating, then uploading.	Live status (`generating`, `uploading`) over a realtime channel.
Ready (or Failed)	The finished file is persisted and billed, or the attempt failed and can be retried.	Media Drive, plus a terminal notification.

How a request flows

Generation is fire-and-forget: the request never blocks while a model runs. When you start a generation, AnyBot creates a tracking row (status pending), enqueues the job on a dedicated media worker, and hands your browser a realtime subscription, the acknowledgement, not the result.

The media worker runs on a dedicated media queue with its own concurrency pool, separate from the request-serving path, so generations don't slow down the app you're clicking through. It works in stages and publishes live progress as it goes:

Refine. Your brief is turned into one brand-aware prompt using your agent's persona and Materials Library.
Generate, the chosen model renders your media. Progress shows generating.
Upload, each result is stored and given a stable file reference. Progress shows uploading.
Finalize, AnyBot persists the files, meters the usage, and publishes a terminal event. Your files appear in Media Drive.

Because the result is delivered over a realtime channel scoped to your generation, it survives a page navigation or a hard refresh, the Tasks Center re-hydrates in-flight work from the backend, which is the source of truth.

Worked example

You ask for four product images. The request returns immediately and the Tasks Center shows one in-flight task. Moments later the status flips to generating, then uploading. When it finishes, four images land in your Media Drive and your image pool is metered for exactly four. If one of the four failed to store, you get the three that succeeded, and you're only billed for three.

Retries never double-charge

Metering happens only when a finished file is delivered, a "winner-only" rule. If a generation fails, nothing is charged for that attempt. Retrying a failed generation replays the exact stored request as a new attempt (linked to the original for your records) and bills only if it succeeds. So retrying is always safe: you never pay twice for one result.

What runs on this pipeline

Image, video, and presenter/avatar generation all share this pipeline: the request, progress, and delivery mechanics are the same for every media type.

Usage is metered in AnyBot Tokens across 6 ABT pools, each media type drawing on its own pool so one never eats another's quota. See Media billing for the dimensions.

FAQ

Do I have to wait on the page while it generates? No. The request returns instantly and progress streams to the Tasks Center; you can navigate away and come back.

Where do finished files go? Into your Media Drive, organized in your space and reusable through the unified picker everywhere in the app.

What happens if a generation fails? You're not charged, and eligible generations can be retried. The retry is a fresh attempt that only bills on success.

Do video and talking-head clips use this pipeline? Yes, both reuse it. Video is funded on every paid plan, and presenter clips bill per minute from the avatar pool.