# Realtime video APIs vs batch video generation: which fits?

> Choose live sessions or completed video jobs by interaction, latency and cost. Understand streaming output, state, controls and when to use both patterns.

Source: https://windflow.online/compare/realtime-video-api-vs-batch-generation · Updated 2026-10-11 · Windflow Labs

Use a **realtime video session** when the user needs to watch and control a model while it runs. Use a **batch generation job** when the product needs a completed clip. The right choice follows the interaction and result you need, rather than the word streaming in an API description.

Windflow provides realtime sessions for live editing, continuous generation and interactive worlds. Some session models generate windows of video at a time, so a live session still has model-specific first-frame and control boundaries.

## Compare the application lifecycles

| Dimension | Live session | Completed-video job |
|---|---|---|
| Application unit | A running model session | A submitted generation task |
| User experience | Connect, wait, watch, steer and end | Submit, track progress and retrieve a result |
| State | Kept across supported controls | Determined by the job's documented inputs |
| Output | A generated or edited media track | A finished clip or documented progressive output |
| Cost estimate | Ready duration and independent concurrency | The provider's job and output pricing rules |
| Failure handling | Reconnect, end and release ownership | Retry, cancel or resubmit according to the API |

This describes two integration patterns. A provider may support both; inspect the endpoint you actually plan to use.

## When a session earns its complexity

Choose a session for an explorable world, a camera effect or a scene that reacts to direction during playback. The state and controls make the experience possible, and your UI needs to reflect the session lifecycle.

Do not start a paid live session merely to populate an asset thumbnail. If a user has already supplied a complete brief and will only inspect the final result later, a completed-output workflow can be easier to operate.

## Streaming is not one behavior

A status stream can report queue progress. An output stream can deliver progressive text or file bytes. A generated video track can provide playable frames. An interactive session adds supported control inputs that affect the running generation state.

Ask whether a new prompt changes the current session, when the change applies, whether source motion continues during camera editing, and what happens to model state after a disconnect. Those answers establish interaction more reliably than a transport label.

## Price the workload you actually use

For a Windflow session, multiply each model's ready seconds by its published rate. Queueing and startup are free. A ready session can continue billing while video is paused, so release ownership when done. Several viewers of one session share its state and meter; independent sessions are billed independently.

For a video job, use the endpoint's current billing rules and the number of outputs you need. Do not equate seconds of generated video with seconds of wall-clock ready time. The [pricing guide](https://windflow.online/guides/how-per-second-pricing-works) explains Windflow's unit.

## Use both when it serves the workflow

A production tool can use finished assets for approved references and realtime sessions for an interactive preview. Save a recording only when the user asks, and make export and retention explicit. The session's generated track is the live experience; storing its output is a separate product step.

For live direction across shots, evaluate [LongLive 2.0](https://windflow.online/models/longlive-2). For generated video with sound, inspect [LTX-2.5](https://windflow.online/models/ltx-2-5). In each case, test first-frame and prompt-application timing rather than assuming instant control.

## FAQ

### Should I use a realtime API for every AI video feature?

No. Use it when the experience requires supported inputs and controls during generation. A completed clip can be a better fit for a fully specified task reviewed later.

### Does a WebSocket guarantee persistent model state?

No. A WebSocket is a transport. Check whether the endpoint preserves model state and accepts controls that affect generation within the same session.

### Is a ready session the same unit as a generated video second?

No. Ready time measures how long the session is in its billable state. Output duration measures the video produced. Budget with the unit your provider actually bills.