Background Generations

Why background generations?

Some AI models — especially video generation — take several minutes to complete. Without background generations, you’d be stuck staring at a loading screen, unable to do anything else in modelBridge.

Background generations solve this. When a generation takes longer than expected, modelBridge automatically moves it to the background so you can keep working. This means you can:

Browse and explore other AI models while your video renders
Start additional generations on different models — build a pipeline of content
Adjust settings, review costs, or manage your project
Work on your Premiere Pro timeline — the plugin doesn’t block anything

You’ll hear a chime and see a notification the moment your generation is ready, no matter where you are in the plugin. Think of it like sending a render to a render farm — you hand it off, keep working, and get notified when it’s done.

This is especially powerful for professional workflows where you need to generate multiple assets simultaneously — a hero video on one model, B-roll on another, and a voiceover on a third, all running in parallel.

How it works

When you click Generate, modelBridge begins polling fal.ai for your result. If the generation doesn’t complete within a category-specific time window, it automatically moves to the background. This isn’t something you choose — it just happens.

The handoff thresholds vary by model type because different categories have different typical generation times. Video models stay in the foreground longest, image models the shortest. The thresholds are tuned so that most generations complete in the foreground — background handoff only kicks in for generations that are genuinely taking longer than expected.

When handoff occurs, a notification appears confirming that your generation has moved to the background. The first time this happens, the notification includes extra context explaining what background generations are and what to expect. After that, the message is shorter — you already know the drill.

Once in the background, the generation appears as a row in the Active Generations Panel at the bottom of the plugin. This panel is always visible, regardless of which model or tab you’re viewing.

The Active Generations Panel

Each row in the panel shows:

Status dot — orange (running), green (completed), red (failed)
Model name — which model is generating
Elapsed timer — how long it’s been running
Expand chevron — click for detailed progress, guidance text, and links

5-step progress tracking

Expand any row to see exactly where the generation is:

Sent — your request has been submitted to fal.ai
Queued — waiting for GPU allocation (shows queue position when available)
Generating — the AI model is producing output
Downloading — retrieving the result
Importing — adding to your Premiere Pro project

Steps turn green as they complete. The active step pulses. Everything updates in real time.

What you can do while it runs

Once a generation moves to the background, the plugin is fully usable:

Switch models — browse the catalog, explore new models, read descriptions
Start new generations — generate on a different model while this one runs
Adjust settings — change your API key, toggle preferences, manage costs
Edit your timeline — Premiere Pro is never blocked by a generation

The one thing you can’t do is start another generation on the same model that’s currently running. That model stays locked until its generation completes or fails. This prevents conflicting results and keeps your cost tracking accurate.

Estimated time

If you’ve run the same model before, modelBridge shows an estimated completion time based on your previous generations. The more you use a model, the more accurate the estimate becomes.

For models you haven’t used before, modelBridge shows a category-based average instead:

Video models — “Usually under 5 min”
Image models — “Usually under 1 min”
Audio models — “Usually under 2 min”

Estimates appear in the expanded row during the first few minutes of a generation, along with a clock icon and formatted time.

Staged communication

modelBridge doesn’t just show a spinner and leave you wondering. Messages update as your generation progresses — from initial submission through active processing to final delivery. The longer a generation runs, the more context the panel gives you about what to expect.

Early messages confirm that the generation is running and remind you that you can keep working. As time passes, the messages reassure you that longer waits are normal for certain model types, explain what might cause extra processing time, and eventually warn you as the monitoring limit approaches. Links to your fal.ai dashboard and this documentation page appear when a generation runs longer than usual.

Notifications

When a generation finishes — success or failure — two things happen:

Completion chime — a soft two-tone sound plays (if enabled in Settings). This is useful when you’re focused on editing and not watching the panel.
Snackbar notification — a floating notification appears at the bottom of the panel with the model name and result status. You can act on it immediately or dismiss it.

If you’re on a different tab (like Costs or Settings) when a generation completes, a notification dot appears on the Generate tab so you know there’s something waiting for you.

Completed results stay in the Active Generations Panel until you preview them. They aren’t auto-dismissed — you can come back to them whenever you’re ready.

Parallel generations

You can have multiple background generations running simultaneously on different models. Each gets its own row in the Active Generations Panel with independent:

Status tracking and elapsed timer
5-step progress indicator
Staged communication messages
Completion notifications and chimes

There’s no limit to how many you can run at once. Start a video generation on Kling, switch to FLUX for an image, queue up a voiceover on a speech model — each runs independently and notifies you separately when done.

The only constraint is one generation per model at a time. You can’t run two generations on the same model simultaneously.

Timeout

modelBridge monitors each background generation for up to 30 minutes. If the generation hasn’t completed by then, monitoring stops and you’re notified.

This doesn’t mean your generation failed. It may still be running on fal.ai — the plugin simply stops checking. You can:

Check your fal.ai dashboard — the link is provided in the timeout notification. If your generation completed, you can download the result from there.
Try again — if the result never appeared, something likely went wrong on fal.ai’s side. Regenerating is usually the best next step.

The 30-minute limit exists because indefinite polling wastes resources and could mask a genuine failure. In practice, the vast majority of generations complete well within this window.

Recovery

Background generations are resilient to interruptions:

Close and reopen the panel — active jobs are automatically resumed when the panel reopens
Restart Premiere Pro — recovery data is stored locally, so your generations are picked up on next launch
Switch models freely — the panel tracks all jobs regardless of which model you’re currently viewing

Recovery works by saving generation state (model ID, request ID, status URL) to local storage when a generation enters the background. On next load, any unfinished generations with valid recovery data are automatically re-polled from where they left off.

If enough time has passed, recovery data expires — but by then, the result is either available on your fal.ai dashboard or the generation has long since timed out on fal.ai’s end.

When things go wrong

Failed generations stay visible in the panel with a red indicator. They’re never auto-dismissed — you always see what happened. There are three main failure scenarios:

fal.ai returns FAILED

The generation definitively failed on fal.ai’s side. You were not charged for the attempt. The error message explains what went wrong, and you can retry with adjusted settings.

Connection lost

modelBridge lost contact with fal.ai during polling. Your generation may still be running — you just can’t see its status. A Check status now button appears so you can manually re-check. If the generation completed successfully, the result downloads and imports automatically.

Monitoring timeout

After 30 minutes of polling, modelBridge stops checking. Your generation may still complete on fal.ai. A link to your fal.ai dashboard is provided so you can check manually.

Error categories

Error messages use color-coded categories so you can quickly understand what happened:

Red — something about your input needs fixing (wrong dimensions, unsupported format)
Amber — billing or content policy action required (check your fal.ai account)
Blue — temporary issue on fal.ai’s side, usually resolves on retry

Category differences

Different model types have different characteristics when it comes to background generation:

Category	Typical time	Foreground window	Notes
Video	1 to 5 minutes	Longest	Most likely to enter background
Audio	30 seconds to 2 minutes	Medium	Speech and music models vary widely
Image	5 to 30 seconds	Shortest	Rarely enters background unless queue is long