Most AI video generations finish in one to four minutes; heavier pipelines like multi-clip reels or story generation run five to fifteen minutes, and VdoBloom applies a hard 30-minute ceiling (1,800 seconds) to a single video job before it is marked failed and the credits are refunded. Image generation is much faster — typically well under a minute, with ceilings of five to ten minutes depending on the model. Anyone promising instant AI video is either using a very short, very low-resolution model or counting only the queue acknowledgement.

Those numbers are not marketing estimates. They are the actual timeouts and polling intervals configured in the VdoBloom backend for each model family, so they represent the outer bound of what the platform will wait for before giving up on your behalf.

Realistic wait times by job type

Job type	Typical wait	Hard ceiling	Example cost
Image generation (1K)	10–40 seconds	5–10 minutes	Nano Banana, 6 credits
Image edit	15–60 seconds	5–10 minutes	Flux Kontext Pro, 7 credits
5s text-to-video, 720p	1–3 minutes	30 minutes	WAN 2.7 720p 5s, 32 credits
5s image-to-video, 1080p	2–5 minutes	30 minutes	WAN 2.7 1080p 5s, 48 credits
10s video, premium tier	3–8 minutes	30 minutes	Kling V2.5 Turbo 10s, 80 credits
High-fidelity tier	4–10 minutes	30 minutes	Seedance 2 1080p 5s, 203 credits
Veo 3 quality	3–10 minutes	30 minutes	160 credits
Story generator frames	1–2 minutes	—	Per-frame image cost
Story generator videos	5–15 minutes	Per-clip 30 minutes	Per-clip model cost
Video edit / character replace	5–15 minutes	30 minutes	WAN 2.7 video edit 720p 5s, 32 credits

Credits convert at the Lite plan rate of $15 for 300 credits, so one credit is $0.05. A 32-credit clip is $1.60 of your allowance; a 203-credit Seedance 2 1080p clip is $10.15. Full plan and pack pricing lives on the pricing page.

What is actually happening while you wait

Generation is asynchronous, not streaming. When you hit generate, four things happen in order.

Validation and credit deduction. Near-instant. Your prompt, model and settings are checked and the credits are held.
Submission to the provider. A job is created upstream and you get a task id back, usually within a second or two.
Queue time. This is the variable part. You are in line behind everyone else using that same model globally, and it is largely outside our control.
Render and delivery. The model renders, the file is fetched, stored and attached to your account.

The backend polls the provider on a fixed interval while this runs — typically every 3 to 15 seconds depending on the model family, and as slow as every 3 minutes for a couple of heavyweight editing models where faster polling would only waste calls. That polling interval is why a clip that finished at 1:47 might appear at 1:52. It is not extra render time.

Why two runs of the same model take different times

Five factors move the number, roughly in order of impact.

Queue depth. The largest single variable. Peak hours in US and European time zones can double the wait on popular models with no change in render time at all.

Duration. Close to linear. A 10-second clip generally takes about twice as long as a 5-second one on the same model, which mirrors the pricing — WAN 2.7 720p is 32 credits at 5s and 64 at 10s.

Resolution. Steeper than linear. Moving Seedance 2 from 480p to 1080p at 5 seconds takes the cost from 38 credits to 203, and render time climbs sharply alongside it.

Model tier. Fast variants exist precisely to trade quality for latency. Seedance 2 Fast at 480p 5s is 24 credits against 38 for the standard model, and it returns noticeably sooner.

Pipeline depth. Anything that chains stages — story generation, multi-clip reels, lip-sync with a separate audio pass, upscaling after generation — adds each stage’s wait end to end.

You do not have to sit and watch

This is the part most people miss. Video jobs are tracked server-side, not in your browser tab. VdoBloom runs a recovery service that reconciles any job left in a pending or processing state — including across a server restart or a deployment — by re-checking task status with the provider and updating your account when it completes.

Practically, that means you can close the tab, lock your phone, or start a second generation, and finished videos will be waiting in your creations library. The in-page progress indicator is a convenience, not the mechanism.

When a job does not come back

If a video passes the 30-minute ceiling it is marked failed and the credit deduction is reversed by the refund path in the credit system. You are not charged for a job that never produced a file. What is not refunded is a generation that completed successfully but that you did not like — which is the practical argument for testing at a cheap resolution tier before committing to 1080p.

If a job sits in processing well past the typical range for its model, the useful move is to start a second attempt at a lower tier rather than waiting it out, since the stuck job will resolve or refund on its own.

How to get results faster

Pick the fast variant while you are still iterating and switch to the quality tier only for the final render. Generate at 480p or 720p until the composition is right. Prefer 5-second clips over 10-second ones when the extra runway adds nothing — you halve both the wait and the cost. Queue several variations at once instead of running them one after another, because queue time overlaps rather than stacking. And for image work, use image generation to lock in the look before spending video credits animating it.

If you are starting a video from a photo of a real person, make sure you have that person’s consent before you animate them — that applies whichever model or effect you choose.

Model-by-model durations, resolutions and aspect ratios are listed on each page in the model directory, and you can time a run yourself at no cost on the free AI video generator. Paid plans also remove the watermark, which matters if you were planning to re-render later for a clean file — see watermark-free generation.

Frequently asked questions

How long does a 5-second AI video take to generate?

On a mainstream 720p model, typically one to three minutes end to end including queue time. Premium and 1080p tiers push that toward four to ten minutes. The hard ceiling before VdoBloom fails the job and refunds you is 30 minutes.

Can I close my browser while a video generates?

Yes. Jobs run server-side and a recovery service reconciles anything still pending, including across deployments and restarts. Your finished video appears in your creations library whether or not the tab stayed open.

Why is AI image generation so much faster than video?

A video is many frames that must also stay temporally consistent with one another, so the compute is orders of magnitude higher than a single still. That is reflected in the timeouts too: image models on VdoBloom are given five to ten minutes, video models thirty.

Does paying more make generation faster?

Not directly — there is no priority queue you can buy. But the fast variants of a model are engineered for lower latency, and a bigger credit balance lets you run several variations in parallel instead of serially, which is the real speed-up.

Am I charged if the generation times out?

No. A job that fails or times out triggers an automatic refund of the deducted credits. Only a successfully delivered video is charged.

How Long Does AI Video Generation Take? Real VdoBloom Wait Times per Model