How long an AI video takes: 3 working days, hour by hour

We make a 15-second clip in 3 working days and a 30-second one in 3 to 4 working days, counted from the moment you approve the script. Inside those three days there are 8 to 11.5 hours of our work, and generation takes half an hour to an hour of that. Below we break the turnaround down hour by hour and compare it with the 10 working days of AI studios and the 12 to 18 days of a traditional production.

How many days it takes to make a clip with a neural network

Fifteen seconds takes 3 working days, thirty seconds 3 to 4 working days. The difference is not the size of the file but the number of shots: the more shots, the more generations, selection and cuts.

Here is what the turnaround is made of. The hours in the table are ours, for our process.

Stage	Hours	Day	What happens
Brief and script	2–3	1	We work through the product and the task, and write the text for 15 or 30 seconds
Storyboard and starting frames	1–1.5	1	We break the script into shots and prepare the reference images
Generation with retakes	1–1.5	1–2	The machine computes for 0.5 to 1 hour, selection takes the rest
Editing, sound, captions	2–3	2	Runs in parallel with generation, not after it
Check against the product	0.5	2–3	Frame by frame comparison with the real object
Revisions after approval	1–1.5	3	One or two rounds, included in the turnaround
Export and file handover	0.5	3	Master file and versions for each platform
Total person-hours	8–11.5

Three days for eight hours of work is not a sign that we work slowly. It is that there are pauses between the stages. There are three of them, and none can be removed.

The first pause is the machine. While the model computes the next shot, there is nothing to do with that shot, and no amount of parallel work will speed it up. The second pause is yours: you read the script and watch the draft, and that is right, because fixing the text is cheaper than rebuilding a finished clip. The third pause is technical: between generation and editing the files have to be downloaded, brought to a single format and checked for artefacts.

That is exactly why the promise of a video in an hour only turns up where "video" means one generation with no script, no selection and no checking.

How many minutes a neural network takes to generate one shot

A neural network computes one short clip in 60 to 130 seconds, and quality doubles that time. Kling 3.0 in fast 720p mode returns a result in 60.4 seconds, in quality 1080p in 116.2 seconds, which is almost twice as long for the same scene (Artificial Analysis, measurements for July 2026). Veo in 1080p takes around 127 seconds by the median over a month of observation (Higgsfield).

One generation gives 5 to 10 seconds of video. So even a perfect fifteen-second clip is assembled from at least two or three calls to the model, and a thirty-second one from four to six.

There is also an upper bound that service advertising does not mention. On heavy models at maximum quality one generation takes 8 to 18 minutes (Creative Bloq). If the scene is complex and that is the mode you need, compute time grows tenfold, and that changes the plan for the whole day.

A separate word on queues. Services cap the number of simultaneous jobs: ModelArk allows 10 parallel generations per account (BytePlus documentation). While there is no free slot, the next shot simply waits.

How many retakes one finished shot takes

One accepted shot takes 3 to 7 generations, and the median across five independent practitioner write-ups is four to five. The model computes from scratch every time and every time slightly differently, so selection counts towards the time just as much as the computing does.

A shot is usually rejected for one of four reasons. The camera move went somewhere other than where it was asked to go. The logo or the text on the packaging warped. The object changed shape or lost a detail halfway through the clip. The light did not match between neighbouring shots, and it shows on the cut.

The first three are caught in seconds of viewing, the fourth only when the shots are compared side by side. So selection is not done clip by clip but in sequence: we watch whether the shots follow one another without a jolt.

The starting frame costs more than the rest. To get the reference image the motion will grow from, practitioners spend 10 to 20 generations (Kreaitoff). The reason is simple: a weak starting frame gives a weak clip, and no amount of editing saves it afterwards.

The scale shows up well in an extreme example. Keven Mendibur, the author of a short film, ran 3,229 generations and spent 242 hours (Creative Bloq). That is not how we work, but the proportion there is the same as on a fifteen-second clip: minutes of machine time per attempt and hours of human time on selection.

For working out a turnaround, one number matters: the multiplier. Four retakes per shot means the real machine time is four times what the model spec says.

How long a neural network takes to generate a 15 and a 30 second clip

The machine computes fifteen seconds in 23 to 31 minutes and thirty seconds in 46 to 62 minutes. The arithmetic is simple: the number of shots multiplied by the retakes and by the time of one generation.

Running time	Shots	Generations at 4 retakes	Compute time, Kling Pro 1080p	On a heavy model at maximum quality
15 seconds	3–4	12–16	23–31 minutes	1.6–4.8 hours
30 seconds	6–8	24–32	46–62 minutes	3.2–9.6 hours

This is our own calculation on data from Artificial Analysis for July 2026 and from Creative Bloq, not a quote from a source. We did not find a single piece of material where generation time is multiplied by the number of retakes, although without that multiplication any "two minutes per clip" means nothing.

The right-hand column explains why a turnaround cannot be promised as the same for every job. The same thirty-second clip takes about an hour in Kling quality mode and almost a working day on a heavy model at maximum quality. The mode is set by the scene: smooth product shots get by on fast, complex motion and fine texture need quality.

How many hours the script and the storyboard take

The script with the storyboard takes us 3 to 4.5 hours and is done on day one. In a traditional production the same stage stretches over weeks: approving the concept alone takes 3 to 10 working days, and the script anything from 2 days to 2 weeks (Videoglobal).

The difference is not typing speed. In the classic setup the script serves the shoot: a location, props and actors are found for it, and a mistake costs a shooting day. For us the script serves the generation, and the price of a mistake is a few minutes of recomputing.

The storyboard matters more than it looks. We prepare the reference images before video generation is switched on, because it is the starting frame that decides how alive the motion turns out. An hour spent here saves three hours on selecting retakes.

How many hours editing, captions and checking take

Editing with captions takes 2 to 3 hours, and the point here is not the number of hours but that they run in parallel with generation rather than after it. While the model computes the next shot, the finished ones are already on the timeline. Practitioners keep the editing software open alongside the generator for exactly this reason: otherwise the continuity of motion between shots breaks (Creative Bloq).

On the market this stage is separate and sequential. Simple editing takes 1 to 2 days (DTF), and a traditional production allows around 2 working days for the sound mix alone (Videoglobal).

Checking the finished clip against the real product adds half an hour. We compare the frames with the original photographs on six points, and we do it before delivery, not after.

How long revisions take

One or two rounds of revisions sit inside our three days and take 1 to 1.5 hours. The industry norm is 2 to 3 iterations (Videoglobal), and on the market each round costs a separate 3 to 5 working days.

It works that way because the script is agreed before production starts. You see the text and the storyboard before we switch on generation, so revisions after assembly are about details: the length of a shot, the wording on a caption card, the order of scenes. There is nothing to reshoot in that setup.

Large studios sell rounds of revisions as part of a package: price lists carry rates for three and for five corrections. It is an honest approach, and it also shows what a revision really costs when production goes through filming.

How many days an agency takes to make a clip

A traditional production makes a simple clip in 12 to 18 days, an AI studio in 5 to 10 working days, and a freelancer on a job board takes up to 2 weeks for the editing alone. Below are the numbers for each type of contractor, with sources.

Who	Turnaround for 30 seconds	Price	Revisions inside the turnaround	Script included
Blind Chameleon	3–4 working days	from $114	yes	yes
AI studio	5–7 days or 10 working days	$714 to $7,143	usually 2 rounds by contract	yes
Traditional production	12–18 days, up to 2 months in practice	hundreds to many thousands of dollars	approvals sit outside the turnaround	yes
Freelancer on a job board	up to 2 weeks for editing	about $429	by agreement	no

How many days a traditional production takes

A simple 15 to 30 second clip takes 12 to 18 days, a mid-range one 24 to 35, a premium one 45 to 70 (raketadigital). Video under a minute takes 10 working days on average (obscura.pro). Industry estimates are wider: 2 to 3 weeks for an advertising clip and 4 to 8 weeks for a corporate film (Oblako Production), and with 3D graphics from 2 to 4 months (FAZA).

How many days an AI studio takes

The spread here is the widest. Infomult quotes 10 working days for 30 seconds, Sxema from 10 working days at a price from $7,143, VideoHunter 5 to 7 days from $714, AVA studio 2 to 4 weeks with a record of 5 days. So even among those who work with neural networks, a turnaround of a week and more counts as normal.

How many days a freelancer on a job board takes

On job boards like fl.ru the typical turnaround for the editing alone reaches 2 weeks at a price of about $429, and the script is usually not part of the job: the client brings it.

Why an agency's real turnaround is twice the contract one

Contract turnarounds usually leave out the time for approvals, and obscura.pro and Videozayats write about this independently of each other. Both say it plainly: a contract month becomes two or more in practice.

Two cases show where exactly the time goes. On the Abiroy project the production itself took 3 weeks, while preparation stretched into 3 months of waiting for materials from the client. On a project for SIBUR one animation was discussed for 4 days and assembled in a couple of hours.

Hence the practical conclusion worth holding on to when you compare offers: compare not the figure in the contract but what is inside it. A 12-day turnaround with no approvals and a 4-day one with them are different things, and the second is shorter not by eight days but by a month.

Doubling the speed on this market is achieved in exactly one way. In the TookToDo case a month turned into two weeks, and the reason was the speed of the client's replies, not the studio's production capacity.

There are three questions worth asking any contractor before signing, and they save more time than any negotiation about a discount. Does script approval fit inside the stated turnaround. How many rounds of revisions are included and how many days each one takes. What event starts the count: payment, the brief, or the approval of the script.

The answers to those three questions give you the real turnaround. It almost always differs from what is written in the offer, and the difference is measured in weeks, not days.

What day the 3 working days start counting from

The count starts from your yes on the script. We hand the script over the same day, which is 2 to 3 hours of work, so our day zero is measured in hours.

In a traditional production the same stage takes 3 to 10 working days and sits outside the contract turnaround (Videoglobal). Their day zero is measured in weeks, and that difference is where most of the gap comes from.

Weekends are not counted. An order on Friday with the script approved the same day means a finished clip on Wednesday. An order on Monday means Thursday.

What to send so the clip is ready in 3 days

On day one we need four things: photographs of the product at original resolution, a description of the task in a couple of sentences, the platform the clip will run on, and the numbers for the caption cards, if you need any.

The photographs matter more than everything else. The more angles, and the less retouching and infographics on them, the more closely the clip will match the real object. One daylight shot with no processing is worth more than ten processed ones.

A useful set looks like this: a wide shot from three sides, an end face or a cross-section if the object has thickness, a macro of the material texture, the detail you consider the main advantage, and the object folded or taken apart if it does that. Eight shots is usually enough, fifteen is better.

What is better not to send: cards with infographics already laid over them, screenshots from someone else's catalogue, and frames pulled out of video. Shape can be recovered from those, texture cannot, and texture is what separates a clip that looks like the product from a clip that looks like a render.

One rule about feedback. We hold 24 hours in reserve for your reply on the script and on the draft, and within that time the turnaround does not move. It is the speed of replies, not the power of the equipment, that separates two weeks from a month in any production.

Can a clip be made faster than 3 days

Yes, rush assembly is possible for a separate fee. On the market rush costs plus 30 to 50% on the price for a 48-hour turnaround (VideoHunter), and with us the manager works out the terms for the specific job: a lot depends on how many shots are in the script and whether there are ready photographs of the product.

What rush assembly covers

What gets shortened is everything that can be parallelised: editing starts from the first finished shots, generation runs at the maximum number of simultaneous jobs, revisions are agreed in chat within the hour.

What rush assembly does not shorten

Agreeing the script before production starts always stays. Without it the risk grows that the finished clip has to be rebuilt from scratch, and the rush turns into a delay. The check of the finished clip against the real product stays too: it is the gap between the picture and the object that comes back later as customer returns.

How to find out the exact turnaround for your clip

The exact turnaround depends on the running time, the number of shots and whether the photographs are ready. The calculator on the home page works out price and turnaround for a specific configuration: you pick the length, the number of clips and the voice option, and you see the total before placing the order.

What the clip itself costs and what the price is made of is covered in how much an AI video ad costs and how much it costs to make a video with a neural network.

Sources

• Artificial Analysis, generation time measurements for video models, July 2026.
• Higgsfield, median generation time over 30 days of observation.
• Creative Bloq, breakdown of a short film production: 3,229 generations, 242 hours.
• Kreaitoff, practice of preparing starting frames.
• BytePlus ModelArk, documentation: limit on parallel jobs.
• raketadigital, obscura.pro, Oblako Production, FAZA: turnarounds in traditional video production.
• Infomult, Sxema, VideoHunter, AVA studio: turnarounds and prices at AI studios.
• Videoglobal: stage by stage turnarounds and the iteration norm.
• Videozayats: approvals outside the contract turnaround, the Abiroy, SIBUR and TookToDo cases.
• DTF: editing turnarounds.

The figures were taken on 31 July 2026. Turnarounds and rates on the market change, so before you compare offers, check whether approvals are included in the stated turnaround.