Video production is the work of turning an intention into finished moving image. The intention might be a product launch, a category explainer or a paid-social test. The output might be one film or forty cutdowns.
The craft splits into three phases, and the split matters because the cost of a change rises steeply as you move through them. Deciding something in a document is cheap. Deciding it on set is expensive. Deciding it in the edit is sometimes impossible.
The three phases, and what each one actually decides
Pre-production is where the film gets designed. The brief becomes a treatment, the treatment becomes a script and shot list, and the logistics become a schedule. Casting, locations, product samples, permits and rights all get settled here.
Production is the filming. Its job is narrower than people expect: capture the coverage the plan calls for, protect product accuracy, and leave the edit enough options. A shoot day is mostly logistics wearing a creative hat.
Post-production is edit, sound, colour, graphics, captions and delivery. It is also where a single shoot becomes many assets, which is the phase most advertising projects underestimate.
Why advertising changes the shape of the work
A brand film made for one channel and an advertising shoot made for paid social are the same craft with different constraints.
Advertising needs versions. The same idea has to work vertically and squarely, with sound and without it, at six seconds and at thirty. Those are not export settings. They are framing and pacing decisions that have to exist in the shot list before anyone rolls.
The practical consequence: shoot for the crop you will need, not the crop you are monitoring. Framing that reads well in a wide edit can lose the product entirely once it is cut vertically.
Platform requirements shift, and they differ by placement rather than by platform. Meta publishes per-placement guidance, TikTok and YouTube publish their own, and none of them stays still long enough to be worth memorising. Check the current specification when you plan the delivery, not from memory.
Captions are not optional polish. WCAG 2.2 sets captions for prerecorded audio at Level A, its baseline conformance level: "Captions are provided for all prerecorded audio content in synchronized media, except when the media is a media alternative for text and is clearly labeled as such." Most viewers also start without sound, so captions serve both accessibility and performance.
Rights, which is where projects quietly go wrong
Talent, music, stock footage, locations and product claims all carry permissions, and each permission has a scope: which markets, which media, how long.
A film cleared for organic social and then run as a paid advert in three countries is a rights problem, not a media problem. It is discovered late, usually by someone outside the marketing team, and the remedy is often a reshoot.
Agree the intended use before the shoot and write it into every agreement. If the campaign might expand, buy the wider scope while you have leverage.
A worked example
A skincare brand wants to launch a serum. The objective is trial among people who already buy the category but have never bought this brand.
Pre-production settles the hard things: three concepts narrowed to one, a demonstration that shows texture without claiming a clinical result, one location, one presenter, and rights covering paid media in four markets for two years.
The shot list is written backwards from delivery. The team knows it needs a vertical hero cut, a square feed cut, three opening variants and a product-only sequence for retargeting. So the day is planned to capture a clean vertical frame of every beat, plus wider coverage for the square versions.
On the day, the discipline is unglamorous: match the product to the approved packaging, keep continuity on the texture shots, and get the presenter's lines in short takes so the edit can rearrange them.
Post assembles one master narrative, then derives the variants from it. The three opening variants exist to test one belief: that showing texture first beats showing the presenter first.
The result is one shoot supporting a test, rather than one film and a hope.
What drives scope
Runtime is a poor proxy for effort. A thirty-second film can cost more than a three-minute one.
What actually moves scope: how many distinct concepts, how many locations and moves, whether there is cast and how they are cleared, how much of the frame is built rather than found, how much post is involved, how many versions get delivered, and how wide the rights are.
Two projects with identical runtimes and different answers to those questions are not comparable, which is why quoted day rates tell you very little on their own.
Choosing a production model
Studio buys control. Consistent light, predictable schedule, no weather. It suits product-led work where the product must look exact.
Location buys context and credibility, and costs flexibility.
Creator-led buys the register of a recommendation rather than an advert. It suits demonstration and objection handling, and it needs clear rights if the material becomes paid media.
Animation suits explaining something invisible: a mechanism, a comparison, a process.
AI-assisted production has moved quickly. It is now genuinely useful for variation, for scenes that would be disproportionate to build, and for getting more versions from a fixed budget.
It is weakest in two places: where a specific real product must be depicted exactly, and where a claim rests on the footage being a record of something that happened.
Treat it as one method among several, chosen per shot rather than per project.
Most real projects end up hybrid, and the decision should be made shot by shot against what each shot has to prove.
Where this breaks down
A shoot cannot fix an unclear proposition. If nobody can say what the film is supposed to make someone believe, more coverage will not help.
Volume is not the same as optionality. Forty near-identical cutdowns give a media team forty ways to run the same idea. Planning three genuinely different openings gives them something to learn from. See performance creative for how that testing logic works.
And production quality has a ceiling on its contribution. Beyond a competent, on-brand execution, further polish tends to show up in how the team feels about the work rather than in the result.
Handover, briefly
The deliverable is not a file, it is a set: masters, per-placement versions, caption files, a stills selection, and a record of what was licensed and for how long.
That last item is the one people skip and later need.
Where Adsome fits
Adsome produces video and image advertising for e-commerce and consumer brands, combining conventional production with AI-assisted methods where they genuinely serve the shot. You can see finished work in our AI work and client cases.
Start from a scoped problem, not a runtime. "We cannot test enough openings" and "our product never looks right on camera" lead to very different plans.
Relevant tutorials
The step-by-step production side lives in the tutorials:
- An AI-native ad production workflow, end to end walks the whole pipeline.
- Anatomy of a high-converting AI video ad takes a finished ad apart.