Orelon logoOrelon
Tarifs

How to Scope and Budget an AI Video Production Workflow

16 nov. 2026 · Par Orelon Team

Explorez les modèles vidéo IA

Parcourez quelques créations de la communauté pour trouver l’inspiration, puis ouvrez n’importe quel modèle pour continuer à créer dans Orelon.

A practical guide to scoping AI video projects: what drives effort, how to brief for comparable quotes, and how to plan iteration without endless loops.

A request for an estimate on a cinematic AI video used to be a short conversation. Now the same brief can be answered with a laptop, a reference board, and a queue of generations — and the proposals that come back vary enormously, because no two of them price the same things. One supplier quotes per finished second, another per shot, a third per revision round, a fourth per deliverable package. Two estimates for what looks like identical output can differ by a factor of five, and both can be defensible.

That spread is not a reason to avoid the format. It is a reason to learn what actually moves the number. What follows is a practical breakdown of the variables that drive effort in AI-assisted cinematic work, how to brief so proposals become comparable, how to plan iteration before it spirals, and how to tell a lean quote from one that will quietly become the expensive option.

Why AI production budgets ignore the old rules

Traditional production scales with time and people. Every shooting day adds crew, equipment, locations, catering, insurance, and transport; post-production scales with edit hours and finishing passes. The mental model is pleasantly linear.

AI-assisted production breaks that model in two directions at once.

What fell out of the cost structure

Travel, permits, physical sets, and large crews largely disappear from the line items. A single creator can produce a look that once required a lighting team and a rented stage. Generation effort is also decoupled from audience size: a film for an internal meeting and a film for a global campaign can share nearly the same pipeline.

What appeared in its place

The new drivers are less visible but very real. The number of shots. The number of attempts each shot needs before it is usable. Resolution, frame rate, and duration targets. The consistency work that makes separate generations read as one film. And the human finishing that turns raw material into something you would put in front of a paying audience.

Those five variables explain most of the spread between competing proposals, and a serious estimate should be built against them rather than against a generic day rate.

The five variables that move almost every estimate

Strip away supplier-specific packaging and the same handful of factors remain.

Shot complexity and camera language

A locked-off wide of a landscape is cheap. A tracking shot of a character walking through a crowd, turning to camera, and speaking is expensive — not because it is impossible, but because every element has to hold across frames and every retry multiplies. Complex camera moves, hands, reflections, legible text in frame, and multi-person interaction all push the attempt count up. When you read a shot list aloud, listen for the phrases that signal risk: "and then", "turns", "unfolds", "reads". Each one is a place where a model has to cooperate with you.

Runtime versus edit density

Total runtime matters, but density matters more. A sixty-second film cut from six long shots is a very different job from a sixty-second film cut from twenty short ones. Short shots feel more cinematic and hide generation flaws, yet they multiply the number of generations, selections, and continuity decisions. A useful shorthand: price the shot count first, then the runtime. If a brief says "fast-paced montage" without a shot count, that is the first question to ask.

Delivery spec: resolution, frame rate, aspect ratios

Delivering for a phone screen is not the same pipeline as delivering for a cinema pre-roll or a large-format display. Higher resolution and frame rate mean longer renders, more storage, and a stricter quality bar — flaws that vanish on a handset become obvious on a six-metre wall. Aspect ratios multiply the work too: a vertical crop is rarely a simple reframe when the composition was built for widescreen, and three ratios can mean three separate framing decisions per setup.

Iteration volume

This is the most underestimated line in any AI video estimate. Generation is probabilistic. You rarely get the exact frame you imagined on the first attempt, so the real cost is the number of attempts needed before a shot reaches an acceptable standard — plus the human time to review, choose, and redirect those attempts. A hero shot with a character and a specific beat of performance might need twenty to forty attempts; an atmosphere shot might need three. Both are normal, and a brief that treats them as equivalent will misprice the project.

Human finishing

Editing, sound design, music licensing, colour consistency, motion graphics, titles, captions, and compliance review remain human work and do not shrink because the footage was generated. On many projects finishing is now the largest share of the budget, and it is the block most often missing from a low quote. If two proposals differ by a large margin, the finishing line is usually where the difference lives.

How to brief so proposals become comparable

Most quote variance is brief variance. If two studios interpret your reference reel differently, their numbers will differ for reasons that have nothing to do with efficiency.

The one-page creative brief

Keep it to a page: audience, the single message, tone in three adjectives, runtime, aspect ratios, delivery date, and a definition of done. The definition of done is the part most briefs omit, and it is the part that decides whether a project ends on schedule. "Approved by the brand lead after two consolidated review rounds" is a definition of done. "Looks premium" is not.

The shot list as a pricing document

List every shot with duration, subject, camera behaviour, and the one thing that must be readable in frame. Then mark each as hero (must be perfect), supporting (must be good), or connective (must work). Suppliers price hero shots differently because they expect more attempts, more review, and more colour matching. Making that expectation explicit protects both sides, and it gives you a way to cut scope later without renegotiating everything.

A reference board that doubles as input

Collect ten to fifteen frames that define the look: light direction, palette, lens character, texture, motion feel. Modern image-to-video workflows let those frames become the actual first frames of your shots, so a good board reduces iteration later rather than just explaining taste. A structured prompt library helps translate visual references into instructions a model can act on consistently across a sequence.

Rights, likeness, and usage boundaries

Confirm what source material you can legally use and how the output will be distributed. Third-party images, music, real people's faces, and recognisable locations all carry conditions that change with the channel: a paid advertisement, an internal training video, and a public social post are different risk profiles. Settle this before anyone builds a look on top of a borrowed image, because unwinding a look built on unclear rights is expensive in a way that no production saving can offset.

Matching the method to the shot

Not every shot deserves the same technique, and matching method to shot type is where experienced teams save the most time and money.

  • Hero shots with character performance reward a dedicated image-to-video route: lock a still you genuinely love, then animate from it instead of hoping a text prompt lands on the right face, the right light, and the right moment.
  • Establishing and atmosphere shots suit fast text-to-video generation with tight prompt control, followed by selection from multiple results rather than refinement of one.
  • Product and logo moments are often better solved with stills, motion graphics, or careful compositing, because brand marks must be exact and a soft edge on a logo is a defect, not a style.
  • Connective tissue — inserts, cutaways, texture — can be generated quickly and swapped freely in the edit without anyone noticing which take was used.

Keeping image and video generation in one workspace reduces the cost of moving between these methods, which is the practical argument for pairing Orelon's AI video generator with keyframe image generation. Starting from an existing video template also shortens the concept phase, since adapting a proven structure is faster than inventing one from a blank page.

Managing iteration before it becomes the whole budget

Iteration is where AI video projects either become efficient or become endless. Three habits keep it under control.

Acceptance criteria written before generation

For each shot, write what "good enough" means: framing, action, colour, duration, and the one detail that must be correct. Without written criteria, review drifts into taste, and taste arguments do not converge. A written criterion also gives the person doing the work permission to stop, which is worth more than it sounds.

A fixed number of review rounds

Agree on two or three rounds and define what counts as a round. A round is one consolidated set of notes, not a stream of individual messages and side comments spread across a week. This single rule prevents most scope creep, because it gives everyone a shared unit of effort and a reason to think before writing fifteen separate observations about the same six seconds.

A decision log

Record which take was approved and why. Three weeks later, when someone asks to revisit a shot, the log shows what was already settled and what the change actually costs. It also builds institutional memory: after a few projects you know that your hero shots take about twice as many attempts as you assumed, and that your finishing hours are consistently underestimated by a third.

A worked example: a sixty-second launch film

Imagine a sixty-second product film for a launch page and paid social, delivered in three aspect ratios.

The shot list totals fourteen shots: three hero shots with a person on camera, four product moments, and seven atmosphere or connective shots. Hero shots need locked keyframes, several animation attempts, and careful colour matching. Product moments are mostly stills with subtle motion. Atmosphere shots are quick generations.

Workstream What it includes Relative effort
Concept and shot list Message, tone, references, shot breakdown Low
Keyframe production Stills for hero and product shots Medium
Animation and generation Attempts, selection, retries per shot High — hero shots dominate
Edit and assembly Cutdowns for three aspect ratios Medium
Finishing Sound, titles, captions, colour, exports High

In this structure the three hero shots typically absorb more effort than the other eleven combined, and the finishing block quietly matches the generation block. A proposal that ignores either one is not a bargain; it is an incomplete estimate waiting to become a change request. If you have to choose where to reduce scope, cutting a supporting shot saves a little and cutting a hero shot saves a lot — but only if the film still makes sense without it.

Comparing proposals without being fooled by the total

Ask every supplier the same five questions before you look at the totals:

  1. How many attempts per shot are included, and what happens if a hero shot needs more?
  2. How many review rounds are included, and what counts as a round?
  3. Which deliverables are specified — aspect ratios, durations, captions, source files?
  4. Who owns the working files, keyframes, and project archive at the end?
  5. What triggers additional work, and how is it estimated in advance?

The lowest number is frequently attached to the narrowest definition of scope. A higher number with clear inclusions is usually the safer bet, because the risk sits with the person who has priced it. It also helps to understand that generation engines behave differently on different shot types; side-by-side comparisons such as Orelon vs Runway or Orelon vs Kling AI are worth reading when a shot keeps failing in one pipeline and succeeds in another.

Mistakes that quietly inflate the number

Six patterns show up again and again in projects that overspend:

  1. Briefing by vibe. "Make it feel premium" gives nobody anything to price. Reference frames and acceptance criteria do.
  2. No deliverable list. Aspect ratios, durations, and file formats agreed late turn into re-renders and re-edits that nobody planned for.
  3. Open-ended revision language. Unlimited feedback loops are the most common hidden cost in creative work, and the hardest to argue about afterwards.
  4. Changing the style mid-project. A palette or lens shift after approval invalidates everything downstream of it, including shots that were already signed off.
  5. Treating raw generations as the finished film. Generation produces material; the edit produces meaning. Budget the edit explicitly.
  6. Leaving audio until last. Music licensing and sound design chosen at the end can force picture changes you thought were locked, and a strong track can change the rhythm of the cut.

Building a repeatable internal process

Teams that produce AI video regularly stop quoting per project and start maintaining a small production system: a shared reference library, a standard shot list template, a defined review cadence, and a naming convention for takes and versions. The system is the cost control. Once it exists, estimates become predictable, because you know from your own history how many attempts a hero shot takes in your own hands.

A useful exercise: log actual effort on three projects — attempt counts, review hours, finishing hours — and compare it against what you originally estimated. Most teams discover their instinct about generation was roughly right and their instinct about finishing was badly low. Correcting that single bias improves every future estimate more than any negotiation tactic, and it makes conversations with stakeholders far simpler, because you can point at a pattern instead of defending a feeling.

FAQ

How long should an AI video project take?

A sixty-second film with a handful of hero shots commonly needs one to three weeks including concept, generation, edit, and finishing. Character-heavy work or multi-language delivery extends that. Generation itself is rarely the bottleneck; approvals and finishing are.

Is it cheaper to generate more shots than I need?

Sometimes. Extra coverage gives the edit room to breathe, but every shot adds selection and continuity work. The efficient middle ground is to over-generate only hero and connective shots, where optionality genuinely helps, and to keep supporting shots lean.

Do I need a traditional production background to estimate this well?

No, but you need discipline about scope. The people who estimate AI video well are usually the ones who write clear acceptance criteria, hold review rounds to a fixed number, and keep a decision log instead of relying on memory.

Why do two teams quote such different numbers for the same brief?

Because they are pricing different things: attempts, review rounds, resolution, finishing, and rights. Comparing totals without comparing inclusions is meaningless, and the gap usually narrows once both sides describe what is inside the number.

What should I own at the end of a project?

At minimum, the finished masters in every deliverable format, plus captions and a project archive. Where possible, negotiate access to approved keyframes and working files — they make future iterations far cheaper and let a different editor pick the film up without starting again.

How do I know a shot is finished?

When it meets the written acceptance criteria and survives one honest viewing at final size with sound. If you cannot describe why a take is wrong, it is not a note — it is a mood, and moods are expensive to chase.

Test the hard shots before you commit to a schedule

Good budgets start with good briefs. Before you ask for a single estimate, write the shot list, define what "done" means for each shot, and decide how many review rounds you can genuinely handle. Then test the pipeline yourself on the two hardest shots — the ones most likely to fail — and you will learn more about your real cost than any proposal can tell you.

You can run that test in an afternoon with Orelon: generate keyframes, animate them, compare takes, and watch how a hero shot behaves before committing to a full schedule. When you are planning something larger, browse the Orelon blog for workflow breakdowns and check Orelon pricing to map generation volume against your scope. Cinematic ideas in motion start with a clear plan — write yours, then let the tooling follow.