Start from the shoot, not the demo
A demo shows a tool at its best on a scene the vendor chose. A production has a scene the script chose, a cast with faces, a client with a policy and a day that is already booked. So the first step is not to open the tool but to write down what this shoot needs settled before the day: how many scenes need boards at shooting density, which sequences need time, which shots have to be designed rather than described — a crane move, a volume, a set that will be built — and which identities have to be the same across all of it. That list is the brief the software is being hired against.
The list usually points at an approach before it points at a product. A dialogue film in real locations needs boards and an animatic and little else; a commercial with a product hero and a rig needs a placed camera; a feature's set piece needs a 3D department whatever else is bought. The previs approaches page compares drawn boards, hand-built 3D and generated frames as approaches, with what each settles and what each costs in shape. Decide the approach there; then use the checklist below on any tool inside it. A tool that is excellent at the wrong approach will be excellent at not solving the problem.
The checklist
Continuity: is it measured? Generate the lead in two shots, a day apart, from different angles. Put them side by side. Then ask how the tool knew they were the same person — a style prompt, a trained model that will be stale when the wardrobe changes, or a lock the frames were made under and graded against, with a pass, a warning or a failure you can read. The second question is what happens when the lock changes: does the work made under the old one get marked, everywhere, or does the production find the stale frames by eye? The character consistency page is the long form of this criterion, and it is first on the list because it is the one every generative tool fails by default.
Cost: is it visible per shot, before the press? Every frame and every take on a generating tool is a spend, and a production makes many more than it keeps. The price should be on the control, in the unit the production thinks in — this frame, this second of take, at this resolution — before it runs, and the plan a person is on should change how many credits they have, not which rooms they can enter. A tool that walls a feature behind a tier is charging twice for the same work.
Handoff: what does the plan leave as? A crew reads documents. The shot list should leave as a spreadsheet; the boards as a PDF; the animatic as a timeline the editor opens — an editor's XML or an EDL — with the holds in place; the placed camera as numbers a grip can rig to; and the whole as one package a first assistant director reads on the morning. A plan that exports only as pictures has exported the least useful part of itself.
Engines: who operates them, where, and can the tool say no? A generating tool is a router. Your script, your cast references and your client's product are sent to a model that someone operates under some jurisdiction's law, and a studio with a confidentiality agreement or a client with a policy about where its material may go needs two things: a record per asset of which engine made it, operated by whom, in which jurisdiction; and a policy, set per project, that the tool enforces before a job is billed rather than promises afterwards. Ask to see the record for one frame. Ask what happens when the policy forbids the only engine that can do the job.
Review: can the people who decide see it without becoming users? The client, the agency, the director on a phone. A board that needs an account to view will be viewed as a screenshot and the note will come back about the wrong frame. A review link that opens the frame, lets the reader comment on it, and keeps the comment against that frame is the minimum; a record of what was approved and when is the thing the production will need six weeks later.
The record: can every frame answer where it came from? A client, a lawyer or a colleague will ask what a frame was made from. The answer should be on the frame — its references, its parents, the lock it was made under — not in someone's memory of a chat.
Leaving: what do you keep? Decide at the start what the production owns if it stops paying — the frames, the boards, the exports, the record — and whether it leaves in formats another tool reads. A plan held hostage is not a plan.
How PREVIZ answers it, and where it does not
Continuity is the Continuity Lock: a locked identity — face, wardrobe, prop, set, look — attached to every generation in every room and scored on every locked render, with a drift result a person reads, and work made under a superseded lock marked stale wherever it is. Cost is on the control: every generating action states its price and, for a take, its duration before it runs, and plans differ by credits, with every room open on every plan. Handoff is the shot list as CSV, the boards as PDF or PNG, the timeline as FCPXML, Premiere XML or EDL, the staged set as a Techvis sheet and a shot list from the camera's numbers, and the whole as the production package. Engines are routed under a per-project jurisdiction policy enforced before a job is billed, and each asset's record names the engine, its operator and the jurisdiction; the trust page and the security page set out what is live and what is a written policy. Review is a link the client opens without an account, and the record of what a frame was made from stays on the frame.
Where it does not answer: it is not a 3D department, and a feature's hardest set piece still goes to one; the staged set is a blocking pass with a camera as numbers, not a construction drawing. It does not schedule, budget or issue call sheets — the pre-production software page draws that line in full. It does not hold continuity for work made outside it. And a drift score is a measurement, not a guarantee: it tells a production where to look, and the looking is still a person's job. The toolkit page describes the rooms as a product; the glossary defines the words this checklist has leant on.
Questions people ask
What is the single most important criterion?
Whether continuity is a measurement or a hope. Any tool can make one good frame; the question is whether the lead in shot forty is the lead in shot three and how the tool knows. If the answer is a prompt that said so, the production will find out on the board; if it is a lock the frames are scored against, with a result a person can read, the production finds out before the board is shown.
How do I compare the cost of tools that price differently?
Price a shot, not a seat. Take a real scene, count the frames and takes it would need at shooting density, and ask what each tool charges to make them — including the ones that will be rejected, because there will be some. Then ask whether the price is visible before each generation runs. A tool whose cost is discovered at the end of the month is not cheaper; it is unpriced.
Why does it matter who operates the engines?
Because a generating tool routes your script, your cast references and your client's product to a model that someone runs under some jurisdiction's law. A studio with a confidentiality agreement, or a client with a policy about where its material may go, needs to know which engine, operated by whom, where — per asset — and needs the tool to refuse a job the policy forbids rather than route it and apologise.
Should a small team use the same checklist?
Yes. A small team has fewer people to catch a drifted frame, less budget to absorb an unpriced month, and a client who is just as unforgiving. The checklist is the same; the acceptable answers are simply reached at a smaller scale.