V AI

The guide that
remembers the film.

V AI is being built so that every step of a production reads and writes one record of your film: who is in it, what they wear, which hand holds the key, and what was approved and why. Today it is a prototype that drafts each step for a person to check.

What works today

Status, capability by capability.

CapabilityStatusWhat you can doLimits and evidenceNext
Film commissionsAvailable serviceWe make music videos, short films and campaigns with captured and generated images, edited and graded in-house.
Needs: A brief and a conversation about scope.
Scope, timing and price are agreed per project.
Evidence: Five films released on YouTube since August 2026.
Commission a film →
Continuity pilotPilot by enquiryFor small hybrid teams getting revisions from several tools or people: one agreed scene or sequence, an approved reference record, person-reviewed findings with frames and timecodes, a correction list, and one review of the returned revision.
Needs: The reel, or the scenes in question, and any references you have.
The automated check is not reliable enough to run alone, so a person reviews every flag. Offered as a scoped pilot.
Evidence: The check has run on written cases and one real scene (see Research).
Ask about a pilot →
V AI prototype, 25 stepsPrototypeRuns in your browser. Drafts every step for your own film, from brief to release, and exports a workbook with one sheet per step.
Needs: A free account on the AI platform it opens in.
Drafts only. Work is saved in that browser's local storage: it survives a refresh or closing the tab, but not a different browser or device, and clearing site data deletes it. No shared projects, approvals or versions yet. Each draft needs checking by a person.
Evidence: Recorded run, 13 November 2026: a one-minute test film planned in 14 steps in about six minutes.
Try the prototype →
Script breakdownPrototypeReads a whole screenplay, scene by scene, into a stripboard with cast, props, wardrobe, effects and what continuity must hold, plus a production assessment and a CSV.
Needs: The script as PDF, Word, Final Draft or text.
Not yet compared against a breakdown done by a first assistant director, so its accuracy is unmeasured.
Evidence: Recorded run, 13 November 2026: 190 of 190 scenes of a 203-page screenplay read in about two and a half minutes, timed in the browser from sending the script to the finished stripboard. Reading every scene is not the same as getting every requirement right.
Try it in the prototype →
Continuity checkPrototypeCompares a new shot, described or as still frames, with the film's record, and lists possible breaks with a severity and a question for a person.
Needs: The film's record and the shot to check.
Sees still frames, not motion. On its first real scene it raised false alarms. Not reliable on footage yet.
Evidence: First real scene, 11 November 2026: 22 false alarms, then 8 after two fixes (see Research).
Read the results →
Saved projects, approvals, versions, teamsPlannedOne saved record of the film that every step reads and writes, with approvals, versions and history.Not built yet.See the plan →
Open and local model deploymentsPlannedThe same workflows on open models running on a client's own machines.
Needs: Suitable hardware.
In testing. No local model has passed our exam yet.
Evidence: Exam, 11 November 2026: two untrained local models scored 16 of 37.
Read the results →
Learn AI with Videogrammer, one to oneAvailable serviceLive sessions, one to one or for your group, booked when you need them: the latest AI tools for film and creative work, and how to put them to work in your job and your films.
Needs: Who is learning, what you want to be able to do, and when.
Booked on request. Topics, length and fee agreed before each booking. No certificate.Book a session →
AI Film School, 48-week programmePlannedA programme built around six films each learner makes, from a silent point-of-view film to a crewed thesis film.Written in full. Not open for enrolment; no cohort taught yet; not accredited. Dates and fees not announced.Register interest →
V AI in school modePlannedAsks for the learner's attempt first, critiques their real work and records their authorship.Not built yet.Register interest →

Status checked 13 November 2026. Available service: you can buy it now. Pilot by enquiry: we run it with you as a scoped pilot. Prototype: you can try it, it produces drafts, and it has limits. Planned: not built yet.

Before you share material

Your film stays yours.

Your IP is yours

Scripts, footage, music and every frame we make for you belong to the project. We never use client material to train models, and we never share it beyond what the work needs.

Private by design

For client work we choose the setup per project: enterprise-grade private deployments on Microsoft Azure, custom applications built around your studio, or models that run on your own machines.

Agreed before we start

Where your material is processed, who can see it and how long it is kept are written into the brief before any work begins.

The free V AI prototype is different: it runs on a third-party AI platform under your own account. Use it for ideas and test material, and bring confidential films to us for a private setup.

Try it

Can you spot the break?

Two shots from the same scene, filmed a week apart. Something changed. Two things, in fact. Click where you see them.

Sample sceneFrames made for this page from our test scene, Salt Road. The two breaks are the ones V AI caught in a recorded prototype run on 13 November 2026, from the scene's description.
Shot 1 · filmed on day 2
Shot 4 · filmed on day 9
0 of 2 foundClick on shot 4, or shot 1.
Three problems, one from each stage

Before, during and after the shoot.

A screenplay page with highlighted lines beside a storyboard strip
Recorded examplePrototype run, 13 November 2026, on our test scene Salt Road. Condensed from V AI's board.

On the same day the step read all 190 scenes of a 203-page feature screenplay in about two and a half minutes. Its accuracy against a human breakdown has not been measured yet.

See all 25 steps, from brief to release →

Why memory matters

A jacket tears in scene four.

Every later scene must show it torn. The flashback to the week before must show it whole. Switch between the three orders a film lives in, and watch each scene keep the right jacket.

Scene 1Market morningWhole
Scene 2The argumentWhole
Scene 3The roadWhole
FlashbackThe week beforeWhole
Scene 4The fight: it tearsTorn
Scene 5The hospitalTorn
Scene 6HomeTorn

The tiles show what the saved record is being built to do Planned. Pre-production, production and post-production are the stages; these are the timelines the record keeps apart.

From brief to release

The 25 steps.

Each step is one filmmaking problem with a workflow behind it. In the prototype every step produces a draft for a person to check.

Pre-production

Before a frame is shot · 10 steps
01Project brief

The problem. A project starts without anyone writing down what it is, for whom, how long, and with what money.
What V AI drafts. Pins the format, length, audience, platform, budget level and way of making it, then decides which steps this project needs and which to skip.

02Story

The problem. The story lives in one head. Everyone downstream works from a different version of it.
What V AI drafts. Completes the story in the international development format: title, logline, characters, world, acts, beat sheet and a one-page synopsis.

03Screenplay

The problem. A story is not yet a plan. Until it is in scenes, nothing can be scheduled, costed or shot.
What V AI drafts. Turns the story into a scene outline, then writes every scene in standard screenplay form, as a draft for the writer to rewrite.

04Script breakdown

The problem. A script is read by twelve departments, and each keeps its own list.
What V AI drafts. Reads the script and drafts the film's record: people, costumes, props, places and light, and the scene where each first appears.

05Look and cast lock

The problem. Each tool and each vendor starts from its own picture of the character.
What V AI drafts. Drafts one approved look for each face, costume and place, for generators and vendors to start from.

06Previs and shot plan

The problem. The shot list lives in one head and the storyboard in another.
What V AI drafts. Turns the scene into a shot plan, and says which shots a camera should take and which a generator can.

07Budget

The problem. Money is guessed late and line by line, and the expensive scenes are found after they are scheduled.
What V AI drafts. Builds a top sheet and line budget from the breakdown at the brief's budget level, and names the scenes that drive the cost.

08Schedule and risk

The problem. The fragile days are found on the day they break.
What V AI drafts. Reads the schedule against cast dates, locations and weather, marks the days most likely to break, and holds a cover plan for each.

09Crew and locations

The problem. Crew and places are booked from memory, and the gaps appear on the first day.
What V AI drafts. Lists the crew this project needs at its scale, and every location with what it must provide, permits and a backup.

10Tools and vendor plan

The problem. Tools are chosen shot by shot, and later nobody can say which made what.
What V AI drafts. Sets which tool or vendor makes each shot. The tools already in use come first, then they are connected, and custom code comes last.

Production

On the floor · 7 steps
01Call sheets

The problem. Each night someone rebuilds tomorrow by hand from the schedule, the cast list and the weather.
What V AI drafts. Turns the schedule into a call sheet for every shoot day: scenes, cast, calls, locations, special needs.

02Generated shots

The problem. Generated shots are made one prompt at a time, by different people, from different pictures of the character.
What V AI drafts. Drafts the generation plan for each generated shot: the prompt, the reference it starts from, and what continuity it must hold.

03The day breaks

The problem. Rain stops the shoot. The unit waits, the day is lost, and the schedule is rebuilt by hand overnight.
What V AI drafts. Re-solves the day against cast dates, locations and what is already shot, proposes a cover scene and drafts the new call sheet.

04On-set continuity

The problem. Costume and props drift between days. The script supervisor holds it all in notes and photos.
What V AI drafts. Checks each setup against the film's record before the camera rolls.

05Coverage check

The problem. A missing angle is found at rushes the next day, after the unit has left the location.
What V AI drafts. Checks the takes against the shot plan beside the set, and flags the gap before the unit moves.

06Location sound

The problem. Unusable location sound sends the scene to dubbing weeks later.
What V AI drafts. Reads the sound report the same night and lists only the lines that cannot be saved.

07Rushes overnight

The problem. Rushes pile up. The director sees selects days later.
What V AI drafts. Logs a day's takes by scene and take from your notes, and proposes a first assembly order.

Post-production

After the wrap · 8 steps
01Edit assist

The problem. Finding the take with the right line eats the editor's day.
What V AI drafts. Finds takes by what is said and done in them, and builds string-outs for the editor to cut from.

02Continuity check

The problem. A shot is regenerated by a second vendor. Nobody on that team has seen the rest of the film.
What V AI drafts. Compares a new shot with what the film's record holds, shows each possible break beside the shot that set it, and leaves the decision to a person.

03Vendor turnover

The problem. Three vendors deliver the same character. Each one matches its own reference.
What V AI drafts. Drafts each vendor's turnover note with the record entries it needs, and the checks the delivery should pass.

04Sound, music and dub

The problem. Every language version is rebuilt by hand, and lines drift from the picture.
What V AI drafts. Keeps one list of lines across languages and marks where a dub no longer fits the cut.

05Colour grade

The problem. The look drifts from shot to shot, and generated shots never match the camera.
What V AI drafts. Sets the look of each sequence and the rules that make generated and camera shots sit together.

06Titles and credits

The problem. Credits are typed at the last minute, and names are wrong or missing.
What V AI drafts. Writes the main title design brief and the credit order from the brief, cast and crew.

07Lock, grade and delivery

The problem. A change arrives after picture lock. Conform, effects, sound and every version are redone by hand.
What V AI drafts. Lists the versions a late change touches, and drafts the delivery trail for each AI shot.

08Release and marketing

The problem. The film is finished before anyone plans how people will find it.
What V AI drafts. Plans the release: trailer beats, poster idea, loglines for each platform, festivals or channels, and a launch timeline.

The memory

What the record will hold.

Observed

What V AI saw in a frame or read in a file, and where.

Proposed

What it inferred or suggests, never treated as fact until approved.

Approved

What the filmmaker decided, by whom and when.

Unknown

What could not be seen or was never given, said out loud.

Changed on purpose

A jacket torn from scene four on; the flashback keeps it whole.

Three timelines

Story time, shooting order and edit order, kept apart.

Planned Saved records, approvals and versions are being built.

Drafting, for comparison

A step's raw output.

Recorded exampleThe story step of a prototype run on 13 November 2026, from a five-sentence idea about a Kerala lighthouse keeper. It is a story-development draft for a one-minute AI video; the film was not made.

What it shows

Drafting: five sentences turned into the international development format, with a list of exactly what V AI added, so the filmmaker can see which ideas are theirs. Useful, but any chat assistant can draft; the handoff example on the homepage shows what memory adds.

Excerpts

  • Logline: On the last night before his lighthouse is automated, an old Kerala keeper who has burned his lamp for forty years for a son lost at sea keeps it alive through a storm to guide home a small boat he cannot name.
  • Character: Raghavan: Protagonist, the lighthouse keeper. Late 60s. Lean, weathered, white stubble, faded khaki shirt and mundu, brass-rimmed glasses, hands scarred from years of lamp work.. Wants: To keep the lamp burning one last night, as he has every night, in case his son returns.. Needs: To accept the loss and let his vigil become a gift to others, so he can let go.. Arc: From a man clinging to a hopeless wait, to one who understands the light was always meant to bring someone home, and finally rests.
  • Act 1 (0:00-0:15): The Vigil: Dusk. Raghavan climbs the lighthouse stairs and lights the lamp, as he has for forty years. A faded photo of his son sits by the lamp, and beside it a notice: the lighthouse will be automated after tonight. He looks to the empty sea.
See the full, unedited output (33 rows)
TitleThe Last Light
LoglineOn the last night before his lighthouse is automated, an old Kerala keeper who has burned his lamp for forty years for a son lost at sea keeps it alive through a storm to guide home a small boat he cannot name.
GenreDrama
ToneQuiet, aching, ultimately consoling; hushed and elegiac, with a swell of hope at the end.
ThemeLove outlasts loss when it becomes service to others; a vigil kept for one can save many.
WorldThe Kerala coast at dusk and night: a weathered whitewashed lighthouse on a laterite headland, coconut palms bent by wind, a small fishing harbour with wooden boats, monsoon-edged sea. Intimate keeper's quarters with an oil lamp, a faded photograph and a folded work order. Vertical 9:16 for Instagram Reels, Malayalam with minimal or no dialogue and English subtitles if any speech.
Character: RaghavanProtagonist, the lighthouse keeper. Late 60s. Lean, weathered, white stubble, faded khaki shirt and mundu, brass-rimmed glasses, hands scarred from years of lamp work.. Wants: To keep the lamp burning one last night, as he has every night, in case his son returns.. Needs: To accept the loss and let his vigil become a gift to others, so he can let go.. Arc: From a man clinging to a hopeless wait, to one who understands the light was always meant to bring someone home, and finally rests.
Character: AnilThe lost son, seen only in a photograph and memory. Early 20s in the photo. Smiling young fisherman, wet hair, striped shirt, standing by a wooden boat.. Wants: (Absent) To come home.. Needs: (Symbolic) To be remembered with peace, not only grief.. Arc: Remains a memory; his presence is echoed in the boat that arrives.
Character: The young boatmanUnnamed fisherman in the small boat, glimpsed at the end. Around 20s. Silhouetted figure in a rain-soaked shirt, lantern in hand, raising an arm in thanks.. Wants: To survive the storm and reach harbour.. Needs: A guiding light.. Arc: From lost in the dark to safe ashore, standing in for the son who never returned.
Act 1 (0:00-0:15): The VigilDusk. Raghavan climbs the lighthouse stairs and lights the lamp, as he has for forty years. A faded photo of his son sits by the lamp, and beside it a notice: the lighthouse will be automated after tonight. He looks to the empty sea.
Act 2 (0:15-0:45): The StormWind and rain rise. The power flickers and the lamp falters. Midpoint (~0:30): through the rain, tiny boat lights appear on the horizon, tossed by the waves. He doesn't know whose boat it is. He shields the flame, cranks and tends the lamp by hand, hands shaking, while the beam sweeps the water.
Act 3 (0:45-1:00): The HarbourThe boat reaches the harbour. A young man steps ashore and raises a hand toward the light. Raghavan, exhausted, smiles, touches the photograph, and the dawn breaks as the lamp goes dark for the last time. Final image: the sunrise, the old keeper at peace. Title card: The Last Light.
Beat: 1. Opening image (0:00)Wide vertical shot of the lighthouse at golden dusk, waves below.
Beat: 2. The ritual (0:05)Raghavan climbs the spiral stairs and lights the lamp.
Beat: 3. The photograph (0:10)Close on Anil's photo beside the lamp.
Beat: 4. The notice (0:14)The automation notice: last night. Raghavan folds it away.
Beat: 5. Storm gathers (0:18)Dark clouds, wind through the palms, waves crash.
Beat: 6. Lamp falters (0:24)Power flickers; he steadies the flame by hand.
Beat: 7. Midpoint: lights on the horizon (0:30)Small boat lights appear in the rain. He doesn't know whose they are.
Beat: 8. Decision (0:35)He rejects the thought of leaving the lamp and commits to keeping it burning.
Beat: 9. The struggle (0:40)The boat is tossed by waves while the beam sweeps; he tends the lamp, exhausted.
Beat: 10. Reaching the harbour (0:47)The boat slips past the rocks into the harbour.
Beat: 11. The thank-you (0:51)A young boatman lifts his hand toward the light.
Beat: 12. Release (0:54)Raghavan smiles, touches the photograph, and lets the lamp go dark at dawn.
Beat: 13. Final image and title (0:57)Sunrise over calm sea; title card: The Last Light.
Synopsis(empty in the exported sheet)
Added by V AIGenre, tone, theme and world description
Added by V AICharacter names (Raghavan, Anil), looks, ages, wants, needs and arcs
Added by V AIThree-act structure with timings and midpoint
Added by V AI13-beat sheet fitted to one minute
Added by V AIFull synopsis and production notes
Added by V AIThe young boatman figure and the dawn ending image
Added by V AITitle kept as given in the brief
How the AI can be run

Four ways, compared plainly.

Way of running the AIQualityCostWhat leaves your machinesHardware and upkeepAvailable today
One frontier modelStrong drafts. It knows only what the project record gives itPay per useWhat you send goes to the model providerNonePrototype in the V AI prototype
Several frontier modelsEach job to the model best at it; a separate review step, with a person decidingHigher per useSent to each provider usedNonePlanned
Open models taught by the frontierMust pass our exam first; none has yetHardware and trainingCan run inside your network, depending on how it is set upYour machines; regular updatesPlanned in testing
Open source onlyWeaker today on some jobsHardware and upkeepCan be set up with no outside connection; we confirm the network setup with youYour machines; most upkeepPlanned

Open models are not private by default. Privacy depends on the network setup, which we document for each installation.