ChatGPT Prompts for Filmmakers (2026): 36 Prompts by Stage

Key Takeaways

• A film prompt works when it carries four things: the deliverable, the real source material, the constraints that cannot move and the exact output format.
• Ask in two passes: the scene breakdown first, then the shots or the schedule. One prompt for both is where skipped scenes and invented coverage come from.
• With the Storyflow plugin installed, end a prompt with "and put it on my Tidewater board in Storyflow" (Tidewater is the example project used throughout) and ChatGPT builds the table, shot cards or strip board there instead of in the chat.
• Through the plugin, ChatGPT builds and reads boards; it does not generate images or send call sheets, and a photo attached in the chat reaches the board only through Storyflow's upload link.
• When an answer is weak, fix the missing part of the prompt instead of asking again. Most bad output traces back to missing source text or an unstated format.

Why a Film Prompt Comes Back Generic

You ask ChatGPT for a shot list and get forty rows that would fit any commercial ever made: a wide establishing shot, a medium two-shot, a close-up for emotion. Nothing is wrong with the rows. They just are not your shoot, because nothing in the prompt was.

ChatGPT does not know your shoot until you paste it

Without the script, the brief, the locations and the days, the model fills gaps with averages. OpenAI's own prompting guide compares a GPT model to a junior coworker who performs best with explicit instructions for a specific output (OpenAI prompt engineering guide). A junior coordinator with no script writes a template. Give the same person the pages, the call times and the client's must-haves, and the list becomes a plan.

One prompt for the whole job skips scenes

"Break down this script and give me a shot list and a schedule" asks for three deliverables, each of which depends on the one before. The model compresses: scenes merge, coverage is guessed, the schedule ignores the location windows you never mentioned. Ask for the scene list, check its count against the script, then build each later deliverable from the checked one.

An answer in the chat is not a document the crew can open

Even a perfect breakdown dies six screens up in a conversation your DP cannot see. The prompts below mark where a result should leave the chat. With the Storyflow plugin for ChatGPT installed, adding one clause sends it to a board the crew and client open by link. Without it, copy the result into whatever document your production already shares.

The Four Parts of a Production Prompt

Every prompt in this guide is built from the same four parts. When an answer disappoints, one of them is usually missing.

Two habits make the four parts work harder. Number everything (scenes, setups, days), because numbered output can be checked and referred to in the next prompt. And ask ChatGPT to say what it does not know: "mark anything you assumed" turns silent guesses into a visible list you can answer.

How to Use These Prompts With the Storyflow Plugin

The prompts work in plain ChatGPT. The board clause needs the plugin, which takes a few minutes to set up once.

Install it and choose the project

Open the Storyflow listing in ChatGPT, select Install plugin, sign in to Storyflow and choose which projects ChatGPT may use: all of them, or the ones you pick. Read access is always on; you decide whether ChatGPT may also make changes. Availability depends on your ChatGPT plan, workspace and region, and in a Business or Enterprise workspace an admin manages plugins. The full steps are in how to install plugins in ChatGPT.

Name the plugin and the board in the prompt

Type @ and pick Storyflow, or choose it from the + menu, then name the project and board: "on my Tidewater project, main board". OpenAI notes that ChatGPT can also pick an installed plugin on its own when it is relevant (Plugins in ChatGPT), but naming it removes doubt when you have several installed. To decide how often ChatGPT asks before it reads or changes the board, open Settings, then Plugins, then Permissions (Managing app permissions in ChatGPT); reads allowed and changes confirmed suits most planning work.

Know what the plugin builds and what it does not

Plugin actions run on ChatGPT's model and use no Storyflow AI credits. An invited free Storyflow account can make 25 plugin actions a week; Plus, Pro and Max have no weekly limit.

Development and Treatment Prompts

These turn a client brief into an idea the client can say yes to. The board clause matters most at the end of this stage, when the treatment has to be shared.

Pitch and treatment

• 1. Five pitches from the brief. "Here is the client brief for a 45-second spot. Write five one-line pitches under 25 words each, each built on a different idea, and say which line of the brief each one answers." You get five lines, each tied to a sentence of the brief, so you can defend the pick on the call.
• 2. The treatment outline. "Using pitch 3, write a treatment for a 45-second commercial: the story in one paragraph, the tone in three adjectives with a reference film for each, the look (light, lens, colour), the sound, and the deliverables. Under 400 words. Put it on my Tidewater board in Storyflow as a document." You get a one-page document on the board the client can open by link.
• 3. The treatment as slides. "Turn the treatment document on my Tidewater board into a six-slide deck in Storyflow: title, story, look, sound, references, deliverables. One idea per slide, 30 words at most." You get a deck on the board that presents fullscreen and exports as PDF.
• 4. The client's questions. "Read the treatment on my board and list the ten questions this client is most likely to ask on the call, each with a one-sentence answer taken only from the treatment. Mark any question the treatment cannot answer." The unanswerable questions are the real output: they tell you what to decide before the call.

References and look

• 5. Real references. "Find six real reference photos for this look: dawn light on whitewashed walls, a cool grade with one warm accent, people in linen. For each, give the source page and one line on why it fits, then add them to a References wall on my Tidewater board." ChatGPT searches the web itself and the plugin places the pictures from those addresses. References are for the team; check rights before anything goes in front of an audience.
• 6. The moodboard skeleton. "Build a moodboard section on my Tidewater board with walls for light, colour, wardrobe, location and casting, and a five-colour palette for a cool dawn grade with one warm terracotta accent." You get the structure and a palette; the pictures in each wall are yours to choose or upload.

Script Breakdown Prompts

The breakdown is the document every later stage copies, so these prompts insist on counts and on marking what was assumed. The ChatGPT script breakdown page covers the board formats in more depth.

The scene pass

• 7. Scenes first, nothing else. "Here is the script. List every scene as a numbered row: scene number, INT or EXT, location, day or night, characters who appear, and length in eighths of a page. Do not list shots. End with the total scene count." Check the count against the script before anything else; a missing scene here is missing everywhere downstream.
• 8. Elements per scene. "For each scene, list cast, extras, props, wardrobe, vehicles, set dressing, special effects and sound needs. Include items the script implies but does not name, and mark those as implied." StudioBinder's breakdown guide advises going through the script a few times so nothing is missed, and notes that the colours used to mark each element category can vary (StudioBinder on breakdown colours).
• 9. The breakdown on the board. "Put the breakdown on my Tidewater board in Storyflow as a table, one row per scene, with the element columns from your last answer." You get a table the first AD and the art department read from the same link.

The department and risk passes

• 10. Split by department. "Read the breakdown table on my board and build a wall for each department (camera, art, wardrobe, sound, locations) listing what each needs, with the scene number on every item." Department heads stop reading the whole table to find their four lines.
• 11. The risk pass. "From the breakdown on my board, list every item that needs a permit, a licence, a specialist or a safety plan: drones, animals, children, water, open flame, road use, licensed music. One line each, with the scene number." This list feeds the permit prompts in the locations stage.
• 12. Revised pages. "The client sent revised pages for scenes 3 and 5, pasted below. Compare them to the breakdown table on my board, tell me which rows and elements change, then update only those rows." You get a change list in the chat and an updated table, instead of a second breakdown nobody marks as current.

Shot List and Storyboard Prompts

A shot list is one entry per camera setup, not per scene. These prompts force that, then turn the setups into shot cards with frames. For the formats themselves, see the ChatGPT shot list page.

Setups and cards

• 13. One entry per setup. "For scene 2, list one entry per camera setup: setup ID (2A, 2B), shot size, angle, movement, lens range, what the shot covers, and minutes to light and shoot. Plan the coverage a 45-second spot needs, not a feature." You get five to eight setups for a dialogue scene, each with a time you can add up.
• 14. Shot cards on the board. "Build shot cards on my Tidewater board in Storyflow for every setup in scenes 1 to 4, in shooting order, with the direction and the kit for each." You get a connected row of shot walls, each with an empty frame, a shots-required checklist, a direction note and an equipment note.
• 15. The coverage check. "Read the shot cards on my board and compare them with the script: list every line of action or dialogue no setup covers, and every setup that covers nothing in the script." This is the prompt that catches the missing insert before the shoot, not in the edit.
• 16. A, B and C lists. "We have nine working hours on day one. Using the minutes on each day-one setup, sort them into A (must have), B (should have) and C (if time), and total the minutes in each list." The totals show whether the A list alone fits, which is the conversation to have with the director today.

Frames and verticals

• 17. Reference frames for the hero setups. "For the six hero setups, find one real reference photo each on the web and place it beside its shot card on my board, captioned as reference." Through the plugin, ChatGPT places pictures from web addresses; it does not generate them. For generated previs frames, Runway's plugin makes them in the chat (Runway for ChatGPT), and you upload the keepers to the board.
• 18. Your own recce stills. "I have recce photos from three locations. Give me one upload link per location, each filling its own wall on my Tidewater board." The plugin returns a link per wall, each good for 15 minutes; upload each location's stills through its own link and they land already sorted.
• 19. Marking the verticals. "Mark which setups on my board also need a 9:16 frame for the six vertical cutdowns, and add a note reading 'Vertical 9:16' inside each of those cards." Doing this before the shoot is cheaper than reframing a 16:9 hero shot that has no room above the subject.

A commercial shot list on one board: two scenes, twelve setups, each with its frame, narration and camera notes.

Location Scouting Prompts

ChatGPT is a fast first researcher for locations and an unreliable final one. These prompts make it cite every claim so you can check it. The ChatGPT location scouting page shows the board layout in full.

Briefs and candidates

• 20. The location brief. "From the breakdown on my board, write a brief for each location: what the scene needs on camera, what it needs off camera (power, parking, toilets, a holding area), the light direction at the planned shoot time, and noise risks." You get one paragraph per location that a scout or a venue manager can answer.
• 21. Five candidates with sources. "Search the web for five rooftop terraces in Lisbon facing east, open to commercial filming. For each: name, area, why it fits the brief, who to contact and the source link. Put them in a table under a Locations wall on my Tidewater board in Storyflow." Treat every row as a lead until someone calls the venue.
• 22. The sun check. "For an east-facing rooftop in Alfama on 12 November, estimate sunrise and when direct sun clears the buildings opposite, and say how confident you are." Confirm the answer in a sun-position tool such as SunCalc, which shows sunrise, the sun's path and shadow length for a place and date.

Permits and the recce

• 23. The permit checklist. "List what the Lisbon film office needs for a two-day commercial shoot with a drone and a public terrace: the office, lead time and documents, with a source link for each claim. If you cannot find a source, say so instead of guessing." Rules differ by city. For scale, FilmLA in Los Angeles says basic permits typically take 3 to 7 business days, asks for proof of insurance, and covers up to five locations within seven days on one application (FilmLA).
• 24. The recce day. "Turn the shortlisted locations into a recce day: route order, time at each, what to photograph (the full space, power points, the sun side, access) and a checklist per location as to-dos on my board." You get to-dos per location; assign them on the board yourself.

Location research on a board: candidate frames mapped on the left, sourced research walls in the middle.

Schedule and Strip Board Prompts

Script order is almost never shoot order. These prompts collect what cannot move before ChatGPT orders anything. The ChatGPT shooting schedule page covers the strip board itself.

Constraints and the board

• 25. The immovables. "Before scheduling, list everything that cannot move: talent availability, location windows, sunrise and sunset, permit hours, the client's attendance. Ask me for anything missing instead of assuming it." You get a short list of questions back, which is the point.
• 26. The strip board. "Build a two-day strip board on my Tidewater board in Storyflow from the breakdown table: sunrise exteriors on day one, interiors on day two, grouped by location." You get a wall per day with a strip per scene, coloured by interior or exterior and day or night, and each day totalled in pages.
• 27. The day math. "For day one, list the setups in shooting order with their minutes, add 30 minutes for each company move, a 60-minute lunch and 45 minutes of contingency, and tell me whether it fits in ten hours." Simple arithmetic, but it is the arithmetic producers skip at midnight.

Changes

• 28. Rebuild after a constraint changes. "The rooftop is only available until 10:00 on day one. Rebuild the strip board on my board with that constraint and list which scenes moved." The strip board does not recalculate on its own, so a rebuild is the normal way to apply a change. The plugin does not delete anything, so remove the old strip board yourself.
• 29. What changed. "What changed on my Tidewater board since yesterday, and who changed it?" The plugin reports changes over the last day or week, which settles "who moved scene 4?" without a thread.

Call Sheet and Shoot-Day Prompts

A call sheet is the one document that must never contain a guess. These prompts make blanks visible.

The call sheet

• 30. Call sheet facts as a document. "Draft the day-one call sheet as a document on my Tidewater board: general call, location addresses, parking, nearest hospital with address, weather, the schedule by setup, and cast and crew calls. Leave every fact I have not given you as a visible blank." You get a document with blanks you fill. The designed call-sheet PDF is a job for Storyflow's in-app AI, which uses Storyflow AI credits and is separate from the plugin; neither sends the sheet. More in how to make a call sheet with ChatGPT or Claude.
• 31. The night-before message. "Write the message I send the crew tonight: call time, address, parking, what to bring and the board link. Under 120 words, no emoji." Paste it into your crew chat yourself.

On the day and after

• 32. The producer's day list. "Make a to-do list on my board for the producer on day one: releases to collect, petty cash, location sign-off, drone notification, wrap checks." To-dos take an assignee and a due date, which you set on the board; ChatGPT cannot set a to-do's assignee. To have ChatGPT fill in who does what, ask for a table with an Owner column instead: it fills that with people who already have access to the board, and a name that matches nobody stays empty.
• 33. The wrap note. "Read the day-one shot cards on my board, list the setups whose checklists are not ticked, and propose where each fits on day two." You get a carry-over list to agree with the director before the crew leaves.

Post-Production and Delivery Prompts

Post work happens in the editing plugins; the board keeps the deliverables list honest.

Selects and versions

• 34. A paper edit from the interview. "@Descript import the interview from my Tidewater folder, transcribe it and find every answer about the first harvest. Then give me a 45-second paper edit in interview order." Descript runs imports and edits in your account and spends its media minutes and AI credits (Descript help).
• 35. The version matrix. "List every deliverable: the 45-second master and six vertical cutdowns of 6 to 15 seconds, in Portuguese and English, with subtitles. Put the matrix on my board as a table, one row per file, with a status column." Fourteen rows is easier to finish than "the versions".
• 36. Subtitles. "@Kapwing add English and Portuguese subtitles to the master and the six cutdowns in my Tidewater workspace." Kapwing's connector generates, translates and dubs subtitles and spends workspace credits (Kapwing MCP).

Worked Example: One Prompt Chain for a Two-Day Brand Shoot

This chain is illustrative, written as a fictional team's project. Tidewater Pictures is a three-person production company in Lisbon making a 45-second spot and six vertical cutdowns for Folha, a sparkling-tea brand, over two days in November. The producer runs nine prompts in one ChatGPT conversation with the Storyflow plugin installed and the Tidewater project granted.

From brief to treatment (prompts 1 and 2)

Prompt 1 returns five pitches. The team picks the third, "A city waking up one terrace at a time, and the first bottle opened at sunrise", because it answers the brief's line about mornings. Prompt 2 writes the treatment as a document on the board: one paragraph of story, three tone words with a reference film each, a look section (low dawn sun, 35 to 50 mm, cool grade with terracotta accents) and a deliverables list.

From script to breakdown (prompts 7 and 9)

The client approves and the copywriter's script arrives. Prompt 7 returns seven scenes and a total of seven, matching the script. Prompt 9 puts the breakdown table on the board: seven rows, columns for cast, extras, props, wardrobe, vehicles and sound. Two items come back marked implied: a delivery bike in scene 4 and a second bottle for a pour in scene 6.

From breakdown to shot cards (prompts 13 and 14)

Prompt 13 runs scene by scene and lands on 26 setups. Prompt 14 builds them as a row of shot cards in shooting order. The director merges two pairs that share a lens and adds an insert of the cap twisting, so the row ends at 25.

Locations and schedule (prompts 21 and 26)

Prompt 21 returns five rooftop candidates with links. Two have no evidence they accept commercial shoots, so the producer drops them and calls the other three; one says yes for the morning only. Prompt 26 builds the strip board around that: four exterior scenes on day one before 10:00 and after 15:00, three interiors on day two.

The call sheet (prompt 30)

Prompt 30 drafts the call sheet as a document with five blanks: the weather, the rooftop contact's mobile, the catering drop time, the second assistant's call and the drone notification reference. The producer fills them the night before.

What the board held at the end

What needed a human: choosing the pitch, dropping two locations with no evidence, the director's merge and insert, and every blank on the call sheet. That is the right split. ChatGPT did the typing and the arithmetic; the team made every decision a client would hold them to.

When the Output Is Weak: How to Edit the Prompt

Asking "try again" changes the wording of a bad answer, not its cause. Diagnose first, then edit the one part of the prompt that is missing.

It invented a location, a person or a fact

The prompt allowed guessing. Add "use only the script and brief pasted above; if something is not there, write UNKNOWN" and, for research, "give a source link for every claim". Then ask for the list of what it marked unknown.

It skipped or merged scenes

The prompt asked for too much at once. Go back to the scene pass alone (prompt 7) and make it end with a count you can check against the script.

The shot list has one entry per scene

The prompt never defined a shot. Say "one entry per camera setup" and give an example ID scheme (2A, 2B, 2C). Five to eight entries for a dialogue scene is a sensible floor for a check.

The schedule ignores what cannot move

The constraints arrived after the request, or not at all. Run the immovables prompt (25) first and paste its answer into the scheduling prompt.

The treatment sounds like every other treatment

The prompt gave no taste. Name two reference films, three adjectives and two words to avoid ("epic", "journey"), and cap the length.

Quick Reference: Which Prompt Builds What

If you are choosing which plugins to install for the post rows, the best ChatGPT plugins for filmmakers compares ten of them by job.

What These Prompts Cannot Do for You

ChatGPT with a board is a strong coordinator. It is not a production office, and a few jobs stay with people and other tools.

Confirm anything in the real world

A location, a permit rule or a sunrise time from a chat is a lead. Someone calls the venue, reads the film office page and checks the sun tool before it goes on the call sheet.

Send, track or chase

Storyflow does not send call sheets or track confirmations, and ChatGPT cannot set a to-do's assignee; it can only fill a table's Owner column with people already on the board. For a large crew with daily call sheets, StudioBinder handles distribution and confirmations outside ChatGPT.

Make a timed animatic or a finished storyboard frame

Storyflow has no animatic or MP4 export, and the plugin does not generate images. For frames, generate them with ChatGPT's own images or Runway, then upload them to the board; for storyboard prompts specifically, see ChatGPT prompts for storyboarding.

The Bottom Line

The prompts that save a producer an evening are not clever. They paste the real pages, ask for one deliverable at a time, say what cannot move and name the output. Run the scene pass, check the count, then build the shot cards and the strip board from what you checked. Install the Storyflow plugin so each answer lands on a board the crew and client can open, and keep the decisions that a client will hold you to for yourself. Start with prompt 7 on your next script and the ChatGPT shot list page for what comes after it.