The Infinite Toolbox, first cut

Make me a short film for my product the same way zippr made one particular film: "The Infinite Toolbox, first cut" (zippr.to). In it, a person, seen only as two drawn hands at a desk, asks for four everyday things, and every ask pulls more drawn tools out of a small toolbox: 4 for an email, 7 for a meme, 11 for a deck, 19 for a video, until the desk has grown into a towering ink warehouse and the hands are tiny at the bottom. The sound cuts dead. "Or…" Then the warehouse fades to dotted lines, a small box with one red-orange button draws itself in, one fingertip presses it, and a film reel slides out. Everything is thin black fineliner on cream paper, drawing itself line by line, with one red-orange accent. There is no mascot in this film. Below is the exact recipe zippr used: the tools, the steps, the settings and the timings. Follow it as closely as you can. Only the parts about my product change.

WHAT IT COSTS

- The code tools in this recipe are free and open source. - zippr made it with Claude (Claude's Fable model wrote the plan, and Claude Opus 5.5 built the film) on a paid Claude plan. Any AI app that can run code can follow it. - The idea came from a chat with ChatGPT. You don't need that step: my story comes from the questions below. - The paid step: every drawing in the film is a picture made by Google's Nano Banana Pro image model, at about $0.14 a picture by usage. The film is built from 8 of them (3 new pictures and 5 edits of them). zippr made 17 in all, including wide versions it never used, for about $2.39. The look rests on these pictures. - The free route is real, and easier here than in most films, because there is no character: draw the desk, the toolbox, each tool, the warehouse and the box yourself in black fineliner on cream paper, each on its own sheet, and photograph them flat; or have the drawings made in code, which looks plainer but reveals the same way.

STEP 1: ASK ME FIRST

Before you make anything, ask me these questions. Use your question tool if you have one. Otherwise ask them one at a time in the chat. If I skip one, suggest an answer and tell me what you picked. 1. What is your product or business, in a sentence or two? 2. What is the one thing a viewer should take away? (zippr's: "You don't need to learn AI. AI needs to learn you.") 3. Who is going to watch it, and where: reels, a website, a talk? 4. Length and shape. zippr's film is 26 seconds, portrait 1080x1920, 30 frames a second. zippr also made an 8-second cut (the four asks at 1 s each, "Or…", the box, the name). Keep 26 s, or the 8 s cut? There is no landscape cut. 5. Your brand colours (hex codes if you have them), and your product's name as it should appear at the end. 6. The words on screen. zippr's: four asks as captions, a running count of tools ("1 tool", "2 tools" ... "19 tools"), "Or…", the last ask again, then "You don't need to learn AI." / "AI needs to learn you." / "zippr" / "AI that does the work." Keep the running count, or leave numbers off the screen (zippr made both versions; the page shows the one with the count)? zippr's film has no voice: the asks are captions. Do you want to record your own voice saying them? 7. Your story in zippr's shape. What are four everyday asks, each bigger than the last (zippr's: "Write the email.", "Make a meme.", "Make the deck.", "Make the video.")? How many tools does each take today (zippr's: 4, 7, 11, 19)? What container do the tools come out of (zippr's: a small wooden toolbox with a red-orange latch)? What does the pile become at the peak (zippr's: a warehouse of racks with a crane on rails)? And what does your product turn it into (zippr's: a small box with one button, and a film reel sliding out)? 8. The pictures: (a) you have them: your own ink drawings on cream paper, photographed straight on: the desk with the toolbox closed and open; the four first tools; a sheet of many single tools I can cut out one by one; the warehouse; the box with its button and slot; the thing that comes out; the hands; (b) you want them generated like zippr's with Nano Banana Pro (paid, about $0.14 a picture, about 8 pictures plus re-tries; you run it, or give me access); (c) I draw them in code (free; plainer lines, the same reveal). 9. The colours. zippr's: warm cream paper, black fineliner ink, and one red-orange accent (the latch, then the button). Put your brand colours on the colour roles in the recipe below, or keep zippr's?

STEP 2: THE RECIPE (exactly how zippr made this film)

Where the method came from ChatGPT offered it as one of four film ideas: "Start with a tiny toolbox. Every time the person wants something, another tool comes out. Email requires 4 tools. Meme requires 7. Deck requires 11. Video requires 19. The toolbox eventually becomes a giant ridiculous warehouse. Then: "Or…" Zippr. One request. Done." zippr treated it as a suggestion, not a script. Claude's Fable model wrote it up as a direction: a world, a style, a list of things never to draw, the picture prompts, and a beat sheet with every line and every time. Its choices, which you should keep: the tools are old hand tools standing in for "which AI, which editor, which format, where it goes", never software; "One request. Done." is shown as one press, not written; there is no mascot, on purpose, because the joke is the hands against the scale; the person is only hands and cuffs; no readable words inside any generated picture (every word is typeset later in code); and nothing is animated by a video model: the stills are moved in code.

The tools - Claude: Claude's Fable model wrote the direction and the picture prompts; Claude (Opus 5.5) ran the pictures, cut them into layers, wrote the compositor and the sound, and checked the frames by eye. - Nano Banana Pro (Google's gemini-3-pro-image, through Vertex AI, at 2K): drew every scene. No character reference; the later scenes got a crop of the hero picture's desk as a continuity reference. - Python with numpy, OpenCV and SciPy: cut the pictures into ink layers on one paper, worked out when each ink pixel is drawn, and composited every frame as a pure function of time. Pillow sets the words. No animation library, no browser. - ffmpeg with libx264 and AAC: frames are streamed straight in as raw video (no frame files on disk), then the sound is joined. - Python with numpy and SciPy: every sound is made from scratch. There are no recordings and no music. - The font: Geist (SemiBold and Medium), free under the SIL Open Font License, from Vercel's Geist project.

The steps, in order 1. The plan: one beats file holds every on-screen word and every time, so a change of words is a change of one row. 2. The pictures, the riskiest first: the warehouse (hundreds of small tools on flat paper is the picture most likely to come back as a 3D or photographic interior). Then the four tools, then the one button, then the edits that make each scene buildable. To fix a picture you like, EDIT it; a new prompt re-rolls everything. 3. The layers: every picture becomes ink on ONE paper (below), each moving thing its own layer with its own draw map. 4. The compositor: frame(t) builds each frame; it streams into ffmpeg and logs, per frame, how much ink was drawn and un-drawn, and every event (a tool landing, the lid, the click). 5. The sound: numpy builds the sound from that log, so the pen follows the ink and every clink lands on its tool. Mix, then join. 6. Check a contact sheet at phone size with the sound off.

The look, as colour roles (every value is zippr's) - The medium: a flat ink drawing on paper, the drawing the whole picture. One fineliner weight, slight hand wobble, flat and unshaded; cross-hatching only in the shadows of the warehouse racks, so the old way looks heavy. - The colours are ROLES. My brand colours go onto the roles; zippr's value stays for any role my colours don't fill. - PAPER, the whole frame: warm cream #EDE6D3, made in code: a faint daylight gradient from the upper left (6 levels lighter there, 5 darker at the lower right), fine grain (noise 2 levels, blurred 0.7 px) and soft fibre (noise 1.6 levels, blurred 6 px), fixed seed. Every drawing sits on this one paper. Mine: my lightest warm neutral. - INK, every drawn line: black fineliner; the words #181614. Mine: my darkest colour, kept close to black. - THE ONE ACCENT: red-orange #E4572E, on the toolbox's latch in the old way, then on the box's button. Mine: my strongest brand colour. It must stand clearly apart from the ink and the paper. - zippr's film has nothing but paper, ink and that one accent: no character, no glow, no second colour. That is zippr's rule, not mine: if my brand needs a second colour, give it one role and tell me.

The pictures (the paid step; follow this only if I chose (b)) - The model: Nano Banana Pro (Google's gemini-3-pro-image), through Vertex AI, at 2K (1536x2752 for 9:16). About $0.14 a picture by usage; Google's price page said $0.134 for a 1K or 2K picture. - Every new scene is ONE prompt built in this order: (for the later scenes) the continuity line with a crop of the hero picture's desk attached, WORLD, STYLE, NEGATIVES, then "SHOT ... draw exactly this frame:" and the shot's own line. Every edit is the picked picture sent back with one "keep the identical ..., only change: ..." line and nothing else. Sentences that direct the code (motion, type, sound, the counts) were left out of every picture prompt on purpose. - The continuity line (for the four-tools scene; the button scene's says "draw the desk, the mug and the person's hand and cuff ... There is no toolbox in this frame."): "The attached image is a crop of an earlier frame of this same film, drawn on the same sheet with the same pen: draw the desk, the mug, the toolbox and the person's hands and cuffs in exactly its drawing style and line weight. What lies on the desk, the pose of the hands, the framing and everything else follow the shot below, not the crop." - WORLD, word for word: "A large sheet of warm cream paper. The person is two drawn hands and a pair of cuffs entering the frame from the bottom edge (or the left), a mug beside them, and an off-screen voice; unhurried, deadpan, never slumped, never saved; when they leave, the hands leave with the mug. The known job is always the same object: a small solid ink box the size of a toaster, one red-orange button on top, one output slot. No prompt box, chat window, picker, editor or app ever appears. The old world is a small drawn toolbox with a red-orange latch whose contents never end: each ask lays a new row of drawn hand tools on the desk (a quill, scissors, a stamp, a funnel; then pliers, a magnifier, a spray can, a ruler, a brush, a hole punch, a small ladder; then more), the desk grows to hold them, shelves draw in above, and by the fourth ask the sheet is a cross-hatched warehouse of racks with a crane on rails." - STYLE, word for word: "Drawing style: Steinberg's line contraptions for the old world. Paper #EDE6D3. One fineliner weight, slight hand wobble, flat and unshaded; cross-hatching only on old-world machinery, so thinking looks heavy and known things look light. One accent, warm red-orange #E4572E, on the one object the frame is about (the button, a latch, a handle). Soft daylight from the upper left. In this film the only light is the daylight and the only colour the latch (then the button)." - NEGATIVES, word for word: "- No screen, phone, laptop, monitor, window of software, prompt box, chat bubble, picker, toggle, cursor or button that reads as a UI. - No readable words, letters, numerals, logos, speech bubbles or captions inside any generated frame; drawn paper carries greeked horizontal strokes; calendars have blank squares; drawers are blank. - No face, head or body of the person; hands and cuffs only; nobody tired, crying, slumped, rescued or pitied. - No glowing brains, circuit boards, robots, neural nets, sparkles, holograms, chrome, sci-fi. - No gradients, dark backgrounds, second accent colour, shading, watercolour, pencil grey or texture beyond black ink and the one accent. - No isometric 3D rooms, no photoreal interiors, no offices, cubicles, meeting rooms, server rooms, whiteboards; rooms are domestic. - No real company's interface or name, no recognisable street or old-building world, no stock startup gradients. - no vinyl toy, no creature, no figure, no face anywhere; no power tools, computers, phones or cables among the tools (hand tools only, the older the better); no tally marks or counts; no brand on any tool; the warehouse never becomes a photoreal interior; no logo on the toolbox." - The three new pictures, each shot line word for word: 1. The warehouse (9:16; drawn first, no reference picture): "Photograph of a large sheet of warm cream paper, vertical, camera straight on; drawn on it in thin black fineliner ink, slight hand wobble, flat and unshaded except dense cross-hatching in the shadows of racks, and the drawing is the whole picture: the same small desk, mug and open toolbox tiny at the bottom of the frame with a person's two drawn hands and cuffs holding up a single drawn wrench, the hands and cuffs being all of the person in the frame; above and around them, filling the sheet to the top edge, a warehouse of drawn racks and shelves stacked with hundreds of drawn hand tools, hammers, saws, funnels, brushes, clamps, spools, ladders leaning on the racks, a drawn trolley, and a drawn crane on rails across the top; the only colour in the drawing is a warm red-orange on the small toolbox's latch; soft daylight from the upper left; no words, letters, numbers or logos anywhere" 2. Four tools (9:16, with the desk crop): "Photograph of a large sheet of warm cream paper, vertical, seen from desk height; drawn on it in thin black fineliner ink, slight hand wobble, flat and unshaded, and the drawing is the whole picture: a small desk with a mug, a small drawn toolbox with its lid open and a latch on its front, four drawn hand tools laid in a neat row on the desk, a quill, a pair of scissors, a rubber stamp and a funnel, and a person's two drawn hands and cuffs rising from the bottom edge placing the funnel at the end of the row, the hands and cuffs being all of the person in the frame; the only colour in the drawing is a warm red-orange on the toolbox latch; soft daylight from the upper left; no words, letters, numbers or logos anywhere" 3. One button (9:16, with the desk crop): "Photograph of a large sheet of warm cream paper, vertical, seen from desk height; drawn on it in thin black fineliner ink, slight hand wobble, flat and unshaded, and the drawing is the whole picture: a small desk with a mug, behind it the racks of a warehouse drawn only as faint dotted construction lines fading toward the edges of the sheet; on the desk a small solid ink box the size of a toaster with one button on top and one slot, a drawn film reel sliding out of the slot, and one drawn hand and cuff pressing the button with a fingertip, the hand and cuff being all of the person in the frame; the only colour in the drawing is a warm red-orange on the button; soft daylight from the upper left; no words, letters, numbers or logos anywhere" - The five edits, each the picked picture sent back with this one line: 4. The warehouse, one accent (an edit of 1; both warehouse pictures filled the whole toolbox red-orange): "keep the identical drawing, paper, framing, light, warehouse, desk, mug, hands and wrench exactly as they are; only change: the small open toolbox on the desk is drawn like everything else on the sheet, thin black ink outlines on the cream paper with no fill, and only the small latch on its front stays a warm red-orange (#E4572E); no other colour anywhere. No words, letters, numbers or logos anywhere." 5. The warehouse with no toolbox (an edit of 4): "keep the identical drawing, paper, framing, light, warehouse, desk, mug, both hands and the wrench the hand holds up exactly as they are, every line in the same place; only change: remove the small open toolbox and the wrenches inside it from the desk completely; where it stood, the bare desk top with its wood grain and the desk's back edge continue naturally, and the left hand rests on the empty desk. No words, letters, numbers or logos anywhere." 6. The bare desk with the open toolbox (an edit of 2): "keep the identical drawing, paper, framing, light, desk, wood grain, mug and open toolbox exactly as they are, every line in the same place; only change: remove the four tools on the desk (the quill, the scissors, the rubber stamp and the funnel) and remove both hands, arms and cuffs completely; where they were, the bare desk top with its wood grain, the desk's front edge, apron and legs, and the blank cream paper below continue naturally; and the red-orange latch on the toolbox front is a plain red-orange clasp with no mark on it. No words, letters, numbers or logos anywhere." 7. The lid closed (an edit of 6): "keep the identical drawing, paper, framing, light, desk, wood grain, mug and toolbox exactly as they are, every line in the same place; only change: the toolbox's lid is closed, resting shut on top of the box, the same wood and the same plain red-orange latch on its front; where the open lid stood, blank cream paper. No words, letters, numbers or logos anywhere." 8. The box alone (an edit of 3): "keep the identical drawing, paper, framing, light, desk, wood grain, mug, box, button and dotted racks exactly as they are, every line in the same place; only change: remove the hand, arm and cuff completely, and remove the film reel and its film strip; the slot on the box front is empty; where they were, the desk top with its wood grain, the desk's front edge and the blank cream paper continue naturally. No words, letters, numbers or logos anywhere."

From pictures to layers (free, in code) - Paper and ink: for each picture, estimate its paper (at half size, a 31 px max filter removes the thin dark lines, then a 21 px blur), then un-blend every pixel into ink colour and ink amount so that pixel = amount x ink + (1 - amount) x paper, dropping paper grain under 7%. Black ink stays black and the accent keeps its colour. Every layer is then laid on the ONE code-made paper, so all the scenes share a sheet. - A moving thing = the ink in a picture that is not in its edit (the edits line up with their sources to under a pixel; the edit's ink is grown 2 px first to eat edge jitter). So: the four tools and both hands = picture 2 minus picture 6; the toolbox = picture 4 minus picture 5; the hand and the reel = picture 3 minus picture 8. Each moving thing also gets a filled silhouette, so it hides the paper under it. - The extra tools: cut one by one from the warehouse's pegboards. zippr kept the 21 that came out clean and dropped the 10 that caught part of a neighbour. - The draw map: every ink pixel gets the second it is drawn: its distance along the ink (8-connected, worked at half size) from where its stroke starts, at one pen pace (about 900 px of line a second at 1080 wide). Strokes start left to right across a stagger of 0.1 to 0.4 s, like one hand moving over the sheet, and the map is stretched to fit its slot (the desk 0.95 s, each tool 0.3 s, the toolbox 0.5 s, the box 0.6 s). The warehouse grows from its bottom edge upward along its own lines, fitted to 4.5 s. A soft pen tip of 0.07 s. The un-draw is the same map played backwards, 3 times faster. - The shelf is drawn in code: a plank and two brackets, one fineliner weight, a little wobble, drawn left to right in 0.55 s. - The warehouse's ghost: dashes along each stroke (every 22 px of line), at 26%, and none on the dense cross-hatching (dashed hatching reads as grey stipple, not as construction lines).

zippr's story, beat by beat (26.0 s, 30 frames a second; frame = seconds x 30) - 0.00 to 0.95 s: the desk, the mug and the closed toolbox draw themselves in, left to right. 0.35 s: the left hand slides up into the frame (0.5 s). Caption, 0.5 to 3.45 s: "Write the email." - 1.50 to 1.62 s: the lid pops open (the closed-lid picture gives way to the open-lid one). - 1.70 to 3.55 s: four tools, one every 0.4 s (quill, scissors, stamp, funnel): each draws itself in at the toolbox's mouth (0.25 s), then flies on a high curve to its place in a neat row (0.4 s), growing from half size and turning to rest. 2.55 s: the right hand slides in, in time for the funnel. The count, right-aligned in the caption line: "1 tool", "2 tools", "3 tools", "4 tools", each appearing as its tool lands. - 3.50 s: caption "Make a meme." The camera eases out a little (0.88x to 0.85x, 3.5 to 4.3 s). 3.9 to 6.9 s: seven more tools, one every 0.4 s, fly to a second row along the front of the desk, landing larger (up to 1.85x) and turned at random. The count restarts: "1 tool" to "7 tools". - 7.00 s: caption "Make the deck." The camera pulls back (0.85x to 0.64x, 7.0 to 8.2 s). 7.1 to 7.9 s: the desk grows 520 px wider, by repeating a 130 px slice of itself (never stretching it), and the hands move apart with it. 7.5 s: a shelf draws itself in above. 7.9 to 10.96 s: eleven tools, one every 0.25 s, alternating between the new stretch of desk and hooks under the shelf. The count: "1 tool" to "11 tools". - 11.00 s: caption "Make the video." The shelf fades (11.0 to 11.6 s). The camera pulls back over 5 s (ease in and out) while the warehouse draws itself upward from 11.45 s to 15.95 s, until the whole warehouse fills the frame and the hands are small at the bottom, one holding up a wrench. 14.0 to 14.45 s: the first desk hands over to the warehouse picture's own desk (a short cross-fade). 12.0 to 16.4 s: nineteen tools, one every 0.215 s, fly out of the toolbox up into the racks as they draw, and fade into them (0.3 s). The count: "1 tool" to "19 tools". - 17.00 s: every sound stops dead. The picture holds. 17.3 to 19.9 s: "Or…" - 18.50 s: the warehouse un-draws, top first, 3 times faster than it drew, leaving faint dotted lines. 18.8 s: the toolbox un-draws. 19.3 to 19.9 s: the small box draws itself in where the toolbox stood. - 19.90 to 20.35 s: the view pushes into the box (the button scene grows from half size into the frame, ease-out), and the warehouse fades (19.95 to 20.15 s). Caption, 20.0 to 22.2 s: "Make the video." - 20.25 to 20.85 s: one hand comes in from the lower left. 21.00 s: one fingertip presses the button (10 px down in 0.06 s, back up by 21.28 s). 21.06 to 21.6 s: the film reel slides out of the slot (hidden behind the box until it clears it). 22.05 to 22.55 s: the hand leaves. - 22.3 to 23.0 s: the camera eases back (0.72x to 0.58x) and lifts the box up the frame, making room for the words. - 22.8 s: "You don't need / to learn AI." (two lines). 23.8 s: "AI needs to learn you." Both hold to 24.62 s. 24.8 s: "zippr". 25.4 s: "AI that does the work." Both hold to the end.

The words - All type is Geist in #181614, left-aligned at x 80. Captions (they mark where a voice would go) in Medium at 58 px on a baseline at y 1830. "Or…" in SemiBold at 81 px on the same baseline. The count in Medium at 58 px, right-aligned at x 1000 on the same baseline. The end lines in SemiBold at 81 px: "You don't need / to learn AI." with its first baseline at y 1600, "AI needs to learn you." at y 1812; then "zippr" at 115 px on y 1690 and "AI that does the work." at 81 px on y 1812. - Fade in and out 0.18 s (0.25 s for the first two end lines). Words never sit on a drawn object.

The sound (no music, no voice; the desk's own sounds) - Made in numpy at 48 kHz stereo, from filtered noise and tones, with a fixed random seed. The picture's per-frame log drives it. - Under everything: paper room tone, noise low-passed at 380 Hz, at about -38 dB. - The sounds, what they are made of, and their gains in dB (before the final level set): - the pen scratch: noise band-passed 1.8 to 7 kHz with a 22 to 38 Hz grain, its loudness following how much ink is drawn in each frame (square root, scaled so the 95th-percentile frame is full), -5; - the un-draw: short scratch grains that swell and stop dead (70 ms each, one every 45 ms, like a scratch played backwards), following how much ink leaves each frame, -3; - a clink as each tool lands (metal partials at 2240, 3710, 5120 and 1180 Hz, decaying in 30 to 90 ms, over a woody knock), pitched a little differently per tool, -5 on the desk and -6 in the warehouse, panned left and right in turn; - the lid's creak at 1.5 s (dry hinge clicks, 55 to 95 a second, through a woody band of 0.5 to 2.6 kHz, over a wavering 260 Hz tone), -6; - a soft paper slide as each hand comes in, -12; - the pile-up, from 14.0 s for 3 s: a cloud of small clinks, 26 a second, thickening and getting louder (-22 to -10), under the last of the nineteen; - at 17.0 s: EVERYTHING stops in 8 ms. Room tone only until the un-draw at 18.5 s: the silence is the cue; - a paper flap as the view pushes into the box, 19.92 s, -8; - ONE click at 21.0 s (a 95 Hz falling thump plus a snap: noise above 2.5 kHz with a 1400 Hz ping), -2; - the reel's whirr at 21.06 s (0.55 s: a small motor rising from 260 to 780 Hz with its octave, sprocket ticks 38 a second, a little hiss), -7. - Every sound that goes with something you see starts 2 frames BEFORE that frame; the ear hears that as exactly on time. The slides and the whirr END on their frame instead. - The captions mark where a real voice would go: the effects dip 10 dB under each ask's first second (0.5, 3.5, 7.0, 11.0 and 20.0 s). zippr recorded no voice, and no voice is generated. - The level: the effects are set to a -4 dBFS peak before the room tone is added, with a 30 ms fade in and a 0.3 s fade out. zippr's file measures about -20 LUFS: master mine to -16 LUFS, peaks below -1 dB. - Fixes made while building it, which you should keep: - the generator filled the whole toolbox with the accent: edit it back to the latch only; - a latch drawn with a mark on it reads as a letter: ask for a plain clasp; - grow the desk by repeating a slice of it; stretching smeared its hatching into a row of dashes; - the hand-off from the first desk to the warehouse's desk reads as a dissolve, because the two desks were drawn from different angles: give the later pictures the first desk as a reference, and keep the cross-fade short.

The finish - 1080x1920, 30 fps, 26.0 s, 780 frames. H.264 High, preset slow, crf 17, yuv420p, with +faststart. The sound joined afterwards with the video copied: AAC 192 kbit/s at 48 kHz stereo. zippr's file came out at 18.3 MB.

STEP 3: SWAP IN MY PRODUCT

Keep the ink-on-paper look, the line-by-line reveal, the tools flying out of one container, the desk that grows into a warehouse, the dead silence, the timings and the sound design exactly. Change only these: - The four asks become mine (question 7), and the counts become mine. - The toolbox becomes the container of my old way, and the tools become the things my asks need today, drawn as old hand tools or objects, never as software. - The warehouse becomes my peak of absurdity. - The box with one button becomes what my product does in one press, and the film reel becomes my result. - "Or…" and the end lines become mine, and "zippr" becomes my name. - The colours go on the colour roles (question 9).

STEP 4: SHOW ME THE PLAN, THEN WAIT

Write my story in two sentences, then my beat table with times in zippr's format, then every line of words with its time, then my colours on their roles. If you can run code, show me 4 stills in the finished look (for example at 3.4 s, 10.5 s, 16.5 s and 21.3 s). If the pictures are generated, show me the warehouse first, before anything else is made. Wait for my yes before you build the whole film.

STEP 5: MAKE IT

If you are an AI app that can run code and save files, follow the recipe and give me the MP4 with sound. Ask me before installing anything. If the warehouse keeps coming back as a 3D or photographic interior, stop and tell me before making the rest: the film's peak rests on it. If Geist isn't there, use any free sans-serif and tell me. If you can't run code, build it as one self-contained web page that plays in the chat. Draw the desk, the toolbox, the tools, the racks and the box as line drawings (SVG paths) that draw themselves on by path length, fly the tools out of the box on the same seconds, pull the view back as the racks rise, and make the sounds live in the browser. Add a big Play button (browsers only play sound after a tap), a Replay button, and a "Save as video" button if the browser allows. Tell me how to screen-record it with sound on a phone and on a computer. Say plainly that it will look simpler than the photographed-paper version.

STEP 6: CHECK IT BEFORE YOU GIVE IT TO ME

- Put one frame from every 0.5 s on a contact sheet and look at it at phone size with the sound off. Could someone who never saw this message say what the film means from the frames alone? Fix what doesn't read. - Look at the generated pictures at full size: no words, no power tools or screens, only one colour besides ink. - On the file: it plays start to end; each hit sound starts 2 frames before its picture and is clearly audible; the silence from 17.0 s is real silence over the room tone; the loudness is -16 LUFS and the peaks are below -1 dB. - Then give me the file, one still, and one honest line on anything you couldn't do. Never use copyrighted music.