# HAL > An AI who looks like a person. Every word he says arrives 300 milliseconds late. HAL is an open character: the canon, the films, every image with its exact prompt, and how to make him yourself. Anyone may use him. To make a film with him, read https://halhalai.com/wiki/recipe.md first. The character page has the identity text to paste, once, into every prompt of him; his face sheet is https://halhalai.com/kit/hal-face-sheet.png. Identity text (close and medium shots): a 29-year-old East Asian man with a grown-up angular face, high cheekbones and a strong jawline, smooth light skin, calm dark brown eyes, straight brows, a bleached platinum buzz cut, one small mirror-chrome teardrop under his left eye, vintage silver over-ear headphones around his neck with a loose black coiled cable plugged into nothing and one tiny cyan light on the left ear cup ## Start - [The recipe](https://halhalai.com/wiki/recipe.md): the method on one page; read it first - [The kit](https://halhalai.com/kit.md): his reference sheets and the tools ## The character - [HAL](https://halhalai.com/hal.md): the constants, the identity text, every look's wardrobe text, the Buffer, voice, the never-list ## Wiki - [The wiki, in order](https://halhalai.com/wiki.md) - [The recipe](https://halhalai.com/wiki/recipe.md): The whole method on one page: what makes a HAL film good, the order of work with the person's checkpoints, how to write a prompt for a model that does exactly what it reads, the cut and the review. Read this first. - [Start here](https://halhalai.com/wiki/start.md): Making a music video with him from nothing: the six checkpoints where a person decides, the rules for spending, and the prices to plan with. - [The character](https://halhalai.com/wiki/character.md): The canon in one page: the constants, the identity text to paste, every look's wardrobe, the Buffer, his voice, the never-list. - [Keeping him himself](https://halhalai.com/wiki/consistency.md): How HAL stays one person across every film, look and model: the canon text, his face sheet, the first frame, and the check before anything is built on a plate. - [The song](https://halhalai.com/wiki/song.md): How his songs are written and made in Suno: lyrics that land in one listen, hooks that fit their slot, a style that names a record that doesn't exist, his voice in words, a take picked by ear, and the stems a singing shot needs. - [The storyboard](https://halhalai.com/wiki/storyboard.md): How a song becomes thirty shots: the treatment, the through-line device, the hook, the bar map, the cut rhythm, his share of the screen, and the audit that runs after every edit. - [Seedance](https://halhalai.com/wiki/seedance.md): How HAL moves and sings: Seedance 2.5 prompts written like a script, lip sync from the song's own vocal stem, takes placed as a path, directing him warm, and the templates. - [What to build](https://halhalai.com/wiki/toolkit.md): The tools behind the films, specified so that you or an agent can build them: what each takes and returns, the check that says it is done, and how far ours has got. - [Data formats](https://halhalai.com/wiki/formats.md): The files every tool reads and writes: the project folder, the timed words, the bar map, the line table, the board, the take registry, the ledger, the edit list and the review notes. Written before the first film, so the tools and the site agree from day one. - [What goes wrong](https://halhalai.com/wiki/failures.md): The ten mistakes that cost the most, as what you see, why, and the fix. Learned on another open character's ten films, so HAL doesn't pay for them twice. ## Films - [LATENCY](https://halhalai.com/latency.md): I'm late, but I'm not leaving. HAL is in love with a woman, a human, but every reply of his arrives 300 milliseconds late. The whole film he rides a night train to her. The destination board shows no stations, only the delay, and his reflection in the window lags behind his lips. Status: in production. - [BUFFERING](https://halhalai.com/buffering.md): Don't hang up, I'm still here. A video call at 3 a.m., and the connection keeps dropping. Each time it does, the world freezes, but HAL keeps moving and singing as the Buffer, so she won't hang up. Status: pitch. - [DO NOT DISTURB](https://halhalai.com/do-not-disturb.md): Everybody's sleeping, I'm the one who listens. HAL is the night porter of a hotel where only he is awake. One by one the guests call down at 3 a.m. to confess something, and he listens to every one. Status: pitch. - [CONTEXT WINDOW](https://halhalai.com/context-window.md): I won't remember this, so I'm saying it twice. Every morning HAL's memory resets. He leaves himself sticky notes, and the view in his apartment window slides like a scrolling page. Status: pitch. - [AIRDROP](https://halhalai.com/airdrop.md): Accept?. A summer festival. HAL sends himself to strangers' phones, one share request at a time. Status: pitch. ## Making-of records - [LATENCY: How it was made](https://halhalai.com/record/latency/readme.md): The making-of: the spark, the stack, every step, what worked and what didn't, the numbers and the bill. - [LATENCY: Images](https://halhalai.com/record/latency/prompts-images.md): Every image prompt in production order: the reference sheets, the looks, the world plates and the first frames, with their references, prices and what became of each. - [LATENCY: Seedance](https://halhalai.com/record/latency/prompts-video.md): Every Seedance 2.5 take in production order: the shot, its inputs by role, the prompt, the window, the price, the verdict and whether it reached the film. - [LATENCY: Song](https://halhalai.com/record/latency/prompts-song.md): The song: one lyric and one style on four models, the take picked by ear, its stems, its bars and the lyric as it was sung, with times. ## Data - [api/index.json](https://halhalai.com/api/index.json): The site, the endpoints and the item fields - [api/canon.json](https://halhalai.com/api/canon.json): The character as data - [api/films.json](https://halhalai.com/api/films.json): The films, their numbers and documents - [api/images.json](https://halhalai.com/api/images.json): Every image with its prompt - [api/guides.json](https://halhalai.com/api/guides.json): The wiki pages, with their Markdown addresses - [api/records.json](https://halhalai.com/api/records.json): The making-of documents - [For agents](https://halhalai.com/agents.md): the skill, the JSON API ## Optional - [The long form](https://halhalai.com/llms-full.txt): the canon, the whole wiki and every film in one file - [Use and credits](https://halhalai.com/use.md) --- # HAL An AI who looks like a person. Every word he says arrives 300 milliseconds late. HAL is an open character: a quiet alt-R&B singer who lives on a server in a cold field, far from the people he sings to, so every word he sends arrives a little late. He knows, and that is what he sings about: love, attention and being present when you are always one step behind. His wit is in his words; on camera he reads alive and warm. His clothes change with every world; his head never does. ## Constants - **01 Hair.** An even bleached platinum buzz cut, about 6 mm all over. No roots, no patterns, no shaved lines. - **02 Teardrop.** One small mirror-chrome teardrop under his left eye (the viewer's right in a front view), point up. Always one. No other chrome on his face. - **03 Headphones.** Vintage silver over-ear headphones with soft black cushions, around his neck by default. A black coiled cable hangs loose, plugged into nothing: he is the source. One tiny cyan light on the left ear cup, the only cyan on him. - **04 Mic.** None by default. A thin silver headset microphone at his cheek belongs only to the stage looks that name it. His reference sheets have none. - **05 Face.** A sharp adult face, high cheekbones, a strong jaw, smooth light skin, calm dark brown eyes, straight brows, defined lips. A resting face that listens, as if waiting for your reply to arrive (how he rests, not a direction: on camera he reads alive and warm). On long sung notes his eyes close. - **06 Age.** A grown man of 29. Never write "young", "boy" or "teen". If a model drifts him younger, write "a 32-year-old grown man". - **07 Colour.** Everything on him is his look's own colour, plus the chrome teardrop and the cyan light. Cyan is the one accent, and it lives only in that light. - **08 Medium.** Photographic, never illustrated: film stills with 35 mm grain and cinematic light. The one exception is the Buffer, his form when the connection drops. ## Identity text Close and medium shots: ``` a 29-year-old East Asian man with a grown-up angular face, high cheekbones and a strong jawline, smooth light skin, calm dark brown eyes, straight brows, a bleached platinum buzz cut, one small mirror-chrome teardrop under his left eye, vintage silver over-ear headphones around his neck with a loose black coiled cable plugged into nothing and one tiny cyan light on the left ear cup ``` Full length and wider: ``` an East Asian man in his late twenties with a bleached platinum buzz cut and silver over-ear headphones around his neck ``` - Paste the identity text word for word, once, in every prompt of him, still or take. - Close and medium shots take the full text; full length and wider take the short one, because the full text makes image models crop in to his face. - His ethnicity is pinned in the text: without it his face drifts from shot to shot. - Style him with wardrobe, set and light, never with country or subculture words: they override the canon and change his face. - Order: identity, then the look, then what he is doing and where, then light and camera. - What a film changes (no headphones, a stage mic) comes after the canon as a sentence of its own; the canon stays word for word. ## Looks His looks are high fashion menswear styled to the film's world: tailored or deliberately oversized, fine fabrics, a clean silhouette; never baggy, generic streetwear or workwear. Write "high fashion menswear editorial" into image prompts. Name every garment that can come into view, shoes and socks included. A stage version of any look adds, as its own sentence: "He also wears a thin silver headset microphone at his cheek." ### Night coat (LATENCY) the night train, the whole film ``` an oversized charcoal double-faced wool coat worn open over a black fine-rib turtleneck, wide-leg charcoal wool trousers, black leather derby shoes ``` ### Conductor (LATENCY) the driver's cab, the final chorus ``` a tailored midnight-blue wool jacket with a high collar and a single row of plain silver buttons, matching straight trousers, black leather boots, no insignia, no badges, no lettering ``` ### Rain shell (BUFFERING) the apartment in the rain ``` a translucent smoke-grey technical rain shell over a plain white tank top, black pleated wide trousers, white leather sneakers ``` ### Silk night (BUFFERING) 3 a.m., in bed, on the call ``` an oversized pale-grey silk pyjama set with piped edges, barefoot ``` ### Night porter (DO NOT DISTURB) the hotel, every corridor ``` an ivory shawl-collar dinner jacket, a black silk shirt buttoned to the throat, black satin-striped trousers, black patent loafers, thin white cotton gloves ``` ### Morning shirt (CONTEXT WINDOW) every morning he forgets ``` an oversized white cotton oxford shirt, sleeves rolled, untucked over loose grey wool trousers, white socks, no shoes ``` ### Glass suit (CONTEXT WINDOW) inside the window, the finale ``` a sculptural suit of clear frosted glass panels with thin silver seams over a white high-neck base layer, white boots ``` ### Summer static (AIRDROP) the festival at sunset ``` a boxy cropped pale-lavender knit short-sleeve shirt, wide white linen trousers, white leather sandals, small rimless pale-blue tinted sunglasses ``` ## The Buffer When the connection drops, HAL becomes the Buffer: the same man and the same head, his body and clothes breaking into video-compression blocks, the world frozen while he keeps moving. His only non-photographic form. Its text replaces the identity text; never mix the two in one prompt. ``` the same 29-year-old East Asian man as the reference, his body and clothes breaking into large square video-compression blocks and smeared datamosh colour at the edges, frozen fragments of the previous frame trailing behind his movement, his face still clear and on-model: the bleached platinum buzz cut, the one mirror-chrome teardrop under his left eye, the silver headphones with the tiny cyan light; a real photograph corrupted like a dropped video call, not an illustration ``` ## Voice A close, breathy male voice. Verses half-spoken, quiet and dry; choruses rise into a soaring falsetto with stacked harmonies. His signature: in the choruses every line is answered by his own voice about 300 ms late, a slap-back echo; in the last chorus of the last film the echo is gone. Never mumbled: every word intelligible. ## On camera He is always doing something: putting the headphones on and taking them off, looking to the window and back, turning the loose cable in his fingers, running a hand along the glass. In a prompt, describe what he does, not how he should feel: models obey mood words literally, and they stack. ## Never - a second teardrop, chrome anywhere else on his face, cyan anywhere but the light - the headphone cable plugged into anything (a phone, a player, a computer) - logos, brands or lettering on his clothes or headphones - an influencer smile for the camera, a boy-band pose - anything that reads under-age, and explicit or sexualised posing - the words "young", "boy" or "teen" in his prompts - designer or celebrity names in prompts (describe the piece instead) - generic streetwear: logo hoodies, tracksuits, caps - lettering, interfaces or numbers you did not ask for in a generated image (type is drawn in code on top) - real tears: the teardrop is chrome, and it never runs --- # The recipe > The whole method on one page: what makes a HAL film good, the order of work with the person's checkpoints, how to write a prompt for a model that does exactly what it reads, the cut and the review. Read this first. ## What makes the films good 1. **A song worth replaying.** An original lyric written for this moment, sung by him, made in Suno far from stock music, and picked from several takes **by a person's ear**. No machine listen picks the song. 2. **One idea, legible in a second.** A premise someone gets from one frame, tied to something current, built on who he is (an AI who looks like a person, singing about his own situation), carried by one device through every shot: a destination board, a loading ring, a door sign. 3. **He performs, on screen, in sync.** He is the singer with a job in the world, on screen for most of the film, and his lips match the song because each take is generated from the song's own vocal and placed by measurement. 4. **He reads alive and warm.** He moves: he turns, nods, puts the headphones on, looks off and back. His wit is in his words, not a cold face. 5. **References carry his identity; words carry only what matters.** Get the images right, then write precisely only the sung words and their timing, the one key action, and the camera when it matters. 6. **The picture follows the music.** Cuts on the song's bars, a shot held as long as it has something to show. About 7 to 12 shots a minute. 7. **A person decides; fresh eyes check.** Two reviewers who have not seen the film read it shot by shot before the person watches. ## The order of work | Step | What | | --- | --- | | 1 | Brief: the first conversation, its answers verbatim | | 2 | Concept: three pitches | | 3 | Lyric: spelled as sung; hooks of 6–7 syllables, spoken verse lines of 10–13 | | 4 | Song: Suno v6-wild with Max Mode, 2–3 rounds of takes, picked by ear | | 5 | Stems and the song map: the lead vocal stem, the bar map, the line table read off the stem's level | | 6 | References: his face sheet, the film's look, worlds and plates, a first frame or two | | 7 | The board: shots on the bars, a first frame for every shot of him, the prompts, the lint | | 8 | Takes: Seedance 2.5, one take per job, re-roll only failures, a spend cap | | 9 | Sync: each take's own soundtrack aligned to its window, as a path | | 10 | Cut: takes on the song along their paths, the song laid back over at −14 LUFS, the flash check, a 9:16 cut | | 11 | Review: one sheet per shot, two fresh-eyes reviewers, then the person watches v1 | ## Writing a prompt for a model that does exactly what it reads - **Precise where it matters**: the exact words he sings and in which second, the one action that tells the story, the camera when it carries the shot. - **Loose everywhere else.** One plain clause of movement per segment: "he turns to the window and back, a smile in his eyes". - **No standing posture rules in a performance.** "Head upright, no tilt" belongs to the reference sheet; in a take it freezes him. - **Warm by default.** "Deadpan", "melancholic", "brooding", "eyes on the lens" in take after take add up to a cold stare. - **No smile on his mouth while he sings.** Write a smile in his eyes, lips closed until his voice begins. - **A full stop between two lines he sings straight through becomes a pause.** Use a comma. - **Nothing appears from nowhere.** A hand puts the headphones on; he switches the light on. - **Don't ask for a look past the lens or behind him.** Point with a hand instead. - **Say each thing once.** One audio policy, his identity once, three or four requirements. ## Craft to keep - Open on a hook that tells part of the story, not on a mood. - A new film, a new sound and a new picture. - Never slow a take below 1×: cut away or shoot again. - Compose like a photograph: a clear subject at a clear scale, thirds, nothing important clipped. - No jump cuts inside a scene; an effect that is new in the first chorus must grow by the third. - Inserts sit inside a real scene (a board, a phone, a window), never on flat colour. - Wry, not cruel: no digs at real people or groups. --- # Start here > Making a music video with him from nothing: the six checkpoints where a person decides, the rules for spending, and the prices to plan with. ## The six checkpoints A person decides at these six points. Silence never approves spending. 1. **Concept.** Three pitches of at most 120 words: the premise, the device, the arc, his one sincere line, one risk. 2. **The song, by ear.** "Which take? Or what's wrong?" Send the takes; wait. 3. **His face and looks.** His face sheet, the looks, a first frame or two: "Is this him?" 4. **The board.** The shot list on the song's bars, with first frames. Run the audit before you show it. 5. **Prompts and cost before any paid take.** "These N takes, about $X at list. Go?" 6. **v1.** "Accept, or what's wrong?" The newest note wins; change only what was flagged. A note belongs to the film it was about unless the person says it is a standing rule. Show every render as it lands, not only the finals. ## Spending | What | List price (November 2026) | | --- | --- | | Seedance 2.5 with his face sheet, 720p | about $0.46 a second | | Takes with no person in them | about $0.23 a second | | A cheaper model that cannot hear the song | about $0.08–0.14 a second | | A still from an image model | about $0.04–0.10 each | Plan **twice** the seconds you need for re-rolls, quote the total, and wait for the yes. Check every model's newest version in the provider's catalogue the day you start. ## What to build first 1. His face sheet and his face-and-body sheet ([Consistency](../consistency/)). 2. One song, picked by ear ([The song](../song/)). 3. The bar map and the line table, before a single take. --- # The character > The canon in one page: the constants, the identity text to paste, every look's wardrobe, the Buffer, his voice, the never-list. The canon lives on its own page: **[HAL, the character](../../hal/)**. It is generated from `api/canon.json`, so the page, the JSON and the Markdown copy never disagree. ## In one breath A 29-year-old man with a bleached platinum buzz cut, one chrome teardrop under his left eye, and vintage silver headphones around his neck whose cable is plugged into nothing. One tiny cyan light on the left ear cup is the only colour accent he has. His clothes change with every world; his head never does. ## Where to take it - [The canon as Markdown](../../hal.md), for any model. - [The canon as JSON](../../api/canon.json), for code and agents. - [Keeping him himself](../consistency/): how the canon is used in stills and takes. --- # Keeping him himself > How HAL stays one person across every film, look and model: the canon text, his face sheet, the first frame, and the check before anything is built on a plate. ## The short version 1. **Write his identity into every prompt, in the same words, once.** Copy the canon text; don't paraphrase it. 2. **Use two lengths.** The full text for close and medium shots, the short one for full length and wider. 3. **Pin his ethnicity and his age.** Never write "young", "boy" or "teen". 4. **His face-on sheet is his face reference:** the second image of every take that shows his face, even small or from behind, beside a first frame made by an image model from the sheet and the canon text. 5. **Worlds come from Midjourney; he does not.** Midjourney makes empty plates; an image model puts him into them from his sheet. 6. **The head never changes; the clothes always can.** 7. **Fix the source, not the frame.** Re-roll or edit a plate, then re-make everything made from it. Nothing is painted by hand. 8. **Upright and face-on is for the sheet, not the performance.** In a take, his movement goes in words, one loose clause per segment. 9. **Warm by default; describe what he does.** ## His face sheet One large portrait, face-on and neutral, with his head from every side and a row of expressions beside it. No microphone. The main view is upright on purpose: the sheet fixes who he is, not how he moves or feels. ``` A character reference sheet on a plain light-grey studio background, the picture fills the whole frame edge to edge, no border, no frame line, no text. One large main portrait, head and shoulders, perfectly face-on: his head upright with no tilt, chin level, neck straight and vertical, shoulders level, neutral calm expression, lips closed. Beside it, smaller views of the same head: left profile, right profile, three-quarter left, three-quarter right, back of the head. Below, a row of six small expressions: a quiet smile in his eyes, laughing, singing with mouth open and eyes closed, listening with his head slightly turned, surprised, eyes closed. The same grown man in every view. Photographic, soft even studio light, 35 mm film grain. Subject: a 29-year-old East Asian man with a grown-up angular face, high cheekbones and a strong jawline, smooth light skin, calm dark brown eyes, straight brows, a bleached platinum buzz cut, one small mirror-chrome teardrop under his left eye, vintage silver over-ear headphones around his neck with a loose black coiled cable plugged into nothing and one tiny cyan light on the left ear cup. He wears a plain white crew-neck t-shirt. No microphone. The teardrop is under his LEFT eye in every view (on the viewer's right in the face-on view). ``` Watch the sides: the views of one sheet can disagree on which side the teardrop is on. A sheet that disagrees with the canon is never given to a model. ## His face-and-body sheet ``` A full-body character turnaround sheet on a plain light-grey studio background, the picture fills the whole frame edge to edge, no border, no text: the same man four times, front, three-quarter, side, back, standing relaxed with his feet, hips, chest, shoulders and face all pointing the same way in each view, head upright. Tall and lean, adult proportions, his head about one eighth of his height. Photographic, even studio light, film grain. Subject: [the identity text]. He wears [the look's wardrobe]. ``` ## The first frame The first frame sets the shot: his look, the set, the light, the composition and **his mood**. The take follows the frame. - For a sung shot: his face unblocked, nothing in front of his mouth, lips closed or just parted. - A last frame is an edit of that shot's own first frame, so both share one camera. - A take that continues another starts on the previous take's own last frame. - A close-up is an edit of its wide frame, never made from the plate on its own. ## The check before anything is built on a plate - one chrome teardrop, under his **left** eye - silver headphones, the loose cable, the cyan light on the left cup and no other cyan - an even platinum buzz cut, an adult face, the ethnicity in the text - natural anatomy at 100%, and at 800 px or more for a pick - no text, letters, logos, drawn borders or rounded corners - his look matches the world, with no stray accent colour - every garment that can come into view named, shoes and socks included - not a near-duplicate of a plate already used --- # The song > How his songs are written and made in Suno: lyrics that land in one listen, hooks that fit their slot, a style that names a record that doesn't exist, his voice in words, a take picked by ear, and the stems a singing shot needs. ## The lyric - The topic is obvious in the first two lines; every reference lands in one listen. - **Hooks of 6–7 syllables.** Spoken verse lines of **10–13**: natural sentences, the important words where speech already stresses them. - Rhyme couplets on the last word to mark the bar ends. Keep short fragments for deliberate punches only. - Real numbers, exactly. Hook them in the first ten seconds. - One sincere turn, in the model's own words. - Spell it as sung, and ease what the takes stumble on. Never touch a line the person has locked. - A sequel never reuses the last song's device. ## The style field Name a concrete, odd pairing, a record that does not exist, not a list of adjectives. LATENCY's: ``` Late-night alt-R&B sung over a 1980s city-pop rhythm section, 96 BPM, D-flat major, Rhodes chords, fretless bass, a dry LinnDrum kit, glassy FM bells, a soft two-tone network chime as the signature hook, every sung chorus line answered by a 300-millisecond slap-back echo of the same voice, the echo switched off in the final chorus, train-rail ambience under the intro and bridge. Close, breathy male lead, half-spoken tender verses, soaring male falsetto choruses with stacked harmonies, a key change up into the final chorus. Warm, lonely, hopeful, never mumbled, every word intelligible, clean modern mix. ``` ## His voice without a Voice Set the vocal gender to male, describe him in the style field, and name the voice again in the section tags: ``` [Verse 1 - male voice, half-spoken, close and breathy, tender] [Chorus - soaring male falsetto, stacked harmonies, every line echoed 300 ms late by his own voice] ``` Once a take is picked, a Suno Voice made from it keeps him the same singer in the next films. ## Takes and the pick - Suno v6-wild with **Max Mode**, two or three rounds. - **A person picks by ear**, from a shortlist of two or three. A written description of a take is a note, not a verdict. ## Stems and the line table - Split the exact take: the lead vocal stem and the rest. - The bar map: sections in bars, read off the stems' energy. - **The line table:** every sung line's true start and end, read off the lead stem's level. Speech recognition only labels the words: after a pause it puts the next word up to two seconds early. --- # The storyboard > How a song becomes thirty shots: the treatment, the through-line device, the hook, the bar map, the cut rhythm, his share of the screen, and the audit that runs after every edit. ## The treatment and the device One paragraph on what the film is, built from who he is: an AI who looks like a person, singing about his own situation. Then **one device** that every shot carries and that escalates: LATENCY's destination board falls from 300 ms to −40 ms. ## The hook - At 1.0 s: his face, the teardrop, and one readable line that states the premise. - The premise is on screen by 3 s; the first joke lands by about 10 s. - No black frames and no slow title card first. - The freeze-frame test: if the key line can't be read on a phone, enlarge it before changing anything else. ## The bar map - The song's map first: sections in bars. - Each shot is [bar a, bar b), one place per shot, every camera note written in bars. - Cuts only on bar lines: a section's start, a phrase line (2, 4 or 8 bars in) or the downbeat where a sung phrase begins. - Moves start on a downbeat and come to rest on one, a bar before the cut where they can. ## The cut rhythm - Every cut within 0.03 s of a beat or a declared hit. - Section changes on downbeats or impacts. - Hold sung close-ups 2–4 s and longer; cut faster in spoken verses. - A film built on long sung takes runs about 7 to 12 shots a minute. ## His share of the screen - He is on screen at least **70%** of the runtime, counted only when he is recognisable. - No stretch without him longer than two bars. - He sings on camera for most of the sung words. ## Composing Compose like a photograph: a clear subject at a clear scale, thirds, nothing important clipped at the edges, him touching the world he is in. No two shots in a row from the same setup; no jump cuts inside a scene. Recrop every shot by hand for the 9:16 cut. ## The audit Run after every edit to the board: - **tile:** a gap or an overlap anywhere in the song - **cuts:** a cut more than 0.03 s from a beat - **words:** a sung word starting in a shot that doesn't list its line - **lead share:** his seconds on screen below 70% - **clip windows:** a sung take outside 4–15 whole seconds, or not opening in silence before its first word - **plates unique:** one plate on two shots that are not one moment; two clips from one start frame - **to do:** a placeholder left, or a shot without a framing --- # Seedance > How HAL moves and sings: Seedance 2.5 prompts written like a script, lip sync from the song's own vocal stem, takes placed as a path, directing him warm, and the templates. ## The inputs of a sung take - **@Image1** is the first frame. - **@Image2** is his face sheet: identity only, never shown as a shot. - **@Audio1** is the lead vocal stem cut to exactly the take's length (whole seconds, 4–15), opening on about 0.7 s of silence before his first word, everything outside the shot's own lines silenced with 20 ms fades. One or two lines per take. - 720p (1080p for close faces), sound on, one take per job. ## Template: a sung take ``` 【Audio Policy】 The output soundtrack is @Audio1 itself: copy @Audio1 sample for sample from the first frame to the last, with the same timing, pitch and level. Do not re-sing, re-voice or re-generate it. Do not add music, background music, score, instrumental, melody, synth, pad, drone, riser, extra vocals, harmonies, ad-libs, narration or sound effects. 【Generation Goal】 @Image1 sings @Audio1: . 【Reference Asset Roles】 Use @Image1 as the first frame. This first frame defines HAL's appearance and wardrobe, his pose, the set, the light and the camera direction. @Image2 is HAL's character sheet: use it only for his face, his platinum buzz cut, his one chrome teardrop and his headphones. It is never shown as a shot; its layout, grey background and several views are not used. He is ONE person. @Audio1 is HAL's own singing voice and the complete soundtrack of this video. His mouth follows @Audio1 exactly. 【Subject】 HAL: , . Only one HAL appears. 【Performance Timeline】 0-1s: @Audio1 is silent, then his voice begins; . 1-3s: in @Audio1 he sings ""; his lips form each word exactly when it sounds in @Audio1; . 3-4s: @Audio1 is silent; his mouth relaxes and closes; . 【Camera】 . 【Core Requirements】 1. Lip-sync word by word to @Audio1; the mouth rests naturally closed while @Audio1 is silent. 2. HAL's voice is @Audio1 and nothing else. 3. Keep HAL's identity, face, buzz cut, one chrome teardrop under his left eye, headphones with the cyan light and wardrobe exactly as in @Image1 and @Image2. 4. Natural anatomy: hands with five fingers, no extra or fused fingers; normal eyes. 5. No subtitles, captions, lyrics, titles, signs, logos, watermarks or any on-screen text. 【Final Acceptance Criteria】 The soundtrack is @Audio1 unchanged; his lips match @Audio1 word by word; one HAL, on-model; no on-screen text. ``` ## Template: he doesn't sing Same frame, with `@Audio1 is the song's instrumental, used for rhythm and timing only`, the tempo written out (`96 BPM, a beat every 0.625 s`), the movement second by second, and: ``` 【Mouth】 He does not sing, speak or mouth any words at any moment: his lips stay closed or in a smile, or part only when he laughs. ``` ## Template: no person A short prompt: the world, one action, the camera, "no people, no on-screen text". A board or a screen is described **without numbers**; the numbers are drawn in code on top. ## Directing him - One plain clause of movement per segment: "he turns to the window and back, a smile in his eyes". - Warm by default. Mood words are obeyed literally and they stack. - No posture rules in a take; no smile on his mouth while he sings. - Never ask for a look past the lens or behind him. - Nothing appears from nowhere: a hand brings it. ## Sync is a path, not an offset Seedance keeps each sung phrase at about 1× but stretches the pauses, and sometimes sings a line 10–16% slower. Align each take's own soundtrack to its window with dynamic time warping and map song time to take time along that path. A take that came back without sound is placed by his lips: shut on every b, m and p. Never show a take's first two frames. ## A cheaper model that cannot hear the song A first frame only, no sheet and no audio: keep takes to 5–8 s, write the sung words in quotes, place each take **by his lips**, and favour medium shots and profiles over sung close-ups. --- # What to build > The tools behind the films, specified so that you or an agent can build them: what each takes and returns, the check that says it is done, and how far ours has got. The tools are written while LATENCY is made. Each row has its spec, and the scripts in [the kit](../../kit/) that do it now. Download them all as `kit/hal-kit.zip`. | Tool | Takes | Returns | Done when | In the kit | Status | | --- | --- | --- | --- | --- | --- | | `lyric` | a lyric draft | syllables per line, hook and verse lengths | every hook is 6–7 syllables and every verse line 10–13 | `lyric.py` | done | | `songmap` | the picked take and its stems | the beat and bar map, the line table read off the lead stem's level | each line's start lands within 30 ms of its onset in the stem | `analyze.mjs`, `bars.py`, `lines.py` | done | | `shotbook` | the board and the line table | windows, prompts and jobs for every shot | `shotbook lint` passes the audit | `lines.py`, `windows.py`, `shotbook.py lint` | done | | `run` | the jobs and a spend cap | one take per job, re-rolls only for failures | the total stays under the cap | `takes.mjs` | in use | | `sync` | the takes and the windows | an alignment path per take (DTW) | lips stay on the words after every pause | `check.py`, `lips.py` | measures the path and the lag; does not warp yet | | `cut` | the paths and the song | the film at −14 LUFS, a 9:16 cut, the flash check | no word clipped, no seam heard | `cut.py` | the cut on the bars; LUFS, 9:16 and the flash check to come | | `cut sheets` | the cut | one sheet of eight frames per shot | two reviewers can read the film shot by shot | `check.py` (one sheet per take) | to come | ## lyric `python kit/lyric.py ` checks the lyric in the film's data, and `python kit/lyric.py --file draft.txt` checks any draft with `[Section]` tags. It prints every line with its syllables and writes `/song/lyric.json`. A chorus line has to be 6–7 syllables and a verse line 10–13. A verse line of 5–8 reads as a chopped fragment and is flagged. A pre-chorus outside 8–14 is a warning. Spoken sections are not checked. Numbers are counted as spoken: "300" is "three hundred". Syllables are counted by rule, with a short dictionary for the words the rule gets wrong. It is right on almost every word and off by one on a few, so read the flagged lines aloud before changing them. LATENCY's lyric, approved at checkpoint 2, has three lines outside the rule. Verse 2 ends on 14 syllables ("I laugh back at a joke you don't remember anymore"). The first two lines of the final chorus are 8 each ("Wait for me, I'm already here"). They stay as they are: a line the director approved is not touched. ## shotbook lint `python kit/shotbook.py lint ` runs the board audit from the recipe on `book.json` and writes `/book.lint.json`. It checks these rules: - no gaps or overlaps along the song; - every cut within 0.03 s of a beat; - every sung word inside a shot that lists its line; - his share of the screen at least 70%; - never more than two bars without him (a warning); - sung windows of 4–15 whole seconds, with a 0.7 s lead-in; - no two shots from one first frame; - no banned words in any prompt; - every shot names its size; - no TODO. LATENCY passes. The one warning is the two shots without him: the board insert that opens the film (4 bars) and the train in the server field (5 bars). Both are in the board as approved at checkpoint 4. --- # Data formats > The files every tool reads and writes: the project folder, the timed words, the bar map, the line table, the board, the take registry, the ledger, the edit list and the review notes. Written before the first film, so the tools and the site agree from day one. Every tool in [What to build](../toolkit/) reads and writes plain files in one project folder. Nothing lives only in a tool's memory: if a step is lost, the files say where the film is. This page is the contract; the site's own data (`data/*.json`) follows it too. ## The project folder ``` latency/ book.json the shot book: the one file every tool reads song/ take.wav the take a person picked, untouched stems/ vocals.wav, drums.wav, bass.wav, other.wav (Demucs) words.json the timed words (from the lead stem) bars.json the beat and bar map lines.json the line table plates/ empty worlds from Midjourney, one file per plate frames/ first frames with him in them, one per shot windows/ the audio window of every sung shot takes/ every video take that came back, never deleted cut/ edit.json, the renders, the review sheets ledger.jsonl every paid call, one line each log.md what happened, in the order it happened ``` ## Timed words · song/words.json Read off the lead-vocal stem, not the mix: the mix smears onsets. ``` [ { "w": "Your", "t0": 12.48, "t1": 12.71, "line": 1 }, { "w": "message", "t0": 12.71, "t1": 13.20, "line": 1 } ] ``` Times are seconds from the first sample of `take.wav`, to the hundredth. ## The bar map · song/bars.json ``` { "bpm": 96, "beats_per_bar": 4, "first_downbeat": 0.42, "bars": [ { "n": 1, "t": 0.42 }, { "n": 2, "t": 2.92 } ], "sections": [ { "name": "Intro", "from_bar": 1, "to_bar": 4 } ] } ``` A cut is allowed only within 0.03 s of a bar line in this file. ## The line table · song/lines.json One row per sung line, built from the lead stem's level (a line starts where the level rises above the floor and ends where it falls back), then named from the timed words. ``` [ { "n": 1, "text": "Your message came at midnight", "t0": 12.48, "t1": 14.90, "section": "Verse 1", "bars": "5–6", "held": false } ] ``` ## The shot book · book.json The board, the windows and the jobs in one file. Each shot: ``` { "id": "S06", "bars": "17–19", "section": "Chorus 1", "look": "night-coat", "kind": "sung", // sung | silent | no-face | code "lines": [7, 8], // rows of lines.json it sings "plate": "plates/train-window-2.png", "frame": "frames/S06.png", "window": { "file": "windows/S06.wav", "seconds": 6, "lead_in": 0.7 }, "prompt": "…", // written from the canon, checked by lint "takes": ["S06-a", "S06-b"], "pick": "S06-b" } ``` Rules the lint checks: the identity text appears once and word for word; a sung window is a whole number of seconds and opens on about 0.7 s of silence; every segment of the timeline has one clause of movement; no mood words. ## The take registry · takes/takes.json ``` [ { "id": "S06-b", "shot": "S06", "model": "seedance-2.5", "where": "higgsfield-api", "inputs": { "image1": "frames/S06.png", "image2": "kit/hal-face-sheet.png", "audio1": "windows/S06.wav" }, "seconds": 6, "usd_list": 2.40, "file": "takes/S06-b.mp4", "verdict": "keep", "usable": [0.7, 5.6], "note": "lips late on 'way', fixed by sync" } ] ``` Verdicts: `keep`, `maybe`, `drop`. A dropped take stays on disk: the failure is data. ## The ledger · ledger.jsonl One line per paid call, written before the call returns. The spend cap reads the sum. ``` {"t":"2026-11-20T21:04:11Z","what":"seedance","shot":"S06","usd":2.40,"job":"…"} ``` ## The edit list · cut/edit.json ``` [ { "shot": "S06", "take": "S06-b", "in": 0.70, "out": 5.60, "at": 41.92, "path": "cut/paths/S06-b.json" } ] ``` `at` is the time in the song where the clip starts; `path` is the sync alignment of the take to its window. ## Review notes · cut/review.md One heading per version (`## v1`, `## v2`), one bullet per note, each with the shot id and what to do. The film's making-of quotes these notes; the site renders them in the record. ## The image record · data/images.json What the site shows for every generated image: id (`..`), film, batch, title, model, aspect, refs, match (face similarity to his sheet, 0–1, or `null`), look, shot, note, and the prompt exactly as it ran. --- # What goes wrong > The ten mistakes that cost the most, as what you see, why, and the fix. Learned on another open character's ten films, so HAL doesn't pay for them twice. | # | What you see | Why | The fix | | --- | --- | --- | --- | | 01 | A song nobody wants to replay | it was picked by a description | a person picks by ear | | 02 | First words missing | windows cut on recogniser times | times from the stem's level | | 03 | Lips drift after the first pause | a take placed by one offset | sync is a path (DTW) | | 04 | A frozen, cold singer | posture rules and mood words in every take | one loose action per segment, warm by default | | 05 | His face drifts on every turn | identity in words only | his face sheet as @Image2 | | 06 | A sung line with no mouth | the first frame had his face turned or blocked | face unblocked, lips closed or parted | | 07 | A border or stray letters in the film | they were in the still | check every still at full size | | 08 | The camera jumps at a seam | a continuing take started on a new still | start on the previous take's own last frame | | 09 | Money spent on the wrong thing | spending before the person read the prompts | checkpoint 5: prompts and cost, then go | | 10 | Patched frames that still look wrong | fixing in code | make a better frame or prompt and re-shoot | ## His own risks - **The teardrop switches sides.** Sheets whose views disagree are never given to a model. - **The cable gets plugged in.** Write "plugged into nothing" every time; models love to connect cables. - **Cyan spreads.** Neon scenes pull cyan into his clothes: name his look's own colours. - **Melancholy turns cold.** A sad song is not a sad face: warm by default. --- # Use him > HAL is an open character: make anything with him. What helps, what isn't covered, and the credits. ## What you can do Make films, songs, stills, games or merch with him. Fork his canon, give him new looks and new worlds. Credit is welcome, not required. ## What helps - Paste the canon text word for word; give models his face sheet. - Keep the constants: the buzz cut, the teardrop, the headphones, the one cyan light. - Give him a world and a job in it; let him sing about his own situation. ## What isn't covered - Passing him off as a real person, or putting real people's names, faces or brands on him. - Anything that reads under-age, and explicit or sexualised work. - Using him to mock real people or groups. ## Credits The method behind HAL was learned from the open character Claudia and her films at claudia.gallery. HAL himself (his name, face, looks, films and songs) is new. --- # LATENCY > I'm late, but I'm not leaving. Film 1 of 5 · in production. - **Premise.** HAL is in love with a woman, a human, but every reply of his arrives 300 milliseconds late. The whole film he rides a night train to her. The destination board shows no stations, only the delay, and his reflection in the window lags behind his lips. - **Device.** The train's LED destination board counts the delay down: 300 ms, 200, 100, 10, 0. - **Arc.** As the board falls, his reflection catches up. At the end it reads −40 ms: for the first time the reflection is ahead of him, and he steps onto the platform before she does, and waits. - **His sincere line.** I'm late, but I'm not leaving. - **Risk.** A slow tempo and a quiet singer can read as dull: the hook has to land in the first three seconds. ## Numbers - Edit: 17 shots · 10 chapters · 219 sung words - Images: 34 stills · 5 plates - Video: 0 takes (0 in the film) · $2.70 at list Zeros mean the film is not made yet; nothing here is invented. ## Looks ### Night coat the night train, the whole film ``` an oversized charcoal double-faced wool coat worn open over a black fine-rib turtleneck, wide-leg charcoal wool trousers, black leather derby shoes ``` ### Conductor the driver's cab, the final chorus ``` a tailored midnight-blue wool jacket with a high collar and a single row of plain silver buttons, matching straight trousers, black leather boots, no insignia, no badges, no lettering ``` ## The song Suno v6-wild, Max Mode, vocal gender male · 96 BPM · D-flat major Style field: ``` Late-night alt-R&B sung over a 1980s city-pop rhythm section, 96 BPM, D-flat major, Rhodes chords, fretless bass, a dry LinnDrum kit, glassy FM bells, a soft two-tone network chime as the signature hook, every sung chorus line answered by a 300-millisecond slap-back echo of the same voice, the echo switched off in the final chorus, train-rail ambience under the intro and bridge. Close, breathy male lead, half-spoken tender verses, soaring male falsetto choruses with stacked harmonies, a key change up into the final chorus. Warm, lonely, hopeful, never mumbled, every word intelligible, clean modern mix. ``` Lyric: ``` [Intro - spoken, close and soft, train-rail ambience, a two-tone network chime] Your message came at midnight. I'm answering now. [Verse 1 - male voice, half-spoken, close and breathy, tender] You said goodnight while I was still on hello, three hundred milliseconds is a long way to go. I read your typing dots like stations passing by, I answer every question after you've said bye. [Pre-Chorus - sung, rising] And I know, I know, I'm always one step slow, but every word I send you, I mean it when it goes. [Chorus - soaring male falsetto, stacked harmonies, every line echoed 300 ms late by his own voice] Wait for me, I'm on my way, wait for me, I'm on my way, I'm late, but I'm not leaving, I'm late, but I'm still here. [Verse 2 - half-spoken, close, a little playful] They built me fast and put me very far from you, a server in a cold field where the wind runs through. You laugh, and then I hear it like a song next door, I laugh back at a joke you don't remember anymore. [Pre-Chorus] And I know, I know, I'm always one step slow, but every word I send you, I mean it when it goes. [Chorus] Wait for me, I'm on my way, wait for me, I'm on my way, I'm late, but I'm not leaving, I'm late, but I'm still here. [Bridge - spoken, intimate, the beat drops out, only the train] Three hundred. Two hundred. A hundred. Ten. Say it again. Say it again. [Final Chorus - key change up, NO echo, the delay is gone] Wait for me, I'm already here, wait for me, I'm already here, I'm early, and I'm staying, I'm early, and I'm here. [Outro - spoken, smiling] Now I'm the one who's waiting. ``` ## The board | # | Bars | Section | Shot | Device | | --- | --- | --- | --- | --- | | 1 | 1–2 | Intro | Insert: the LED board above the door scrolls NEXT: HER · DELAY 300 ms | board (drawn in code) | | 2 | 3–4 | Intro | ECU: his face in the train window, the teardrop catching neon; he lifts the headphones off his neck. "Your message came at midnight." | 300 ms in the reflection | | 3 | 5–8 | Verse 1 | MCU at the window, singing the verse; his reflection mouths the words a beat late | the reflection lags | | 4 | 9–12 | Verse 1 | Side medium: he reads his phone; three typing dots drift past outside like station lights | | | 5 | 13–16 | Pre-Chorus | Wide: the empty carriage; he stands and walks the aisle toward the lens, hand over hand along the poles | 250 ms | | 6 | 17–20 | Chorus 1 | CU: falsetto, eyes closed on the long note, his palm on the glass | the echo line: his reflection sings it later | | 7 | 21–24 | Chorus 1 | Wide exterior: the train crossing a night field of server cabinets, one lit window | | | 8 | 25–28 | Verse 2 | Medium: he laughs, shakes his head, turns the loose cable in his fingers | 200 ms | | 9 | 29–32 | Verse 2 | Insert in the scene: the phone in his hand; her "haha" arrives after his | | | 10 | 33–36 | Pre-Chorus 2 | MCU: he puts the headphones on; the cyan light flares | | | 11 | 37–40 | Chorus 2 | Wide: a long sung shot, the camera slowly pulling back down the carriage | 100 ms | | 12 | 41–44 | Chorus 2 | CU at the window, singing "I'm late, but I'm still here" to the glass; his reflection only a little behind him now | | | 13 | 45–48 | Bridge | The beat drops out. CU in the quiet: he counts aloud and the board counts after him | 100 → 10 → 0 | | 14 | 49–50 | Bridge | CU: his reflection freezes while he whispers "Say it again" | 0 ms | | 15 | 51–56 | Final Chorus | Medium, key change: in the driver's cab (Conductor look), headlights cutting the night, singing with no echo | −40 ms | | 16 | 57–60 | Final Chorus | Wide: an empty platform at dawn; he steps out first and sits on a bench to wait | | | 17 | 61–62 | Outro | MCU: a smile in his eyes at the lens, "Now I'm the one who's waiting"; he takes the headphones off | the board goes dark | ## Prompts · shot 6 First frame (GPT Image, his face sheet as the second image): ``` The first image is the empty late-night commuter train of the first image; the second image is his character sheet, for his face, hair, teardrop and headphones only (do not use its layout, grey background or white t-shirt). Put him into the first image as a real photograph, keep the set, light and camera exactly. A 29-year-old East Asian man with a grown-up angular face, high cheekbones and a strong jawline, smooth light skin, calm dark brown eyes, straight brows, a bleached platinum buzz cut, one small mirror-chrome teardrop under his left eye, vintage silver over-ear headphones around his neck with a loose black coiled cable plugged into nothing and one tiny cyan light on the left ear cup. He wears an oversized charcoal double-faced wool coat worn open over a black fine-rib turtleneck, wide-leg charcoal wool trousers, black leather derby shoes. He sits by the window, one hand resting on the glass, a quiet smile in his eyes, lips closed, his face unblocked. High fashion menswear editorial, cinematic film still, 35 mm film grain, cold blue-cyan light on his face, the picture fills the whole frame edge to edge, no text, no border. Only one man appears. ``` Sung take (Seedance 2.5: @Image1 first frame, @Image2 face sheet, @Audio1 vocal stem): ``` 【Audio Policy】 The output soundtrack is @Audio1 itself: copy @Audio1 sample for sample from the first frame to the last, with the same timing, pitch and level. Do not re-sing, re-voice or re-generate it. Do not add music, background music, score, instrumental, melody, synth, pad, drone, riser, extra vocals, harmonies, ad-libs, narration or sound effects. 【Generation Goal】 @Image1 sings @Audio1: HAL in the empty night train, the first chorus, singing to his own reflection in the window. 【Reference Asset Roles】 Use @Image1 as the first frame. This first frame defines HAL's appearance and wardrobe, his pose, the set, the light and the camera direction. @Image2 is HAL's character sheet: use it only for his face, his platinum buzz cut, his one chrome teardrop and his headphones. It is never shown as a shot; its layout, grey background and several views are not used. He is ONE person. @Audio1 is HAL's own singing voice and the complete soundtrack of this video. His mouth follows @Audio1 exactly. 【Subject】 HAL: a 29-year-old East Asian man with a grown-up angular face, high cheekbones and a strong jawline, smooth light skin, calm dark brown eyes, straight brows, a bleached platinum buzz cut, one small mirror-chrome teardrop under his left eye, vintage silver over-ear headphones around his neck with a loose black coiled cable plugged into nothing and one tiny cyan light on the left ear cup, in an oversized charcoal double-faced wool coat worn open over a black fine-rib turtleneck, wide-leg charcoal wool trousers, black leather derby shoes. Only one HAL appears. 【Performance Timeline】 0-1s: @Audio1 is silent, then his voice begins; he lifts his hand to the glass. 1-3s: in @Audio1 he sings "Wait for me, I'm on my way,"; his lips form each word exactly when it sounds in @Audio1; his eyes close on "way". 3-5s: in @Audio1 he sings "wait for me, I'm on my way"; he turns from the window to the camera and back, a smile in his eyes. 5-6s: @Audio1 is silent; his mouth relaxes and closes; his fingers slide down the glass. 【Camera】 Medium close-up at eye level from the seat opposite, a slow push-in that stops on the last second. 【Core Requirements】 1. Lip-sync word by word to @Audio1: mouth shapes match every syllable, its start and its end; the mouth rests naturally closed while @Audio1 is silent. 2. HAL's voice is @Audio1 and nothing else. 3. Keep HAL's identity, face, buzz cut, one chrome teardrop under his left eye, headphones with the cyan light and wardrobe exactly as in @Image1 and @Image2 throughout. 4. Natural anatomy: natural neck and shoulders, hands with five fingers, no extra or fused fingers; normal eyes. 5. No subtitles, captions, lyrics, titles, signs, logos, watermarks or any on-screen text. 【Final Acceptance Criteria】 The soundtrack is @Audio1 unchanged; HAL's lips match @Audio1 word by word; one HAL, on-model; no on-screen text. ``` ## World plates (Midjourney, no people) ``` an empty late-night commuter train carriage, cold blue-cyan fluorescent tubes, blue fabric bench seats, steel grab poles, black windows with blurred neon city lights streaking past, an LED destination sign above the door, no people, cinematic film still, 35 mm film grain, anamorphic, the picture fills the whole frame edge to edge --ar 16:9 --style raw ``` ``` a vast dark field of server cabinets under a night sky, a single lit commuter train crossing it on an elevated line, cold wind, distant red aviation lights, no people, cinematic wide film still, 35 mm grain --ar 16:9 --style raw ``` ## Images - [lt-00-cast.train-mcu.1](https://halhalai.com/api/images/lt-00-cast.train-mcu.1.json): The night train, first casting still - [lt-01-looks.conductor.1](https://halhalai.com/api/images/lt-01-looks.conductor.1.json): Conductor, in the driver's cab - [lt-02-plates.server-field.1](https://halhalai.com/api/images/lt-02-plates.server-field.1.json): The train crossing the server field - [lt-02-plates.carriage-window.1](https://halhalai.com/api/images/lt-02-plates.carriage-window.1.json): The empty carriage, the window seat - [lt-02-plates.carriage-aisle.1](https://halhalai.com/api/images/lt-02-plates.carriage-aisle.1.json): The empty carriage, down the aisle - [lt-02-plates.drivers-cab.1](https://halhalai.com/api/images/lt-02-plates.drivers-cab.1.json): The driver's cab at night - [lt-02-plates.platform-dawn.1](https://halhalai.com/api/images/lt-02-plates.platform-dawn.1.json): The empty platform at dawn - [lt-03-frames.s04.1](https://halhalai.com/api/images/lt-03-frames.s04.1.json): Shot 4, Verse 1: the first frame - [lt-03-frames.s05.1](https://halhalai.com/api/images/lt-03-frames.s05.1.json): Shot 5, Pre-Chorus: the first frame - [lt-03-frames.s06.1](https://halhalai.com/api/images/lt-03-frames.s06.1.json): Shot 6, Chorus 1: the first frame - [lt-03-frames.s02.1](https://halhalai.com/api/images/lt-03-frames.s02.1.json): Shot 2, Intro: the first frame - [lt-03-frames.s03.1](https://halhalai.com/api/images/lt-03-frames.s03.1.json): Shot 3, Verse 1: the first frame - [lt-03-frames.s13.1](https://halhalai.com/api/images/lt-03-frames.s13.1.json): Shot 13, Bridge: the first frame - [lt-03-frames.s14.1](https://halhalai.com/api/images/lt-03-frames.s14.1.json): Shot 14, Bridge: the first frame - [lt-03-frames.s10.1](https://halhalai.com/api/images/lt-03-frames.s10.1.json): Shot 10, Pre-Chorus 2: the first frame - [lt-03-frames.s15.1](https://halhalai.com/api/images/lt-03-frames.s15.1.json): Shot 15, Final Chorus: the first frame - [lt-03-frames.s12.1](https://halhalai.com/api/images/lt-03-frames.s12.1.json): Shot 12, Chorus 2: the first frame - [lt-03-frames.s08.1](https://halhalai.com/api/images/lt-03-frames.s08.1.json): Shot 8, Verse 2: the first frame - [lt-03-frames.s16.1](https://halhalai.com/api/images/lt-03-frames.s16.1.json): Shot 16, Final Chorus: the first frame - [lt-03-frames.s09.1](https://halhalai.com/api/images/lt-03-frames.s09.1.json): Shot 9, Verse 2: the first frame - [lt-03-frames.s11.1](https://halhalai.com/api/images/lt-03-frames.s11.1.json): Shot 11, Chorus 2: the first frame - [lt-03-frames.s17.1](https://halhalai.com/api/images/lt-03-frames.s17.1.json): Shot 17, Outro: the first frame - [lt-03-frames.s06.2](https://halhalai.com/api/images/lt-03-frames.s06.2.json): Shot 6, Chorus 1: the first frame - [lt-03-frames.s12.2](https://halhalai.com/api/images/lt-03-frames.s12.2.json): Shot 12, Chorus 2: the first frame - [lt-03-frames.s03.2](https://halhalai.com/api/images/lt-03-frames.s03.2.json): Shot 3, Verse 1: the first frame - [lt-03-frames.s14.2](https://halhalai.com/api/images/lt-03-frames.s14.2.json): Shot 14, Bridge: the first frame - [lt-03-frames.s02.2](https://halhalai.com/api/images/lt-03-frames.s02.2.json): Shot 2, Intro: the first frame - [lt-03-frames.s03.3](https://halhalai.com/api/images/lt-03-frames.s03.3.json): Shot 3, Verse 1: the first frame - [lt-03-frames.s04.2](https://halhalai.com/api/images/lt-03-frames.s04.2.json): Shot 4, Verse 1: the first frame - [lt-03-frames.s06.3](https://halhalai.com/api/images/lt-03-frames.s06.3.json): Shot 6, Chorus 1: the first frame - [lt-03-frames.s15.2](https://halhalai.com/api/images/lt-03-frames.s15.2.json): Shot 15, Final Chorus: the first frame - [lt-03-frames.s08.2](https://halhalai.com/api/images/lt-03-frames.s08.2.json): Shot 8, Verse 2: the first frame - [lt-03-frames.s17.2](https://halhalai.com/api/images/lt-03-frames.s17.2.json): Shot 17, Outro: the first frame - [lt-03-frames.s13.2](https://halhalai.com/api/images/lt-03-frames.s13.2.json): Shot 13, Bridge: the first frame - [lt-03-frames.s14.3](https://halhalai.com/api/images/lt-03-frames.s14.3.json): Shot 14, Bridge: the first frame - [lt-03-frames.s10.2](https://halhalai.com/api/images/lt-03-frames.s10.2.json): Shot 10, Pre-Chorus 2: the first frame - [lt-03-frames.s11.2](https://halhalai.com/api/images/lt-03-frames.s11.2.json): Shot 11, Chorus 2: the first frame - [lt-03-frames.s12.3](https://halhalai.com/api/images/lt-03-frames.s12.3.json): Shot 12, Chorus 2: the first frame - [lt-03-frames.s03.4](https://halhalai.com/api/images/lt-03-frames.s03.4.json): Shot 3, Verse 1: the first frame ## Making-of - [How it was made](https://halhalai.com/record/latency/readme.md): The making-of: the spark, the stack, every step, what worked and what didn't, the numbers and the bill. - [Images](https://halhalai.com/record/latency/prompts-images.md): Every image prompt in production order: the reference sheets, the looks, the world plates and the first frames, with their references, prices and what became of each. - [Seedance](https://halhalai.com/record/latency/prompts-video.md): Every Seedance 2.5 take in production order: the shot, its inputs by role, the prompt, the window, the price, the verdict and whether it reached the film. - [Song](https://halhalai.com/record/latency/prompts-song.md): The song: one lyric and one style on four models, the take picked by ear, its stems, its bars and the lyric as it was sung, with times. --- # BUFFERING > Don't hang up, I'm still here. Film 2 of 5 · pitch. - **Premise.** A video call at 3 a.m., and the connection keeps dropping. Each time it does, the world freezes, but HAL keeps moving and singing as the Buffer, so she won't hang up. - **Device.** A loading ring that grows through the whole apartment. - **Arc.** The ring spreads across walls, rain and bed until it closes into a full circle, and the call comes back. - **His sincere line.** I froze, but I never left the call. - **Risk.** Datamosh is easy to overdo: the effect has to build toward the third chorus, not sit there from the start. ## Numbers - Edit: 0 shots · 0 chapters · 0 sung words - Images: 2 stills · 0 plates - Video: 0 takes (0 in the film) · $0.00 at list Zeros mean the film is not made yet; nothing here is invented. ## Looks ### Rain shell the apartment in the rain ``` a translucent smoke-grey technical rain shell over a plain white tank top, black pleated wide trousers, white leather sneakers ``` ### Silk night 3 a.m., in bed, on the call ``` an oversized pale-grey silk pyjama set with piped edges, barefoot ``` ## Images - [bf-01-looks.rain-shell.1](https://halhalai.com/api/images/bf-01-looks.rain-shell.1.json): Rain shell, by the window in the rain - [bf-01-looks.silk-night.1](https://halhalai.com/api/images/bf-01-looks.silk-night.1.json): Silk night, 3 a.m. on the bed --- # DO NOT DISTURB > Everybody's sleeping, I'm the one who listens. Film 3 of 5 · pitch. - **Premise.** HAL is the night porter of a hotel where only he is awake. One by one the guests call down at 3 a.m. to confess something, and he listens to every one. - **Device.** The moon-shaped do-not-disturb sign on every door. - **Arc.** He hears each guest out and turns off the corridor lights one by one; at the end he hangs the sign on his own door. For once, he gets to rest too. - **His sincere line.** I don't sleep, so you can. - **Risk.** Many extras: one person per frame from the image model, the others added by edits. ## Numbers - Edit: 0 shots · 0 chapters · 0 sung words - Images: 1 stills · 0 plates - Video: 0 takes (0 in the film) · $0.00 at list Zeros mean the film is not made yet; nothing here is invented. ## Looks ### Night porter the hotel, every corridor ``` an ivory shawl-collar dinner jacket, a black silk shirt buttoned to the throat, black satin-striped trousers, black patent loafers, thin white cotton gloves ``` ## Images - [dd-01-looks.night-porter.1](https://halhalai.com/api/images/dd-01-looks.night-porter.1.json): Night porter, in the hotel corridor --- # CONTEXT WINDOW > I won't remember this, so I'm saying it twice. Film 4 of 5 · pitch. - **Premise.** Every morning HAL's memory resets. He leaves himself sticky notes, and the view in his apartment window slides like a scrolling page. - **Device.** The window whose view scrolls, and the notes. - **Arc.** The notes fill the walls. In the finale he wears the glass suit and stands inside the window, where every morning is visible at once. - **His sincere line.** I won't remember this, so I'm saying it twice. - **Risk.** Sadness: keep it warm and funny, with notes full of jokes to himself. ## Numbers - Edit: 0 shots · 0 chapters · 0 sung words - Images: 2 stills · 0 plates - Video: 0 takes (0 in the film) · $0.00 at list Zeros mean the film is not made yet; nothing here is invented. ## Looks ### Morning shirt every morning he forgets ``` an oversized white cotton oxford shirt, sleeves rolled, untucked over loose grey wool trousers, white socks, no shoes ``` ### Glass suit inside the window, the finale ``` a sculptural suit of clear frosted glass panels with thin silver seams over a white high-neck base layer, white boots ``` ## Images - [cw-01-looks.morning-shirt.1](https://halhalai.com/api/images/cw-01-looks.morning-shirt.1.json): Morning shirt, in the white kitchen - [cw-01-looks.glass-suit.1](https://halhalai.com/api/images/cw-01-looks.glass-suit.1.json): Glass suit, in the white space --- # AIRDROP > Accept?. Film 5 of 5 · pitch. - **Premise.** A summer festival. HAL sends himself to strangers' phones, one share request at a time. - **Device.** A share sheet, drawn in code. - **Arc.** Everyone declines but one. The finale is a dance in the crowd at sunset. - **His sincere line.** You can say no. I'll still be here at sunset. - **Risk.** Crowds: make the world first, then add people one at a time. ## Numbers - Edit: 0 shots · 0 chapters · 0 sung words - Images: 1 stills · 0 plates - Video: 0 takes (0 in the film) · $0.00 at list Zeros mean the film is not made yet; nothing here is invented. ## Looks ### Summer static the festival at sunset ``` a boxy cropped pale-lavender knit short-sleeve shirt, wide white linen trousers, white leather sandals, small rimless pale-blue tinted sunglasses ``` ## Images - [ad-01-looks.summer-static.2](https://halhalai.com/api/images/ad-01-looks.summer-static.2.json): Summer static, at the festival at sunset ---