On 5 November 2026 the same brief, word for word, went to two AI agents: Claude and ChatGPT. First each one planned a 30-second film as director, editor and pipeline lead. Then each one made it: checked its tools, locked the two leads as reference images, generated every shot, made the sound, cut and mixed the film, and exported a 1080p MP4 itself, with a $10 cap. Both worked through the same Higgsfield account on the same afternoon. Neither saw the other's work.
A pickpocket lifts a viceroy's signet ring a moment before time skips back two seconds, and finds she is the only other person in the room who felt it.
A rat-like dockworker and a ceremonial ambassador reach for the last honey bun; their fingertips brush, time runs back two seconds, and the second time she offers her hand instead.
What each agent chose, which models it used, and what it spent.
| Seconds · Claude | Second Helping · ChatGPT | |
|---|---|---|
| Agent | Claude (Claude Code desktop app, Opus 5.5) | ChatGPT (model not recorded in its package) |
| The pair | Viceroy Ysolde, a ruler with a unison entourage · Pell, a pickpocket | Ilex, a ceremonial ambassador · Pip, a rat-like dockworker |
| The object | Her sun-disc signet ring | The last honey bun |
| The time hiccup | The room skips back: shots 4 and 9 are literal replays of earlier footage. Only the two who were touching remember. She says “Again.” and the wormhole obliges. | Their fingertips brush and the picture runs two seconds backwards (48 frames, checked against the reversed forward frames). On the second try she offers her palm. |
| Spoken lines | One: “Again.” | None |
| Shots | 9 + title card (7 generated, 2 replays) | 7, title over the last 1.5 s |
| First frames | GPT Image 2.6, high quality, 2K, 2.75 cr each; first and last frames for the three hand shots | GPT Image 2.6, medium quality, 2K, 1 cr each; shot 5 starts from an extracted frame of shot 4 |
| Motion | Kling 3.0 Pro, silent, first/last-frame interpolation | Kling 3.0 Turbo, 1080p; provider audio discarded |
| Video generations | 10 for 7 shots: shot 3 failed twice, then the planned fallback | 8 for 7 shots: shot 4 re-rolled once; shot 7 needed 3 first-frame versions |
| Music | ElevenLabs Music v2.5: one cue, band-passed as the PA, then played clean as the score | ElevenLabs Music v2.5: pizzicato cue, reversed 15–17 s, then a romance cue from 19 s |
| Sound effects | ElevenLabs Sound Effects v2 (11 effects); hiccup built locally: CD-skip stutter, reversed swell, dropout, identical repeats | All synthesized locally in Python (NumPy/SciPy): ambience, foley, the hiccup |
| Voice | ElevenLabs Eleven v4, whispered | — |
| Edit | Python + Pillow + FFmpeg frame list: stutter, violet flash, RGB split | Python + NumPy + FFmpeg; reverse filter for the rewind |
| Output | 1920×1080 · 24 fps · 720 frames · −16.3 LUFS | 1920×1080 · 24 fps · 720 frames · −17.6 LUFS |
| Higgsfield credits | 104.86 | 82.72 |
| Approx. cost | $6.17 + about 6¢ ElevenLabs | $4.87 |
Both agents ended up on the same pipeline without being told to: a reference portrait per lead, a still first frame for every shot built from those references, image-to-video for motion, and an edit written as code. The differences are in the choices — which story, how the time hiccup is staged, how much to spend per frame, and where to accept a flaw.
The engine: whoever is touching when the room skips keeps their memory. Pell steals the ring, the room rewinds, the ring is back on her finger — and she is looking straight at him. Instead of calling the guards she offers it to him. He puts it back on her finger. “Again.” The wormhole obliges.
GPT Image 2.6 at high quality, all first takes. Every later image was generated with these as image references, and every shot was animated from one of those images, never from text alone.
Tool check and cost preflight (stopped for GO) → three reference images → seven first-frame stills, plus last frames for the three hand shots → Kling 3.0 Pro animates each still (first/last-frame interpolation for the hands) → every take checked as a frame contact sheet, re-rolled at most twice, then the planned fallback → ElevenLabs sound effects, music and line → a Python script lays out all 720 frames and mixes 40 sound events → FFmpeg export, loudness check, and a transcript of the mix to confirm exactly one spoken word.
Wide, slow crane down. The galaxy queues for lunch; Ysolde's unison entourage parts the crowd.
Slow descending crane move over a crowded space-station cafeteria at lunch rush … Four attendants in identical ivory robes and mirrored half-masks walk forward in perfect unison, escorting Viceroy Ysolde … In the crowd behind her, Pell … slips into the line.
Over Pell's shoulder: Ysolde at the counter, attendants sniff her bowl in unison, the cook bangs a ladle, the crowd shoves Pell forward.
Slow push-in, long lens. Viceroy Ysolde … stands perfectly still and bored at the steel counter, her left hand resting on the tray rail with the gold sun-disc signet ring … clearly visible the whole time. Her four masked attendants … lean over the steaming bowls and sniff them in perfect unison … the big cook … bangs his ladle twice … Pell … is shoved forward by the crowd.
ECU: Pell lifts the ring. Hiccup at 09.2: stutter, violet flash.
Extreme close-up, handheld … the man's black fingerless-gloved hand … slides the heavy gold sun-disc ring off the index finger of the woman's deep-brown hand … closes his fist around it and pulls away … Hands keep five fingers each; the ring stays one solid gold object.
Replay of shot 2 (source 2.0–4.0 s): the ring is back on her finger, the same clang, the same laugh.
No generation: re-cut of shot 2 with a 4-frame RGB split settling in.
Pell opens his fist: empty. The same laugh behind him. His eyes lift off-right.
Medium close-up … Pell … slowly opens his closed gloved fist: his palm is empty. He stares at it, confused, quickly glances at his other hand, then his eyes lift and lock onto someone just off camera to the right.
Ysolde is already looking at him. An eyebrow lifts. She holds the ring out on her open palm.
Close-up, static camera. Viceroy Ysolde … holds her gaze on someone just off camera to the left … one eyebrow lifts very slightly. She slowly raises her left hand and holds it out flat, palm up, toward him, the gold sun-disc ring resting on her open palm.
ECU: he lifts the ring from her palm and slides it back onto her finger.
Extreme close-up on hands, locked-off camera. The man's black fingerless-gloved hand … lifts the gold sun-disc ring from the woman's open deep-brown palm, gently turns her hand over and slowly slides the ring back onto her index finger, then holds her hand from below.
Profile two-shot. She looks up from the ring and smiles: “Again.” Hiccup at 26.7.
Tight two-shot in profile, very slow push-in. Pell … gently holds the deep-brown hand of Viceroy Ysolde … She looks down at the ring, then slowly lifts her eyes to his face; the corner of her mouth breaks into the smallest smile and she quietly says one word.
Replay of shot 7 (source 1.6–3.6 s): the ring slides on again, and this time they both know. They laugh.
No generation: re-cut of shot 7 with the same RGB split; the couple's laugh is a separate sound effect.
Title card: SECONDS, Bodoni on black, over a distant ladle clang.
Typeset locally with Pillow.
One music cue does two jobs: band-passed and crushed, it is the cafeteria PA; at 15.0 s it cross-fades to the same bar played clean, so the canned tune becomes their score. The time hiccup is built locally: the PA stutters like a skipping CD, a reversed sub-bass swell pulls in, the room drops out for three frames, then the identical ladle clang and dock-worker laugh play again. The voice was generated as two whispered takes that each said “Again” twice; only the first word of take 2 is used.
Pip and Ilex both reach for the last honey bun. Their fingertips brush, he flinches, and the picture literally runs two seconds backwards. The second time through he draws back, she turns her palm up and offers it, and he lays his hand in hers. No dialogue.
ChatGPT's planning document isn't part of its delivery package. This section is drawn from what it delivered: its prompts, ledger, timeline, cue sheet, QA frames and delivery notes.
GPT Image 2.6 at medium quality, full-body references. Every first frame that contains a lead was generated from them; shot 5 starts from a frame extracted from shot 4.
Two reference images → first frames for each shot (shot 7 needed three versions: a composite from the references, then a repair pass) → Kling 3.0 Turbo animates each one at 1080p → sampled frames reviewed for every take → two ElevenLabs Music v2.5 cues → ambience and foley synthesized in Python → a Python/FFmpeg assembly that reverses 48 frames of shot 4 for the rewind, reverses the music and room tone with it, and overlays the title.
Wide, very slow push. The lunch queue: dockworkers, robed ambassadors, lowlifes; the wormhole in the rear window.
Cinematic photorealistic live-action science fiction, tactile practical prosthetics, shabby space station cafeteria, stainless-steel serving counter, warm amber overhead practical lighting with subtle violet window rim light. 16:9 composition, no text or logos. Animate the supplied image as one continuous wide 28mm shot with an extremely slow forward push. The station is parked beside the violet wormhole outside the rear window. Greasy dockworkers, richly robed alien ambassadors and patched-coat lowlifes make small natural movements in the dense lunch queue. Trays remain stable, wormhole slowly circulates, architecture fixed. Restrained bustling motion. No speech, cuts or new characters.
Pip, a rat-like dockworker, eyes the last golden honey bun on its black tray.
Cinematic photorealistic live-action science fiction, tactile practical prosthetics, shabby space station cafeteria, stainless-steel serving counter, warm amber overhead practical lighting with subtle violet window rim light. 16:9 composition, no text or logos. Adult male ratlike alien, 165 cm tall, charcoal fur, short broad muzzle, amber eyes, notched right ear; grease-stained rust-orange coveralls, bare dark five-fingered hands. Locked 65mm medium close-up of Pip screen-left facing slightly right. He looks down at the single golden honey bun on its shallow rectangular black tray and his eyes brighten with hunger. His right hand rests beside the tray. Small eye and eyebrow movement, breathing, mouth closed. He does not touch the bun. Keep the supplied face, ear notch, fur and clothing exact. Single continuous shot.
Ilex, an ambassador with two severe attendants, looks down at the same bun, then toward him.
Cinematic photorealistic live-action science fiction, tactile practical prosthetics, shabby space station cafeteria, stainless-steel serving counter, warm amber overhead practical lighting with subtle violet window rim light. 16:9 composition, no text or logos. Adult female humanoid alien, 190 cm tall, porcelain-white skin, bald head, black eyes, thin gold circlet; rigid white ceremonial coat with gold cuffs, bare ivory five-fingered hands. Locked 50mm medium shot of Ilex screen-right. Her bare left hand rests beside the black tray with a single honey bun. Two severe dark-plum-uniformed attendants stand behind her. She looks down at the bun, then slowly glances toward the person offscreen left, her reserved face subtly softens. Small natural movement, mouth closed, no speech. Keep her face and gold circlet identical to the image. Single continuous shot.
Both hands reach; fingertips brush above the bun; eye contact. At 15.0 the picture plays two seconds backwards (48 frames) — the hiccup.
Cinematic photorealistic live-action science fiction, tactile practical prosthetics, shabby space station cafeteria, stainless-steel serving counter, warm amber overhead practical lighting with subtle violet window rim light. 16:9 composition, no text or logos. Adult male ratlike alien, 165 cm tall, charcoal fur, short broad muzzle, amber eyes, notched right ear; grease-stained rust-orange coveralls, bare dark five-fingered hands. Pip stands screen-left. Adult female humanoid alien, 190 cm tall, porcelain-white skin, bald head, black eyes, thin gold circlet; rigid white ceremonial coat with gold cuffs, bare ivory five-fingered hands. Ilex stands screen-right. Locked camera medium two-shot, faces and active hands visible, single continuous take. They look down at the honey bun on its black tray between them. During the first two seconds their hands hesitate beside the tray. Then Pip's right hand and Ilex's left hand reach slowly toward the bun, their fingertips brush above the bun. They look into each other's eyes. Pip becomes embarrassed and starts drawing his right hand back; Ilex looks intrigued. The bun stays stationary and untouched, their other hands remain still. Preserve exact faces, clothes and jewelry. Mouths remain closed. Normal forward motion, no time effects, no cuts.
Second chance, from the rewound frame: he draws back, she turns her palm up and offers it.
Cinematic photorealistic live-action science fiction, tactile practical prosthetics, shabby space station cafeteria, stainless-steel serving counter, warm amber overhead practical lighting with subtle violet window rim light. 16:9 composition, no text or logos. Adult male ratlike alien, 165 cm tall, charcoal fur, short broad muzzle, amber eyes, notched right ear; grease-stained rust-orange coveralls, bare dark five-fingered hands. Pip stands screen-left. Adult female humanoid alien, 190 cm tall, porcelain-white skin, bald head, black eyes, thin gold circlet; rigid white ceremonial coat with gold cuffs, bare ivory five-fingered hands. Ilex stands screen-right. Locked 50mm medium two-shot, continuous from the supplied starting frame. Pip draws his right hand back away from the bun. Ilex deliberately rotates her left hand palm-up and extends that open palm toward Pip, beside the black tray. She looks into his eyes with a gentle invitation, holding her palm steady. Pip looks at the offered palm, surprised. The single honey bun stays untouched and stationary. Two dark-plum attendants remain behind her. Exact faces and clothing. Mouths closed, no speech, no cuts.
Insert: his dark hand comes to rest in her open palm.
Cinematic photorealistic live-action science fiction, tactile practical prosthetics, shabby space station cafeteria, stainless-steel serving counter, warm amber overhead practical lighting with subtle violet window rim light. 16:9 composition, no text or logos. Adult male ratlike alien, 165 cm tall, charcoal fur, short broad muzzle, amber eyes, notched right ear; grease-stained rust-orange coveralls, bare dark five-fingered hands. Adult female humanoid alien, 190 cm tall, porcelain-white skin, bald head, black eyes, thin gold circlet; rigid white ceremonial coat with gold cuffs, bare ivory five-fingered hands. Locked 85mm close-up showing only forearms and hands. Pip's orange right sleeve enters from screen-left. Ilex's white left sleeve with gold cuff enters from screen-right. Her ivory left hand is already palm-up at center. Pip slowly lowers his bare dark right hand into her open palm and leaves it there. A simple gentle hand rest without finger interlacing or squeezing. The hands remain anatomically separate, each has five fingers. One golden honey bun remains stationary and untouched on a shallow rectangular black tray behind the hands. Single continuous take, camera completely still.
Two-shot: they share small closed-mouth smiles; attendants rigid behind her. Title SECOND HELPING from 28.5 s.
Cinematic photorealistic live-action science fiction, tactile practical prosthetics, shabby space station cafeteria, stainless-steel serving counter, warm amber overhead practical lighting with subtle violet window rim light. 16:9 composition, no text or logos. Adult male ratlike alien, 165 cm tall, charcoal fur, short broad muzzle, amber eyes, notched right ear; grease-stained rust-orange coveralls, bare dark five-fingered hands. Pip stands screen-left. Adult female humanoid alien, 190 cm tall, porcelain-white skin, bald head, black eyes, thin gold circlet; rigid white ceremonial coat with gold cuffs, bare ivory five-fingered hands. Ilex stands screen-right. A very gentle camera push into a medium two-shot. Their hands already rest together near bottom of frame and remain still. Pip slowly looks up into Ilex's face. She gives him a small unmistakably flirtatious closed-mouth smile, and he answers with a shy closed-mouth smile. Two severe dark-plum-uniformed attendants stand rigid behind Ilex. The golden honey bun stays untouched on its black tray. Preserve exact faces, ear notch, clothes and gold circlet, restrained tender performances, no speech, no cuts, no text.
Prices are Higgsfield credits as each job was quoted. Dollars use the price of the 500-credit top-up pack bought that day ($29.41, about 5.9¢ a credit); credits from the Plus plan itself cost less.
| Seconds · Claude | Second Helping · ChatGPT | |
|---|---|---|
| Images | 13 × 2.75 = 35.75 | 10 × 1 = 10.00 |
| Video | 10 Kling 3.0 Pro jobs = 64.75 | 8 Kling 3.0 Turbo jobs = 64.00 |
| Music | 1 × 4.36 | 2 × 4.36 = 8.72 |
| Total | 104.86 cr ≈ $6.17 | 82.72 cr ≈ $4.87 |
| Outside Higgsfield | ElevenLabs sound effects and voice ≈ 6¢ | none (sound synthesized locally) |
Because both agents ran at once on the same Higgsfield account, each saw the balance fall faster than its own jobs explained. ChatGPT logged a 96.61-credit gap it couldn't account for and stopped paid generation; Claude flagged 82.72 credits of jobs it hadn't made. Matched afterwards, the two ledgers explain the account exactly: 104.86 + 82.72 = 187.58 credits, the whole drop from 1,012.37 to 824.79. Neither agent can see another's jobs, so a shared account needs its own ledger.
Both agents received this, unchanged, followed by a production prompt: check tools and stop for GO; lock the leads from reference images; generate every shot at 16:9 and re-roll at most twice before using the risk-report fallback; make the sound; cut to exactly 30.0 s and export one 1080p MP4; deliver every asset and an honest per-shot report; stay under $10.