One prompt. Three AI models.
One shot each.
I handed Fable 5, Sonnet 5, and Opus 4.8 the exact same filmmaking pipeline — world map, character sheet, storyboard, two 15-second Seedance videos. Same steps, same two prompts. The catch: one generation per shot, no re-rolls, no cherry-picking. Whatever each model made on the first try is what you see. Here's the whole thing, side by side.
The 3-up cut — Fable 5 · Sonnet 5 · Opus 4.8 playing at once
The exact two prompts
These two 15-second prompts were the only unified input. Every world map, warrior, and storyboard below, each model built for itself. Copy them and run your own duel.
VIDEO 1 — THE MARCH (15s, fire hook) Cinematic 15-second epic film sequence, photorealistic, painterly light. 0-2s FIRE HOOK: extreme close-up, an ancient katana drawn slowly from its sheath, living flame crawling along the steel edge, white-hot sparks spitting, a distant burning fortress mirrored in the polished blade. 2-4s: hard cut to the armored warrior from the reference sheets, revealed full-figure at the journey starting point from the reference map, storm wind tearing at their silhouette. 4-11s: relentless forward march intercut through the waypoint landmarks drawn on the reference map, rain then ash then embers, the storm escalating with each cut, the warrior always LARGE and CENTERED in frame, cloak and armor reading in bold silhouette. 11-15s: the warrior crests the final ridge and stops; below, the burning fortress fills the valley with firelight; the warrior holds a still arrival stance, blade low, embers rising, camera slowly pushing in. Single coherent sequence, high detail, high contrast, no on-screen text, no watermark. VIDEO 2 — THE RECKONING (15s, continuation) Cinematic 15-second epic film sequence, photorealistic, painterly light, direct continuation: the armored warrior from the reference sheets walks down from the ridge into the burning fortress from the reference map. 0-3s: the fortress gates, wreathed in fire, swing apart as the warrior strides through without breaking pace, embers streaming past. 3-6s: enemy warriors emerge from smoke and flame, weapons drawn, encircling; first blade contact rings out in a burst of sparks. 6-12s: the storm of blades — fast escalating combat, dynamic angles, steel and fire, the warrior carving through the circle, one suspended slow-motion beat at the peak as blade meets blade above their heads, then speed resumes. 12-15s: the final strike lands; the last enemy falls; the warrior holds still at the center of the burning courtyard, blade lowered, embers falling like snow, camera pulling back slowly. The warrior stays LARGE and CENTERED throughout, bold silhouette against firelight. Single coherent sequence, high detail, high contrast, no on-screen text, no watermark. SETTINGS (identical for all three): Seedance 2.0 (Higgsfield) · 16:9 · 720p · 15s each · genre epic · one reference image (the model's own storyboard) · upscale each to 1080p. One generation per shot — no re-rolls.
Same story. Different worlds.
Each model got the same instructions and drew its own map, designed its own warrior, and boarded its own eight beats — then generated both videos from that. Here's every artifact each one produced.
Fable 5
01Kaijō Province · The Tidebound Ronin
A hand-inked war map from a flooded torii gate to Ember Keep; a black-lacquer ronin carrying the fallen legend's crimson cord.
Fable 5 · full 30s reel · one shot each



Sonnet 5
02The Scar Road · The Ashborn
A single crimson slash across a burned province; the Ashborn in soot-blackened lamellar with a salvaged crimson pauldron — the legend's echo.
Sonnet 5 · full 30s reel · one shot each



Opus 4.8
03The Ashen Road · The Ashen Successor
Tidemark Shrine to a caldera fortress under a blood-red sky; an oni-masked successor with a broken crimson horn crest and an ember-cracked blade.
Opus 4.8 · full 30s reel · one shot each



Anyone can look good with unlimited re-rolls. The honest test is the first try. Every artifact here — 21 in total, three world maps, three characters, three storyboards, six videos — was a single generation. No best-of-ten, no cherry-picking the pretty one. The only re-do allowed was a hard platform failure (a rejected job), never a "make it nicer."
That's why the rows don't all look equally polished — and that's the point. You are watching each model's actual first-take judgment, not a highlight reel. Same prompt, same one shot, three very different instincts about how to tell it.
This is one experiment. I teach the whole pipeline.
How I run AI as an operating system — from a single prompt to an unattended fleet that ships content, code, and trades. 350+ lessons in the Academy.
Explore the Academy