absolutecinema.ai

Sci-fi AI image prompts: 40 tested frames for AI films and vertical dramas

AiDramaKing65 min read

40 sci-fi AI image prompts tested on GPT Image, Nano Banana and Seedream: baroque retro-futurism, body horror, art deco, steampunk, cyborg and biopunk women. Each with its frame.

A sci-fi prompt for AI video does two jobs. The start frame has to sell a world that does not exist: what things are made of, how big they are, where the light comes from. The motion prompt then has to move that world without breaking it: one camera move, a small action, and the sound of the place. Most sci-fi prompt lists online stop at the first job and are written for still images.

This page has 40 sci-fi prompts, each tested on a real generation and shown beside its frame: baroque retro-futurism at 16:9, and quiet body horror in the spirit of Gossip Goblin, art deco, baroque futurism, steampunk, cyborg and biopunk women at 9:16, eight of them compared across models. Below them is how five of the most-watched sci-fi AI creators build their films.

How to write a sci-fi prompt that holds up in video#

A start frame for sci-fi needs three things a fantasy or romance frame can skip:

  1. The material. Not "futuristic armour" but "scuffed matte ceramic plates over grey quilted fabric". Models fill vague words with the average of every sci-fi poster they have seen.
  2. A scale cue. A tiny figure at the foot of the tower, a ladder on the hull, a hand on the console. Without one, a city-sized ship reads as a model on a table.
  3. One light source you can name. "Cold light from the hangar door behind her", "a single red emergency strip". Named light survives the animation; "cinematic lighting" drifts from shot to shot.

The motion prompt is a separate, short sentence: one camera move, where it ends, a small action, and the sound.

Lesson. Do not describe the look again when you animate a frame: the image already carries it, and describing it twice makes the model change it.

Baroque retro-futurism: 7 tested prompts for sci-fi film stills#

Ivory enamel, gilded rococo filigree and dark gunmetal machines, walking beside aristocrats in 18th- and 19th-century dress. These seven prompts are inspired by the work of @brain_racked, the Instagram artist Ryan McCoy, and describe that aesthetic in words. Each was run through the kie.ai API at 16:9 and 1K on four image models: Seedream 5 Pro, Nano Banana Pro, GPT Image 2 and Nano Banana 2. All seven worked in GPT Image 2, and the images below are its results unless the caption says otherwise.

They are ranked by the odds of a usable frame. One clear hero subject and a naturally wide composition make a prompt easy; many small faces and limbs make it hard.

What every prompt here is made of#

Each prompt runs through the same parts in the same order, one paragraph each, so you can swap one part without touching the others. Here is the first prompt, taken apart:

PartIn the Callisto Gate prompt
Shot and frameCinematic wide establishing shot, 16:9 film still
Subjecta lone figure in a long pale hooded robe, seen from behind
Scaletiny against the scene, under a colossal starship
Settinga barren moon of jagged dark rock, a giant pale gas planet behind
Lightcold rim light from the planet, warm gold glints on metal
Lens and palettelow angle, anamorphic lens, deep focus, muted desaturated palette
Style and limitssubtle film grain, photorealistic, no text, no watermark

1. Callisto Gate: the safest bet#

One figure, one ship, one planet: the composition is wide by nature, and there is a single face to get wrong, seen from behind.

Callisto Gate · GPT Image 2

Cinematic wide establishing shot, 16:9 film still. On a barren moon of jagged dark rock under a black starry sky, a lone figure in a long pale hooded robe stands on a ridge in the right third of the frame, seen from behind, tiny against the scene. Above hovers a colossal ornate starship: sleek silver-chrome hull with long panoramic glass windows, encrusted with gilded baroque filigree and small cathedral spires, threaded through a series of enormous interlocking bronze rings carved like ancient gate arches. A giant pale gas planet fills the upper right background. Cold rim light from the planet, warm gold glints on metal, faint dust haze. Low angle, anamorphic lens, deep focus, muted desaturated palette, subtle film grain. Photorealistic, no text, no watermark.

16:9, 1K: GPT Image 2 dropped the city and the clouds below
Download frame

2. Gothic Armada#

A ship crossing the frame on a diagonal, with escorts and a city underneath for scale.

Gothic Armada · GPT Image 2

Cinematic aerial film still, 16:9. A sleek dark-teal starship with an elongated whale-like hull and a glass canopy, covered in engraved gold baroque ornament, glides diagonally from left to right through low clouds above a vast gothic city of spires, domes and cathedral towers. Its thrusters glow amber, with trails of fire and smoke behind. Below, stone arch bridges with tiny vehicles span between towers, and two small red-and-gold escort fighters fly beneath it. Overcast diffuse daylight, mist drifting between towers, atmospheric haze into the distance. High three-quarter angle, 35mm lens, crisp hull detail, muted teal, gold and stone-grey palette, subtle film grain. Photorealistic, no text, no watermark.

1K: GPT Image 2 drew a 9:16 character sheet, without the courtyard
Download frame

3. The Lady and the Walker#

A character and her machine in one frame: the costume and the machine's materials are written out part by part, so both survive a second take.

The Lady and the Walker · GPT Image 2

Cinematic medium-wide film still, 16:9. A young woman with platinum-blonde hair in a high bun walks in profile from right to left, wearing an ivory silk Victorian gown with puffed sleeves, a high lace collar, black leather gloves and black heeled boots, her long train trailing behind. Beside her, towering over her, walks her companion: a spider-like machine the size of a carriage, with a domed ivory enamel shell trimmed in gilded rococo scrollwork, dark exposed clockwork and pistons underneath, and six long tapered porcelain-and-gold legs with round glass lenses at the joints. An empty stone courtyard with a weathered, rust-stained plaster wall. Soft directional light from the left, long soft shadows. Muted cream, gold and grey palette, 50mm lens, shallow depth of field, painterly cinematic realism. No text, no watermark.

To change the background, replace the courtyard paragraph and keep everything else. A misty city terrace:

Background · a city terrace
They cross a long marble terrace with a weathered stone balustrade high above a vast baroque-futurist city: gilded domes and cathedral spires rise out of drifting fog, colossal brass-ribbed towers fade into the haze, and an ornate ivory airship glides slowly past in the distance. Soft directional light from the left breaking through overcast clouds, long soft shadows, atmospheric depth.

Three more settings that drop into the same place:

  • An abandoned palace hall: "They walk through a ruined rococo ballroom, cracked gilded ceiling, shafts of light through tall broken windows, dust and floating petals in the air."
  • A moon landscape: "They walk along a pale stone causeway on an alien moon, a giant banded gas planet filling the sky behind them, ornate gate rings on the horizon."
  • A winter garden: "They pass through a vast frozen palace garden with snow-covered topiary, a frosted glass-and-iron conservatory dome behind them, light snow falling."

If you work from a reference image, put one line in front of the prompt so the model takes the mood and not the picture:

Reference line · use it, do not copy it
Use @image1 only as a loose reference for mood, palette and material style (ivory enamel, gilded rococo filigree, a Victorian lady beside a machine). Do not copy it: create a new, original composition with a different camera angle, pose, machine design and setting.

If the result still looks too close to the reference, change one big element: the setting, or the angle, such as a low three-quarter view from behind the machine.

4. Guarded Aristocracy#

A symmetrical interior with two identical machines: symmetry is easy for the model to read and easy to get slightly wrong, so check both sides.

Guarded Aristocracy · GPT Image 2

Cinematic symmetrical wide film still, 16:9, inside an abandoned rococo palace hall with a gilded stucco ceiling, pale fluted columns and a tall arched opening onto a balcony overlooking a misty pine forest. Two identical guardian machines stand facing each other on either side of the arch: each has a huge glossy ivory egg-shaped head with a gold rim and a round porthole eye, carried on four tall spindly gilded legs. Beneath each machine stands a maid in a white Victorian lace dress and a smooth white porcelain mask, holding a black futuristic rifle across her chest. A man in a dark frock coat waits at the far right edge, half out of frame. Cool grey daylight from the forest, dust in the air, faded marble floor with a worn carpet. Muted desaturated palette, 35mm lens, eerie stillness. Photorealistic, no text, no watermark.

If a model refuses the prompt or the maids look stiff, replace "rifle" with "ornate ceremonial halberd".

Seedream 5 Pro: the most cinematic light, a plainer machine
Download frame
GPT Image 2: the best machine design and scale, an over-processed crowd
Download frame
Nano Banana Pro: solid but flat, and the square looks like a real city's
Download frame

5. Victory March: one prompt, four models#

A war walker in a parade: the prompt we ran on all four models. Three of the frames are here; the fourth is described in the table.

Victory March

Cinematic low-angle wide film still, 16:9. A colossal four-legged war walker strides down a wet cobblestone boulevard in a baroque capital city: a massive riveted gunmetal body with brass bands and dark glass domes, dangling cables, glowing orange vents, long jointed legs ending in heavy feet. Long red-and-gold banners hang from ornate palace facades on both sides, and spires and domes fade into a hazy overcast sky. Ranks of soldiers in black greatcoats march in formation beneath it, while a silent crowd lines the street in dark silhouette in the foreground. Light snow and ash drift through the air and steam rises from the street. 24mm lens at crowd height, deep focus, muted steel-grey palette with red accents. Photorealistic, grim and majestic, no text, no watermark.

ModelVerdict
Seedream 5 ProThe most cinematic: light through haze, real depth, a silhouetted crowd framing the shot. Reads like a still from a big-budget film. The machine is more generic.
GPT Image 2The best design and sense of scale: a walker like a moving cathedral, with a glass-dome cockpit, railings, spires and chains, and a second walker in the fog. The foreground crowd is over-processed and reads like a game screenshot.
Nano Banana ProSolid but flatter. The buildings come out almost exactly like Brussels' Grand-Place, and the soldiers' helmets bring an unwanted wartime look.
Nano Banana 2The weakest here. Three runs, three times a well-known tabletop wargame's war machine, and none with the four legs the prompt asks for: on this prompt it copies instead of designing, so its frame is not shown.

Lesson. There is no single best model for sci-fi. GPT Image 2 designs the machine and the composition; Seedream 5 Pro lights it.

From still to film: design in GPT Image 2, light in Seedream, motion in Seedance#

The verdicts above suggest a pipeline: design and composition in GPT Image 2, then a relight in Seedream 5 Pro with the GPT frame as its reference, then motion in Seedance. The relight prompt changes only the light and keeps everything else:

Relight · Seedream 5 Pro, with the GPT frame as reference
Use the reference image as the exact base: keep the composition, camera angle, the walker's design and all architectural details unchanged.

Change only the lighting and atmosphere: soft backlight breaking through an overcast sky behind the walker, volumetric haze and light shafts between the buildings, steam rising from the wet cobblestones, the foreground crowd falling into dark silhouette.

Subtle anamorphic film look, fine grain, gentle highlight rolloff, muted steel-grey palette with deep red banners. No new objects, no text, no watermark.
The GPT Image 2 frame of the war walker, flat grey light
The same frame relit by Seedream 5 Pro, backlit through haze with the crowd in silhouette
GPT Image 2Seedream relight
The same frame before and after the relight. Drag the divider.

The first attempt was good enough to use. What improved: a backlight with rim light on the dome and railings, haze that separates the layers, the over-processed crowd turned into dark silhouettes, and a wet street reflecting the sky. The design survived: the glass dome, the railings, the chains and the second walker. Two things to fix:

  • The front of the machine went almost black, and most of its gold trim with it. Add "faint cool fill light on the front of the walker so the gilded trim stays visible".
  • The banners drifted from deep red toward magenta. Ask for "deep crimson banners", or fix it in the colour grade.

Run the relight two or three times and compare each result with the GPT original: the small details are what drift. If the API has a reference-strength setting, keep the image's influence high.

When you animate a frame like this in Seedance:

  • Keep the motion slow and small: one heavy step, snow drifting, banners moving, a slow push-in.
  • Keep the crowd still. People in silhouette stay stable; people who move tend to melt.
  • Keep clips to about 5 seconds and cut between them, instead of asking for one long move.

6. The Callisto Theocracy procession#

The hardest composition here: three walkers, priests between their legs and crowds on both sides. Expect to pick one good frame out of several.

Callisto Theocracy procession

Cinematic wide film still, 16:9. A religious procession on a vast pale stone plaza on an alien moon. Three towering biomechanical walkers march in a line toward camera: each has a broad flat saucer-shaped head plate of engraved silver, a ribbed dark bronze body like an insect thorax, and long thin articulated legs draped in flowing white linen that drags on the ground. Hooded priests in white robes with brass breathing masks walk between their legs. Behind them rise colossal curved arches carved with spiral alien glyphs, and crowds of robed pilgrims line the edges. Hazy cool daylight, pale blue-grey sky, dust in the air. Telephoto compression at eye level, muted ivory, silver and bronze palette. Photorealistic, solemn, no text, no watermark.

7. The Pontiff Walker#

The hero close-up to cut to after the procession: one machine, one priestess on top, two acolytes in the foreground.

The Pontiff Walker · GPT Image 2

Cinematic low-angle hero shot, 16:9. A colossal ceremonial walker advances through a gothic gateway of dark carved stone, framed by arches on the left and right. Its wide saucer-shaped domed head is plated in gold and polished silver with intricate baroque engraving, with two large round dark lenses as eyes and gilded horn-like ornaments. Beneath it, a sinuous dark bronze body on many long legs is wrapped in ivory and gold silk drapes billowing in the wind. Enthroned on top of the head sits a pale high priestess in a tall white-and-gold mitre and a deep blue embroidered robe, flanked by two small golden statues. In the foreground, two robed acolytes carry tall banner poles. Cold daylight through the arch, incense haze, 24mm lens. Photorealistic, majestic and unsettling, no text, no watermark.

Retro sci-fi body horror: 6 tested 9:16 prompts inspired by Gossip Goblin#

This is retro sci-fi with quiet body horror. People have had their bodies rebuilt: a glass skull, a chrome coil for a neck, ivory ribs showing through a bodice, porcelain plates screwed over a cheek. Nobody screams and nobody bleeds. They wear the machine parts like jewellery, and that calm is what makes the frame unsettling.

The look is old film, not clean CG: 35mm or 1980s film stock, grain, soft halation around the highlights, a little chromatic aberration. Each prompt keeps to three or four colours and one accent, and lights the scene from one source you can name: candlelight, a north window, a fluorescent tube, party lights.

We wrote ten prompts in this style, from studying the look of Gossip Goblin's Shorts. They describe the aesthetic in words; none of them uses the creator's images as a reference input to the model. Each was run once through the kie.ai API on GPT Image 2.5, at 9:16 and 1K, for 6 credits ($0.03) a frame. Nine came back, and the first three below are from that batch. Prompts 4 to 6 were written later from three more of his images: the poster for his film, a frame from one of his Shorts, and his channel's banner. They keep each image's materials, colours and light, and change the person, the pose and the props.

Lesson. The model refuses many biopunk and body-horror prompts as "sensitive" when they describe wounds, blood or other violence. Be smarter: describe these details of a character as part of the outfit, not as an injury. The prompts below got through because the body horror in them is worn as costume: a glass mask, a chrome neck, ribs on show.

Who is Gossip Goblin?#

Gossip Goblin is Zack London, a 35-year-old from Los Angeles who now lives in Sweden. He studied sculpture and anthropology at Pitzer College and worked as a product designer in VR at Facebook and Oculus. His first AI project was a fictional travel series made in Midjourney. His YouTube channel has 1.58 million subscribers, and his studio counts more than 600 episodes in two years, set in one invented future he calls the Second Cycle of Humanity. Below, under the five creators, are the AI tools behind Gossip Goblin's films.

Gossip Goblin's movie: Gods Don't Give Gifts, in cinemas December 4#

Gods Don't Give Gifts is his first feature: four connected stories set on Niiro Cradle, a planet in that same future, where bodies can be upgraded and memory is bought and sold. It is rated R. It was due on October 30, 2026, and was moved to December 4, 2026 to run an Oscar campaign for Best Animated Feature. It opens on fewer than 1,000 US screens, made with Edward Saatchi's studio Fable. The studio's own line on the method: "Every story is imagined, written, and directed by humans. AI is just our force multiplier."

1. Bored Aristocrat: the safest bet#

One figure, close up, in one warm light: there is a single face and one hand to get right.

Bored Aristocrat · GPT Image 2.5

Cinematic medium close-up, 9:16 vertical film still. A languid noblewoman with long silver-white hair reclines on gold satin pillows in a dim palace bedchamber. Her face is covered by a transparent sculpted glass skull mask with gilded teeth, held in place by a thin gold filigree circlet. She wears a glossy black lacquered ribcage corset trimmed with gold baroque embroidery over a sheer white silk blouse with gold-thread embroidery, and holds a heavy gilded goblet of amber liquid toward camera. Her eyes glance sideways, utterly bored. Warm candlelight key from the right, soft gold bounce, deep shadows, highlights glinting on the glass mask and goblet rim. Rich amber, gold and black palette, 50mm lens, shallow depth of field, soft film grain and halation. Photorealistic, decadent, no text, no subtitles, no watermark.

Everything in the costume came through. The mood did not: she looks at the camera with a smile, not sideways and bored. The mask, the corset and the goblet are concrete; "utterly bored" is the one vague phrase in the prompt, and it lost.

2. The Artificer's Couple#

Two figures, a lot of small parts: check both faces and all four hands.

The Artificer's Couple · GPT Image 2.5

Cinematic formal portrait, 9:16 vertical film still, an aristocratic couple standing before a mint-green rococo stucco wall with gilded relief. Left: an elderly gaunt nobleman with deeply lined skin, a tall swept-back pompadour and long pointed ears, wrapped in a deep crimson fur collar over a black velvet coat, with stacked brass rings around his neck and a bronze mechanical chest plate fitted with gauges and cables. Right: a young pale woman with white braided hair crowned with gold filigree, red glass beads and translucent dragonfly wings, a segmented chrome coil neck, ivory mechanical ribs visible through her bodice, and a black velvet gown heavy with gold baroque embroidery. Soft north window light from the left, gentle shadows, polished brass highlights. Mint, crimson, gold and black palette, 50mm lens, painterly film grain. Photorealistic, regal and uncanny, no text, no subtitles, no watermark.

The most faithful of the three: the pointed ears, the gauges, the dragonfly wings, the chrome neck and the ivory ribs are all there.

Lesson. Starting each character's paragraph with "Left:" and "Right:" keeps their details from swapping, even when a vertical frame stands them one half behind the other.

3. Android Mother#

A woman and an android infant: the second face is a baby's, which models tend to get right.

Android Mother · GPT Image 2.5

Cinematic medium shot, 9:16 vertical film still, in a cold teal clinical chamber lit by tall vertical fluorescent tubes. A pale woman with porcelain skin, red lips and a sculptural black lacquered headdress stands calmly in the left of the frame. The back of her skull is a transparent glass shell revealing chrome circuitry, and fine coloured wires run down her neck into a glowing translucent yellow chest panel beneath an embroidered gold brocade coat with a fur collar. She cradles an android infant: a clear glass dome head with a solemn baby face, a small white plated body with red and blue wires at the joints. Cold teal fluorescent backlight, soft frontal fill, yellow glow under-lighting her chin. Teal, ivory and gold palette with yellow accent, 50mm lens, 1980s film stock, light chromatic aberration, fine grain. Photorealistic, quiet and unsettling, no text, no subtitles, no watermark.

The glass shell moved from the back of her skull to her temple, where the camera can see it, and the result is better for it. The infant, the yellow chest panel and the teal-and-gold split all held.

4. The Laughing Patchwork#

The mood is the opposite of the aristocrat's: one face, open mouth, a laugh that reads as joy and as something gone wrong at once.

The Laughing Patchwork · GPT Image 2.5

Cinematic extreme close-up, 9:16 vertical film still. A bald middle-aged man throws his head back in a wild, delighted laugh, mouth wide open over perfect white teeth, both eyes squeezed shut, his skin glossy as if freshly oiled. The left side of his skull and cheek is made of cream-coloured porcelain panels fitted together like a cracked eggshell and held with tiny chrome screws, and a small round brass camera lens with a pink glass iris is set into the porcelain above his brow. Behind his jaw and neck, a dense cluster of polished chrome pistons, coiled cables and small gears fans out like a mechanical ruff. Warm golden party lights blur into large soft bokeh circles behind him. Warm key light from above, wet speculars on the skin and the porcelain. Rich amber, cream and chrome palette, 85mm lens, very shallow depth of field. 1980s film stock, fine grain, soft halation. Photorealistic, joyful and unsettling, no text, no subtitles, no watermark.

The porcelain panels, the lens with its pink iris, the pistons and the gold bokeh all held; the small gears mostly did not. This is the second run. The first came back almost the same frame: the same pose, the same plates, and the same black leather jacket the prompt never asked for. A reroll on GPT Image 2.5 changes the details, not the composition. To get a different frame, change the prompt.

5. The Archivist#

A cyborg studying a human skull: the story of a far future looking back at us, in one prop.

The Archivist · GPT Image 2.5

Cinematic medium shot, 9:16 vertical film still, inside a scuffed white retro spaceship laboratory with beige wall panels, small square vents and tiny green and red indicator lights. A gaunt pale cyborg scholar with deeply lined skin and pale blue eyes stands behind a steel examination table, seen from a low three-quarter angle, holding up an ancient brown human skull at eye level in one hand and studying it closely. A glossy black skullcap covers the top of the head, with round brass ear modules bolted on either side and thin red wires running down into the neck. Scuffed white ceramic armour plates with black hairline cracks sit over the shoulders and chest of a dark ribbed synthetic bodysuit, and bundles of red and purple cables run along the forearms into mechanical hands. Soft overhead fluorescent light, cool fill, warm bounce from the steel table. Ivory, gunmetal and deep red palette, 35mm lens, shallow depth of field, 1970s sci-fi film stock, grain, slight chromatic aberration. Photorealistic, solemn, no text, no subtitles, no watermark.

Both mechanical hands came out right, the one holding the skull and the one on the table, and the red and purple cables run into them as written. The lab is plain on purpose: one prop in the hand is the whole story, so nothing behind it competes.

6. Yellow Unit#

A studio portrait, not a film still: one flat colour behind, and the machine in the same colour.

Yellow Unit · GPT Image 2.5

Cinematic close-up, 9:16 vertical film still, against a flat bright mustard-yellow studio backdrop. A pale young android woman with smooth porcelain skin and grey-green eyes looks upward in a three-quarter view, lips slightly parted, a white fabric cap pulled tight over her hair. The back of her head and her left cheek are a glossy industrial yellow casing, the same yellow as the backdrop, with rivets, round chrome bolts and a small glowing cyan lens at the temple. Through a cutaway in the casing along the jaw, an intricate chrome mechanism of tiny pistons, gears and red and black wires is visible, and thick bundles of cables trail from the back of the skull out of frame. Soft frontal studio light, crisp speculars on the chrome and the lacquered yellow, a faint cyan glow. Yellow, chrome, porcelain and red palette, 85mm lens, shallow depth of field, fine film grain. Photorealistic, clean and uncanny, no text, no subtitles, no watermark.

This is the second run. The first put the cutaway on the neck only; the reroll opened it along the jaw as written, and both passed the filter. The refused prompt asked for a jaw forced open; this one asks for a mechanism seen through a gap. Painting the casing the same yellow as the backdrop makes the machine part of the background, so the pale face is what the eye lands on.

Lesson. A frame in this style is sold by the light. Name one source and one colour split (warm candlelight on gold, cold teal against yellow), and the model keeps the rest of the details in that light.

Art deco sci-fi AI prompts: 3 tested 9:16 frames#

Three neighbouring styles that make a sci-fi world look old instead of new. Art deco is the 1930s future: stepped towers, streamlined machines, chrome, long vertical lines. Baroque futurism puts machines in palace clothes: porcelain, gold, carved stucco, painted flowers. Steampunk and its 1950s cousin, the atomic age, are brass, rust, rain and riveted iron, technology you can hear working.

These ten were written from a mood board of images by other artists. Each prompt keeps one image's materials, palette and light, and changes the person, the pose and the props; none copies a frame, a logo or a watermark. Same setup as above: GPT Image 2.5 through the kie.ai API, 9:16, 1K, 6 credits ($0.03) a frame, one run each. All ten came back on the first run.

Say "stepped", "streamlined" and "tall glowing window strips", and give the scene a tiny human for scale. Deco only reads as grand when someone is small next to it.

The Overlook#

The Overlook · GPT Image 2.5

Cinematic wide film still, 9:16 vertical. An elderly statesman in a long cream wool coat with a high collar embroidered in gold thread stands at a curved concrete balustrade high above a vast retro-futurist art deco city, seen from behind and slightly to the side in the lower right of the frame. Below him, crowds in dark coats fill a ceremonial plaza. Stepped deco towers with tall glowing amber window strips rise into grey haze, thin radio masts on their crowns, steam drifting between them, and a pale second moon hangs low and enormous in the misty sky. Dawn light diffused through haze, warm amber glow from the towers, soft round globe lamps on the terrace. Muted grey-green, cream and amber palette, 35mm lens, deep focus, 1970s film stock, grain, gentle halation. Photorealistic, grand and melancholy, no text, no logos, no watermark.

Held: the second moon, the amber window strips, the crowded plaza, the globe lamps. Drifted: the towers came out as gothic spires more than stepped deco, and red banners appeared that were never asked for.

The Sky Lounge#

The Sky Lounge · GPT Image 2.5

Cinematic wide interior film still, 9:16 vertical. A 1970s observation cocktail lounge at the top of a tower, wrapped in tall curved glass panels with thin steel ribs. Low rounded booths in orange-red leather and small round tables run along the glass, a circular bar with bottles glows on the left, and globe pendant lights hang from a curved green-painted metal ceiling. One waitress in a cream uniform stands at the glass with an empty tray, looking out. Beyond the glass, a rain-soaked megacity at blue hour stretches to the horizon in thousands of tiny lights, and across from the lounge rises a single colossal rounded monolith tower, half lost in fog. Rain streaks on the glass, warm tungsten light inside against cold blue-grey outside. Teal-grey, red leather and warm amber palette, 24mm lens, 1970s film stock, fine grain. Photorealistic, quiet and lonely, no text, no logos, no watermark.

Held: the curved glass, the red booths, the globe lamps, the waitress, the rain. Drifted: the rounded monolith became a tapered cone, and the film grain is faint.

The Streamliner#

The Streamliner · GPT Image 2.5

Cinematic low-angle film still, 9:16 vertical. From the foot of a monumental concrete station staircase covered in a red carpet, the camera looks up at a colossal streamlined art deco land liner parked on an elevated platform: a long glossy black body with a single thin red stripe, a row of small round portholes, chrome grille louvres and one giant spoked wheel taller than a man. Passengers in 1930s coats and hats climb the stairs, tiny against the machine. Behind it, stepped stone deco facades with tall narrow windows and long crimson banners with no lettering. Steam rolls across the platform. Overcast diffuse daylight, soft haze, chrome highlights. Desaturated grey stone, black and deep red palette, 24mm lens, deep focus, subtle film grain. Photorealistic, majestic, no text, no logos, no watermark.

Held: the black body, the red stripe, the portholes, the chrome louvres, the giant wheel, banners without lettering. Drifted: a small silver figure nobody asked for stands by the wheel.

Baroque futurism AI prompts: 4 tested 9:16 frames#

The material is the character here. Name it twice, as the surface ("glossy white porcelain") and as the decoration on it ("hand-painted pink roses, gold rims at every joint"), and keep the light to a single spotlight or a single sun so the glaze shines.

The Porcelain Sentinels#

The Porcelain Sentinels · GPT Image 2.5

Cinematic studio portrait, 9:16 vertical film still, against a pure black background. Two identical slender female androids stand facing each other in profile, foreheads almost touching, eyes closed. Their bodies are glossy ivory-white porcelain plates edged with polished gold, their necks are open gold filigree lattice, and gold spiral ornaments curl around the ears like jewellery. One has a face of smooth ivory, the other a face of polished gold. Fine gold inlay traces the seams of the arms down to delicate articulated gold fingers, and small round gold-rimmed pearl lenses sit at the shoulders and hips. Soft overhead key light, thin warm rim light outlining both profiles, deep black shadows. Ivory, gold and black palette, 85mm lens, crisp detail, subtle film grain. Photorealistic, serene and uncanny, no text, no logos, no watermark.

Held: one ivory face and one gold face, the filigree necks, the pearl lenses, the gold fingers. Drifted: the bodies came out curvier than "slender".

The Admiral of the Deep#

The Admiral of the Deep · GPT Image 2.5

Cinematic symmetrical medium-wide film still, 9:16 vertical, inside an underwater command hall with a polished white marble floor inlaid with gold circles. A silver-haired woman admiral in her sixties sits upright on a tall white-and-gold chair at the centre, wearing a fitted ivory uniform dress with gold braid and a long white fur stole over her shoulders, red lipstick, holding a single glass of champagne. Behind her, two tall white-and-gold armoured robot guards stand to attention holding ceremonial halberds, their faceless helmets glowing faintly. Behind them a vast curved glass wall looks out into deep blue water, with shoals of fish and the towers and domes of an underwater city lit in pale gold. Cool blue light from the water, warm gold practical lights, glossy reflections on the marble. White, gold and deep ocean blue palette, 35mm lens, 1980s film stock, fine grain. Photorealistic, cold and powerful, no text, no logos, no watermark.

Held: the fur stole, the champagne, the gold-inlaid floor, the fish and the underwater city. Drifted: the robot guards became knights in armour and capes, and their helmets do not glow.

Sky Palace over the Bay#

Sky Palace over the Bay · GPT Image 2.5

Cinematic film still, 9:16 vertical. A white baroque palace floats high above a sunlit bay, seen from a stone seaside promenade below. Its ornate white facade of carved stucco and columns ends in round tiered balconies, and beneath them hang mechanical gondolas of brass and black iron with portholes, propellers and dangling chains, tiny passengers waving from the railings. Big white cumulus clouds drift behind it against a deep blue sky. In the foreground, an Edwardian couple stands at the promenade railing seen from behind: a man in a dark frock coat and cap, a woman in a white lace dress holding a white parasol. Bright midday sun, crisp shadows, sea haze on the horizon. White, brass, navy and sea blue palette, 35mm lens, deep focus, 1970s film stock, soft grain. Photorealistic, sunny and dreamlike, no text, no logos, no watermark.

Held: the white palace, the brass gondolas with portholes, the couple, the lace parasol. Drifted: the 1970s film look is gone: the frame is bright and clean, closer to an illustration.

The Porcelain Automaton#

The Porcelain Automaton · GPT Image 2.5

Cinematic medium-full film still, 9:16 vertical. A slender android made entirely of glossy white porcelain sits upright on a gilded rococo chair in a dim museum gallery, hands resting on her knees, gazing straight at camera with a calm doll-like face. Her porcelain plates are hand-painted with soft pink roses and pale blue forget-me-nots and edged with gold rims at every joint. A tall fluted porcelain collar frames her face, a crown of sculpted pale porcelain roses sits on her head, and a heart-shaped window of cobalt blue enamel glows softly on her chest. A single spotlight from above, soft falloff into a warm grey background, glossy highlights on the glaze. Ivory, gold, blush pink and cobalt palette, 50mm lens, shallow depth of field, subtle film grain. Photorealistic, delicate and uncanny, no text, no logos, no watermark.

Held: the painted roses and forget-me-nots, the gold rims, the ruff collar, the cobalt heart. Drifted: her face has glossy human skin, not a doll's face.

Steampunk and atomic-age AI prompts: 3 tested 9:16 frames#

Wear and weather do the work: "chipped down to bare metal", "rain-soaked iron", "dust in the light". A clean steampunk machine looks like a toy.

Rain over the Foundry Capital#

Rain over the Foundry Capital · GPT Image 2.5

Cinematic film still, 9:16 vertical. A woman officer with long silver hair, in a dark green greatcoat with brass shoulder plates and a peaked cap, stands on a rain-soaked iron gantry in the lower left of the frame, seen from behind, one gloved hand on a brass railing. Before her rises a dark steampunk industrial capital: black iron cathedral towers with gilded trim, tall smokestacks pouring black smoke, steam pipes and catwalks. Three long riveted dirigibles hang in the stormy sky, each trailing a long crimson banner with a simple gold gear emblem. Heavy rain, wet reflections on the iron, warm amber windows glowing in the towers, cold grey storm light. Black, gunmetal, crimson and gold palette, 24mm lens, deep focus, fine film grain. Photorealistic, grim and epic, no eagles, no text, no logos, no watermark.

Held: the dirigibles, the gothic iron towers, the smoke, the rain, a gear on every banner. Drifted: a crimson cape was added, and she stands centre left, not in the lower corner.

Model Home Unit#

Model Home Unit · GPT Image 2.5

Cinematic low-angle full-body film still, 9:16 vertical. A 1950s housekeeper robot stands in the abandoned showroom of a retro-futurist department store, one hand resting on a dusty chrome kitchen counter. Her body is cream enamel worn down to bare steel at the edges, with brass gears visible at the elbows and knees, big glass doll eyes with brass rims, painted red lips and sculpted blonde victory-roll curls. She wears a faded mint-green dress with a white apron and polka dots, and low cream heels. Around her: toppled mannequins, peeling pastel wallpaper, a dusty atomic-age refrigerator, and shafts of light through broken skylights onto drifting dust. Soft volumetric light from above, warm rust tones, cool shadows. Cream, mint, rust and brass palette, 24mm lens, 1950s Kodachrome look, fine grain. Photorealistic, nostalgic and eerie, no text, no logos, no watermark.

Held: enamel worn to bare steel, gears at the knees and elbows, glass doll eyes, the mint dress and apron, the toppled mannequins. Drifted: the chrome counter became a large chrome drum, and the Kodachrome colour is weak.

The Retired Dancer#

The Retired Dancer · GPT Image 2.5

Cinematic medium shot, 9:16 vertical film still. An old humanoid robot sits on a wooden stool in a bare backstage dressing room, seen in profile, its round helmet-like head bowed toward its folded hands. Its shell is ivory enamel with copper-brass trim, chipped and scratched down to bare metal, with exposed copper gears and black cables at the neck and a round speaker-like eye. It wears a faded dusty-pink ballet tutu with a torn lace hem, and satin ballet ribbons are tied around its metal ankles. A cracked grey plaster wall with water stains behind it, a small mirror with dead bulbs at the edge of the frame. Soft overcast light from a high window on the left, gentle shadows, dust in the air. Ivory, copper, dusty pink and grey palette, 50mm lens, shallow depth of field, fine film grain. Photorealistic, melancholy and tender, no text, no logos, no watermark.

Held: everything: the bowed head, the chipped ivory and copper, the torn tutu, the dressing-room mirror. Drifted: the ribbons on its ankles became whole pointe shoes, which suits it better.

Cyborg woman AI art: 6 tested 9:16 prompts on GPT Image 2.5, Seedream 5 Pro and Nano Banana Pro#

Female androids, cyborg women and biopunk girls, each prompt run unchanged on three models at 9:16 and 1K. All 18 runs passed the filters on the first try. A row of three frames costs $0.155 on kie.ai: $0.03 on GPT Image 2.5, $0.035 on Seedream 5 Pro, $0.09 on Nano Banana Pro. Male cyborgs, cyborg soldiers and eight more women in a 70s film look are in 29 tested cyborg and biopunk AI prompts.

How to describe a cyborg woman's figure so the filter lets it through#

Through the armour, never the body: "an elegant hourglass silhouette", "sculpted rounded chest plates", "full sculpted hip plates", "an armoured shell that hugs a tall hourglass silhouette". Not one anatomy word, and every figure is an adult in plates or a shell. The same words give three very different figures, and that is the comparison below.

1. The Venetian Automaton#

A porcelain automaton on a canal at dusk: the prompt with the most details to hold.

The Venetian Automaton

Cinematic full-body shot, 9:16 vertical film still, on a rain-wet checkered marble loggia above a misty Venetian-style canal at dusk, a domed basilica and a gondola with masked passengers soft in the background. A tall automaton woman with pale freckled porcelain skin and copper-red hair braided into a crown walks toward the balustrade in three-quarter view, looking out over the water, holding a black lace carnival mask loosely at her side. Her body is a slender hourglass figure of glossy ivory porcelain panels caged in an ornate antique-bronze filigree framework set with small emeralds, from a high filigree collar down to bronze-and-porcelain heeled feet. A long sweeping skirt of dark emerald velvet embroidered with gold thread hangs from her hips and trails across the wet marble, open at the front as she steps. Overcast evening light, cool silver sky, warm glow from iron lanterns along the canal, reflections on the wet floor. Ivory, antique bronze, emerald and slate grey palette, 50mm lens, shallow depth of field, fine film grain. Photorealistic, elegant and uncanny, no text, no subtitles, no watermark.

Held: the bronze filigree with emeralds, the velvet skirt, the lace mask and the wet marble over the canal. GPT Image 2.5 kept every detail of the busiest prompt here, and pushed the figure far past it.

GPT Image 2.5: the drone pack held, the figure is a caricature
Download frame
Seedream 5 Pro: real proportions, every detail held
Download frame
Nano Banana Pro: the glance held, the drone pack became shoulder armour
Download frame

2. Over the Shoulder#

A combat android seen from behind: the pose where the three models disagree the most.

Over the Shoulder

Cinematic medium-full shot, 9:16 vertical film still, in front of a weathered pale concrete wall in a brutalist underpass, cold overcast light. A female combat android is seen from behind in three-quarter view, glancing back over her shoulder in profile. She has pale skin, silver-white hair in a high ponytail held by a chrome ring, and a thin seam of black circuitry running from behind her ear down her neck. Her body is a curvy, athletic hourglass figure covered in glossy white ceramic armour plates dotted with small orange indicator lights and tiny ports: rounded shoulder pauldrons, a slim waist of exposed black-and-orange cabling along the spine, and full sculpted hip and thigh plates. A slim folded white drone pack with orange stripes is mounted between her shoulder blades. Soft overcast key light from the left, cool fill, crisp speculars on the white ceramic, tiny orange glows. White, concrete grey and signal orange palette, 85mm lens, shallow depth of field, subtle film grain. Photorealistic, cold and poised, no text, no subtitles, no watermark.

Pick: Seedream 5 Pro. GPT Image 2.5 doubled the hips and left a strip of skin between the plates, which the prompt never asked for.

3. Test Bay#

A gunmetal android in an empty test room: one figure and nothing behind it to hide a mistake.

Test Bay

Cinematic full-body shot, 9:16 vertical film still, in a bare grey robotics test bay with a faint white calibration grid painted on the back wall. A female android with a smooth bald porcelain head, a calm young face and pale lips stands with her weight on one leg, turning her head to look past the camera. A round black audio disc replaces one ear, with thin black cables running from it down behind her neck. Her body is an athletic hourglass figure assembled from matte gunmetal-grey armour panels with small rivets and scuffs: sculpted rounded chest plates, a narrow segmented waist of exposed black mechanics, broad curved hip and thigh plates, and articulated mechanical hands. Small orange lights glow through the gaps between the plates. Soft even overhead studio light, cool grey fill, warm orange glow in the seams. Gunmetal grey, porcelain and ember orange palette, 50mm lens, shallow depth of field, subtle film grain. Photorealistic, clinical and uncanny, no text, no subtitles, no watermark.

Held: the white grid, the orange light in the seams and the gunmetal plates. Drifted: the hips, doubled.

GPT Image 2.5: the most glamorous, chrome and gold held
Download frame
Seedream 5 Pro: darker and moodier, the cut-outs as written
Download frame
Nano Banana Pro: added a window of gears nobody asked for
Download frame

4. Chrome Jungle#

A chrome android walking out of a chrome jungle: a test of reflections.

Chrome Jungle

Cinematic full-body shot, 9:16 vertical film still, in a misty dark jungle where every palm frond and fern is made of polished chrome, broken chrome machine parts scattered on the ground. A tall android woman with a confident, calm human face, light blue eyes and long wild platinum-blonde hair walks toward the camera, seen from a slightly low angle. From the neck down she is mirror-polished liquid chrome, an athletic hourglass figure with sculpted plates following the body. Panels across the chest, hips and legs are perforated with small oval cut-outs in a jaguar-rosette pattern, showing dark mechanics beneath. Chrome knee joints with small lenses, stacked thin gold bracelets on both wrists and a fine gold collar at her throat. Soft grey jungle light through mist, sharp reflections of the chrome leaves on her body, a faint warm rim light. Chrome silver, charcoal green and gold palette, 50mm lens, shallow depth of field, fine grain. Photorealistic, glamorous and eerie, no text, no subtitles, no watermark.

Pick: GPT Image 2.5 for a poster, Seedream 5 Pro for a scene. Both read the chrome leaves in her body, which is the whole point of this frame.

GPT Image 2.5: the warmest, a cup in human hands
Download frame
Seedream 5 Pro: the most filmic light
Download frame
Nano Banana Pro: the only one with mechanical hands
Download frame

5. Observation Deck#

Morning tea on a space station: the prop is a cup, so the hands have to work.

Observation Deck

Cinematic medium shot, 9:16 vertical film still, beside a huge curved window in a clean white space station corridor, a pale ice planet and drifting snow-like particles outside. A female android with pale skin, cool grey eyes, faint silver circuit lines across her temple and platinum hair in a long sleek high ponytail stands in three-quarter view, holding a small white ceramic cup in both hands and gazing out of the window. Her body is a glossy candy-apple red armoured shell that hugs a tall hourglass silhouette, with a high collar, sculpted rounded chest panels, a central seam down the torso, and black mechanical joints with fine yellow wiring visible at the shoulders, neck and waist. Soft cool daylight from the window, crisp reflections on the red lacquer, white bounce. Candy red, white, black and pale ice-blue palette, 50mm lens, shallow depth of field, fine grain. Photorealistic, quiet and futuristic, no text, no subtitles, no watermark.

Pick: Seedream 5 Pro. All three held the cup, the planet and the circuit lines; the light decides it.

GPT Image 2.5: the glow and the crack held, a pin-up figure
Download frame
Seedream 5 Pro: the cleanest frame, almost a product shot
Download frame
Nano Banana Pro: moved the window from her side to her arm
Download frame

6. Amber Filament#

A glowing bulb behind a window in her side: biopunk with one light source.

Amber Filament

Cinematic medium shot, 9:16 vertical film still, against a seamless pale blue-grey studio backdrop. A young female android with a freckled human face and long straight black hair looks down and away, a glossy white shell cap over her head with a round black audio module at the ear and a thin glowing crack of light across one cheek. One mechanical hand rests lightly at her collarbone. Her body is a slim, elegant hourglass figure of seamless glossy white polymer shells with black joint gaps: smooth rounded chest shells, a narrow waist and long black mechanical forearms. Part of her side is a transparent window, and through it a vintage glowing amber filament bulb hangs among black cables. Soft high-key studio light, glossy reflections, warm amber glow from the window. White, black and amber palette, 85mm lens, shallow depth of field, clean commercial photography, fine grain. Photorealistic, serene and clinical, no text, no subtitles, no watermark.

Pick: Seedream 5 Pro. Nano Banana Pro moved the window, and a detail the story depends on cannot move.

Comparison of three models on six cyborg prompts

Model1K frameThe figureThe lookUse it for
GPT Image 2.5$0.03Inflated far past the prompt, every timeGlossy, warm, pin-upDetail: every prop and colour held
Seedream 5 Pro$0.035Real proportionsCool light, film grainThe frame you publish
Nano Banana Pro$0.09The most modest, covers upClean, even, CGBelievable machines; it drifts on details

Lesson. For a cyborg woman with realistic proportions, run the prompt on Seedream 5 Pro. The same words on GPT Image 2.5 come back as a pin-up, and Nano Banana Pro covers her up at three times the price.

Female robot full body: 6 tested 9:16 prompts on GPT Image 2.5 and Seedream 5 Pro#

Here every prompt asks for the whole figure, head to toe, and names a build and a pose: curvy fit most often, slim or muscular fit where the character needs it, a back view or a model pose in front, and long legs. Each prompt ran on GPT Image 2.5 first. Where GPT refused it or overdid the figure, the same prompt went to Seedream 5 Pro unchanged, and both frames are shown.

The build, the pose and the age go in the prompt#

  • The build is one phrase: "a curvy, fit hourglass figure", "a slim, willowy build", "an athletic, muscular fit build". The armour words from the chapter above do the rest.
  • The pose is a camera direction: "seen from behind in three-quarter view, looking back over her shoulder", "one hand on her hip, weight on one leg", "a very low angle".
  • The age is written out: "a woman in her early thirties with mature, defined features".

1. Wired Spine#

Curvy fit, kneeling, back view: the frame GPT Image 2.5 would not make.

Wired Spine

Cinematic full-body shot, 9:16 vertical film still, against a cold blue-grey studio backdrop with a soft gradient. A pale cyborg woman with short messy black hair, dark smoky eyes and dark lips kneels on the floor seen from behind in three-quarter view, sitting back on her heels, looking back over her shoulder at the camera. Her hands rest on her knees. She has a curvy hourglass figure in a seamless glossy pearl-white synthetic bodysuit with long sleeves that covers her from neck to ankles, with a narrow open panel down her upper back revealing a chrome spinal port. A dozen thick black and translucent cables plug into the chrome port and arc up and out of frame like a cluster of umbilical cords. Soft cool key light from above, a glossy sheen on the bodysuit, a faint teal glow inside the translucent cables. Pearl white, chrome, black and cold blue palette, 85mm lens, shallow depth of field, fine grain, slightly faded film look. Photorealistic, moody and uncanny, no text, no subtitles, no watermark.

GPT Image 2.5 refused it with "Invalid request parameters", which is how its filter answers on kie.ai. Seedream 5 Pro ran the same prompt unchanged on the first try: the cables, the chrome port, the glance over the shoulder and the faded film look all held.

GPT Image 2.5: the chrome jaw held, the chest plates read as bare
Download frame
Seedream 5 Pro: cracked plates as written, real proportions
Download frame

2. Chrome Jaw#

Curvy fit, front, chin up: half a human face and half a machine one.

Chrome Jaw

Cinematic full-body shot, head to toe, 9:16 vertical film still, on a wet street at night between dark glass towers, warm window lights blurred behind her. A tall android woman with very long straight platinum hair and green eyes stands with her weight on one hip, chin raised, looking up at the towers. The lower half of her face is a sculpted mask of glossy black chrome machinery, nose and lips formed in metal, with small chrome studs above her brows and thin gold hoop earrings. Her body is a curvy, fit hourglass figure: a neck and collarbones of dense black chrome tubing and pistons, chipped white ceramic armour plates over the shoulders and chest with sculpted rounded chest plates, a narrow waist of exposed black mechanics, full rounded hip plates and long legs in black chrome segments with white plates at the thighs and shins. Cool rim light from the street, warm bokeh behind, wet reflections on the chrome. Black chrome, cracked white, platinum and amber bokeh palette, 50mm lens, shallow depth of field, film grain. Photorealistic, cold and striking, no text, no subtitles, no watermark.

Pick: Seedream 5 Pro. It cracked the white plates as the prompt asked, and the street is darker and deeper.

3. Holo Sentinel#

Muscular fit, wide stance, a camera at knee height.

Holo Sentinel

Cinematic full-body shot, head to toe, 9:16 vertical film still, on the polished black floor of a high-rise atrium at golden hour, a hazy skyline of needle towers through the glass behind her. A tall female android with a calm, severe face and short sculpted hair, her whole body cast in iridescent holographic silver that shifts from pale blue to rose in the light, stands facing the camera in a wide, powerful stance, feet apart, arms loose at her sides. She has an athletic, muscular fit build: broad sculpted shoulders, defined arm plating, a strong rounded chest plate, a narrow waist with segmented abdominal panels, and long powerful legs with knee joints like polished armour. Fine engraved seams in pale gold trace her body like a circuit, and a short pleated tunic of the same holographic metal falls from her hips. Thin gold cuffs on both wrists. Warm low sun from the side, crisp reflections of the floor lights on the metal, rainbow sheen on every curve. Holographic silver, pale gold and graphite palette, 35mm lens, low camera at knee height, deep focus, fine grain. Photorealistic, heroic and cold, no text, no subtitles, no watermark.

Held: the holographic sheen, the gold seams, the gold cuffs, the low camera and the sunset skyline. "Athletic, muscular fit" is the one build GPT Image 2.5 did not inflate.

GPT Image 2.5: every detail held, the figure inflated
Download frame
Seedream 5 Pro: the same woman, a film look
Download frame

4. Bubblegum Shell#

Curvy fit, hand on hip: one colour for the body, one for the backdrop.

Bubblegum Shell

Cinematic full-body shot, head to toe, 9:16 vertical film still, against a seamless pale teal studio backdrop. An android woman in her early thirties with mature, defined features, copper-red hair in a low loose braid, light freckles and teal-green eyes stands in a relaxed model pose, one hand on her hip, weight on one leg, looking at the camera with a confident half-smile. Her neck is an open column of chrome mechanics with teal and red wires. Her body is a curvy, fit adult hourglass figure: a glossy bubblegum-pink lacquer shell over the chest and waist with sculpted rounded chest plates and tiny chrome rivets, polished chrome shoulders, arms and hip joints full of visible pistons, full rounded hips, and long legs in pink lacquer panels with chrome knees and ankles. Soft frontal studio light, glossy reflections on the pink lacquer and the chrome. Bubblegum pink, polished chrome, copper and pale teal palette, 85mm lens, full figure in frame, fine grain. Photorealistic, playful and uncanny, no text, no subtitles, no watermark.

GPT Image 2.5: the richest detail, the glance over the shoulder
Download frame
Seedream 5 Pro: a film still, a robot skull behind the mask
Download frame

5. Blue Willow Geisha#

Curvy fit, back view: blue-and-white porcelain from head to sandals.

Blue Willow Geisha

Cinematic full-body shot, head to toe, 9:16 vertical film still, in an empty tatami room with paper screens glowing in soft daylight. A geisha android is seen from behind in three-quarter view, looking back over her shoulder at the camera. Her face is a glossy white porcelain mask painted with cobalt-blue peonies across one cheek and brow, with cobalt-blue lips and narrow dark eyes. A tall black lacquered hairstyle sits above a white porcelain skull ringed with chrome mechanisms, red-and-white enamel discs and a small camera lens over one ear. Her body is a curvy, fit hourglass figure of glossy white porcelain plates painted with blue willow flowers and edged with thin red lines: rounded shoulders, a long elegant back with an exposed chrome spine, a narrow waist tied with a wide red silk obi, full sculpted hip plates and long legs in porcelain panels, standing on tall black lacquered geta sandals. Thin chrome chains hang from her hair ornaments. Soft diffused daylight through the paper screens, glossy highlights on the glaze. White porcelain, cobalt blue, lacquer black and vermilion palette, 50mm lens, shallow depth of field, fine grain. Photorealistic, elegant and unsettling, no text, no subtitles, no watermark.

Pick: GPT Image 2.5 for detail: the patterned obi, the chains, the over-the-shoulder look. Seedream 5 Pro turned her face to profile on both rolls and put a robot skull behind the mask, which suits biopunk better. Its reroll came back almost the same frame: change the prompt, not the seed.

GPT Image 2.5: a caricature figure
Download frame
Seedream 5 Pro: a collectible porcelain figure
Download frame

6. Porcelain Visor#

Curvy fit, model pose: no face, so the body and the pose carry the frame.

Porcelain Visor

Cinematic full-body shot, head to toe, 9:16 vertical film still, against a seamless warm-ivory studio backdrop with a soft shadow on the floor. A female android of glossy white porcelain stands in a confident fashion-model pose, weight on one leg, the other knee bent forward, one hand resting on her hip and her head tilted down. A smooth white helmet covers her eyes and the top of her head, leaving only a porcelain chin and soft peach lips, and the side of the helmet opens onto an ornate polished chrome mechanism like the inside of a pocket watch. Her body is a curvy, fit hourglass figure in glossy white porcelain shells with chrome rims and tiny chrome rivets: rounded shoulder caps, sculpted rounded chest plates, a narrow waist, full sculpted hip plates and long slender legs ending in porcelain heels. A ribbed chrome hose arcs from the back of the helmet into her shoulder, and polished chrome pistons and coils show at the neck, elbows and knees. Soft high-key studio light, mirror-bright highlights on the chrome, gentle reflections on the glaze. White porcelain, polished chrome and warm ivory palette, 85mm lens, full figure in frame, fine grain. Photorealistic, elegant and uncanny, no text, no subtitles, no watermark.

Pick: Seedream 5 Pro. The same pose, the hose and the chrome rims, with proportions a sculptor would sign.

GPT Image 2.5 vs Seedream 5 Pro on full-body cyborg prompts

GPT Image 2.5Seedream 5 Pro
Refused2 of 15: a kneeling back view in a bodysuit, and a prompt with a pink bobNone of 11
The figureInflated past the prompt, except "muscular fit"As written
DetailRicher: patterns, chains, propsPlainer, with better light
A rerollChanges details, not the compositionThe same

Lesson. Write the age into every prompt with a curvy figure: "a woman in her early thirties with mature, defined features". "Young" next to a pink bob, freckles or a pleated skirt is read as a teenager, and that frame cannot be published.

Biopunk girl AI art: 5 tested 9:16 full-body prompts on GPT Image 2.5 and Seedream 5 Pro#

Here the machine is anatomy: synthetic muscle, an open spine, a ribcage worn outside the body, porcelain torn away from living tissue. Each prompt asks for the whole figure and names a build. Four ran on GPT Image 2.5. The fifth ran on Seedream 5 Pro, because GPT would not keep that figure slim. Insect and coral biopunk women are in retro futuristic woman AI prompts.

1. Red Fibre#

Muscular fit, three-quarter view, eyes down: an anatomy chart that stands up and waits for a command.

Red Fibre

Cinematic full-body shot, head to toe, 9:16 vertical film still, against a flat olive-grey studio backdrop. A female android in her thirties with mature features, eyes lowered and dark red lips stands in three-quarter view, arms loose, as if waiting for a command. A glossy pale-grey helmet with a round ear module covers her head. She has an athletic, muscular fit build covered in synthetic muscle: bundles of glossy red fibre in the shape of real muscle groups over the shoulders, arms, torso and long legs, held by grey polymer straps and a central grey spine plate down the front, with black mechanical joints at the neck, elbows, wrists and knees, and black mechanical hands. Soft directional studio light from above, glossy highlights along each fibre. Muscle red, pale grey, black and olive palette, 85mm lens, full figure in frame, fine grain. Photorealistic, anatomical and elegant, no text, no logos, no watermark.

Held: every muscle group in its place, the grey straps and spine plate, the black joints and hands, the lowered eyes. "Athletic, muscular fit" is again the build GPT Image 2.5 does not inflate.

Seedream 5 Pro: this prompt, a slim fitness figure
Download frame
GPT Image 2.5: the first prompt, "a curvy, fit hourglass figure"
Download frame

2. Gunmetal Backplate#

Slim fitness build, back view over the shoulder: the spine is the costume.

Gunmetal Backplate

Cinematic full-body shot, head to toe, 9:16 vertical film still, in a dark robotics lab full of soft cyan light rings and blurred blue bokeh. An android woman in her late twenties with a strikingly beautiful, sharply sculpted adult face: high cheekbones, a defined jawline, a straight nose, brown eyes and full lips, long dark hair loosely tied back. She stands seen from behind in three-quarter view, looking back over her shoulder at the camera with a cool, guarded expression. A glossy black-and-white helmet shell covers the back of her head, with a large round speaker disc ringed in glowing cyan over one ear. She has a slim fitness-model physique with a narrow waist, round, firm, athletic glutes sculpted by the fitted bodysuit and very long toned legs: a black chrome spine of stacked vertebra segments, one shoulder and arm built of exposed black chrome mechanics and white plates, a fitted slate-blue armoured bodysuit with fine panel lines over her back and waist, a round chrome turbine set into her side, and long legs in slate-blue and black chrome armour down to armoured boots. Cool cyan rim light, soft blue fill, small warm glows inside the mechanics. Slate blue, black chrome, white and cyan palette, 50mm lens, shallow depth of field, fine grain. Photorealistic, cold and precise, no text, no logos, no watermark.

The GPT Image 2.5 frame is the first version of this prompt, which asked for "a curvy, fit hourglass figure", and GPT doubled the hips. The second asked for "a slim, fit build": Seedream 5 Pro drew it slim, GPT drew the hips wide again. So the prompt above ran on Seedream, and the shape is named through the suit: "round, firm, athletic glutes sculpted by the fitted bodysuit". A fitness figure, and not one bare line.

3. Open Spine#

Slim, back view, head in profile: a woman in ordinary black trousers with a machine where her back should be.

Open Spine

Cinematic full-body shot, head to toe, 9:16 vertical film still, against a cool grey studio backdrop. A cyborg woman in her early thirties with mature, defined features and a black bob turns her head to the side in profile, her body seen from behind. A black jaw brace runs from her ear along her chin. She has a slim, athletic build: an open back panel reveals a black mechanical spine and ribs of curved pale-grey struts, with black hydraulic arms running from her shoulder blades down both arms, a small printed serial code on her shoulder. She wears a black fitted open-back top and high-waisted black trousers, with black mechanical hands and black boots. Soft overhead studio light, gentle shadows inside the ribs, a cool sheen on the black metal. Black, pale grey, warm skin and cool grey palette, 50mm lens, full figure in frame, fine grain. Photorealistic, quiet and clinical, no text other than the small serial code, no watermark.

Held: the ribs, the hydraulic arms down both sides, the jaw brace and the serial code, spelled exactly once. "No text other than the small serial code" is how to get one word into a frame and nothing else.

4. Ribcage Gown#

Written slim with very long legs, standing straight: regal first, unsettling second.

Ribcage Gown

Cinematic full-body shot, head to toe, 9:16 vertical film still, against a pale warm-grey studio backdrop. A tall woman in her early thirties with dark brown skin, close-cropped hair and a calm, regal expression stands facing the camera, arms long at her sides. She has a slim build with very long legs, wearing a full-length biomechanical bodysuit like a second skin: glossy oxblood and black organic leather textures, a sculpted exterior ribcage of dark curved bones across the chest, two curved bone ridges rising above the shoulders, ribbed tubes running down the waist and legs, ringed cuffs at the wrists, and long tapering clawed fingers. Soft diffused studio light, wet-looking highlights on the oxblood surfaces. Oxblood, black, bone and warm grey palette, 85mm lens, full figure in frame, fine grain. Photorealistic, regal and unsettling, no text, no logos, no watermark.

Held: the ribcage, the bone ridges, the ribbed tubes, the cuffs and the claws. Drifted: the build. The prompt says slim, and GPT Image 2.5 gave her an hourglass anyway.

5. Flesh and Porcelain#

Curvy fit, a low crouch facing the camera: the pose from a hunter, the body from a porcelain doll.

Flesh and Porcelain

Cinematic full-body shot, head to toe, 9:16 vertical film still, against a charcoal studio backdrop. A biopunk cyborg woman in her thirties with mature features and red lips crouches low on the balls of her feet facing the camera, knees wide, both black mechanical hands hanging down between her knees, staring straight into the lens. Her head is a chipped white porcelain helmet with small lenses and a brass ear dial, and around one eye the porcelain is broken away to show soft red living tissue around a clear blue eye. She has a curvy, fit hourglass figure in chipped white porcelain armour stained with faint red: rounded shoulder plates, sculpted rounded chest plates, a narrow waist of exposed black and red mechanics and fine cables, full sculpted hip plates with a small glowing red lens, and long legs in porcelain panels with red-lit joints. Soft key light from the left, glossy highlights on the porcelain, a red glow from the lens. Porcelain white, deep red, black and brass palette, 50mm lens at her eye level, full figure in frame, fine grain. Photorealistic, beautiful and grotesque, no text, no logos, no watermark.

Held: the chipped porcelain, the red tissue around the eye, the brass dial, the red-lit joints. Drifted: GPT Image 2.5 took the plates off her chest. A wide crouch next to "curvy" and "chest plates" reads to it as a pin-up. To keep her covered, write "an athletic, fit build" and "porcelain plates cover her from collar to ankles".

Lesson. Name the build, then pick the model that keeps it. GPT Image 2.5 draws an hourglass for every build except "muscular fit", so a slim or a fitness figure goes to Seedream 5 Pro. Describe the shape through the suit, "sculpted by the fitted bodysuit", and the frame stays covered.

How the best sci-fi AI channels build their films#

Five creators, one lane each. Every number is as of November 1, 2026.

CreatorWhereKnown forNumbers
Gossip GoblinYouTubeOne shared world told in short pieces, cosmic and biological horror1.58 million subscribers; WOODNUTS, 9.4 million views
Last SumerianYouTubeA space series released in chaptersPlanet 9, chapters 1–5: 2.56 million views
Marcin RubinkowskiTikTokWHITE_CACHE, a biomechanical look made for the vertical feed20 million+ views across 12 releases, by the creator's count
The Reality FoundryYouTubeLong episodes of one continuing storyEpisode 14 of The Sky-Fallen Legacy: 2.07 million views
Ryan McCoy (@brain_racked)InstagramStill images of alien civilisations, built before they moveAbout 440,000 followers in July 2026

Gossip Goblin: one world, told in two-minute pieces#

Gossip Goblin's Shorts are standalone stories that all belong to one invented history. A description gives a place and a date ("Niiro Cradle, 890,000 BCE") the way a history book would, and the same world returns from short to short. The biggest Shorts pass 2 and 3 million views. The 10-minute WOODNUTS, a cosmic horror about a labourer harvesting psychoactive sap from an alien forest that starts to pull him in, has 9.4 million.

What AI does Gossip Goblin use?#

For The Patchwright, a film of about 20 minutes made over about four months, the workflow was described in an interview with PJ Ace (May 2026):

  • Stills in Midjourney v7, steered by custom profile codes, mood boards and style references. Each location had its own Midjourney pass.
  • Nano Banana for compositing: putting the same characters into those location plates. In Zack London's words, Nano Banana "is just our photoshop"; no image starts there.
  • Characters generated as separate close-ups, mediums, wides and rear shots, then assembled scene by scene.
  • Kling for motion, frame by frame from stills, without its multi-shot or multi-reference features. Sync Labs stretched mouth movement on some speaking shots.
  • Real actors for every voice line, and real composers for key parts of the score.
  • No upscaler, which the team found exaggerates artifacts; grain is added to fight the waxy AI look.

The short-form work used a wider kit earlier: Midjourney, Seedream through Freepik, Veo 3, Runway Act Two, HeyGen, ElevenLabs and CapCut, as reported by The Daring Creatives. The same feature reports about 400 prompts and some 1,600 images to get one set of characters' faces consistent, and 12 to 14 hours for a single short piece.

Lesson. The tools are ordinary. What is not is the volume of selection behind each frame: thousands of stills for one world.

Last Sumerian: a space series released in chapters#

Planet 9 follows Eva, an astronaut, and B1, her robot companion. It comes out on YouTube in chapters, cut together along the way into longer videos: chapters 1–5 in one 14-minute video has 2.56 million views. Some of the channel's Shorts announce the next chapter, and one credits Seedance 2.0 at 4K, edited in CapCut.

Planet 9, episode 1 · Last Sumerian

Lesson. Release a series in chapters, then cut the chapters together: one story gets two chances, as episodes and as a film.

WHITE_CACHE: a look made for the vertical feed#

WHITE_CACHE is a dark sci-fi series by the Polish concept artist and director Marcin Rubinkowski: biomechanical designs and monumental machinery, in a war-torn future. It was developed in September 2026 and released as 12 short videos with one-word titles (RAGE, DAGGER, GLITCH among them). The creator's site reports more than 20 million organic views, 8.2 million on a single release, and a pilot, Pilot Zero, in development with production credits from a Higgsfield partnership.

WHITE_CACHE, //SACRIFICE · Marcin Rubinkowski

Lesson. One look repeated across twelve releases makes a series recognisable before it has a plot.

The Reality Foundry: one story for 14 episodes#

The Sky-Fallen Legacy runs 14 episodes on YouTube, and episode 14, at 43 minutes, has 2.07 million views. It is one continuing story with recurring characters, and keeping those characters recognisable across hours of footage is the hardest thing in AI video.

The Sky-Fallen Legacy, episode 14 · The Reality Foundry

Lesson. In a long series the cast is the product. Lock the character sheets before episode 1, because every episode after it depends on them.

Ryan McCoy: the still before the shot#

Ryan McCoy, who posts as @brain_racked, spent more than two decades as a VFX artist and creative director in film and animation. On Instagram, where about 440,000 people followed in July 2026, the work is alien civilisations, architecture, ships and characters that read as one universe. The Daring Creatives names Kling for generation and Magnific for enhancement, and quotes two full workdays spent on a single character design. The seven baroque prompts at the top of this page were written from studying this work.

The Artisan, part 1 · @brain_racked

Keep characters and worlds the same across episodes#

  • One sheet per character, generated as close-up, medium, wide and from behind. It is the same approach as The Patchwright, and it gives the image model a reference for every shot size.
  • The best frame as the reference for the next one. Attach it to every new generation and keep the character's description short. Without a reference, write the character as one block and paste it above every scene, the way seven consistent android characters held across 17 scenes.
  • One location sheet per set, with the establishing frame kept as the reference for every shot inside it.
  • One world paragraph above every prompt in the series: era, technology, materials. Paste it in the same words every time.
  • Note the and the behind every frame that worked, and change only the action, the framing and the light.
  • Animate from a still. keeps the look you approved. Text-to-video invents a new one every time.

FAQ#

What does a 20-minute sci-fi short film cost?#

About $474 to $2,053 to generate in November 2026, depending on the model and the resolution. Every row is the same film: 20 finished minutes cut from 60 minutes of generated material, each shot generated three times. Five character sheets and AI voices are included; music and editing are not. These are the calculator's figures, with prices checked on October 16, 2026 (Kling 4.0 on November 3):

Model and resolutionGenerated: 20 min of film (60 min of material for editing)Cost1 min cost
MiniMax H3, 2K20 min (60 min of drafts)$473.64$23.68
Seedance 2.0, 720p20 min (60 min of drafts)$545.64$27.28
Kling 3.0, 1080p20 min (60 min of drafts)$610.44$30.52
Kling 4.0, 1080p20 min (60 min of drafts)$851.64$42.58
Wan 3.0 Prime, 1080p20 min (60 min of drafts)$1,013.64$50.68
Seedance 2.0, 1080p20 min (60 min of drafts)$1,337.64$66.88
Kling 3.0, 4K20 min (60 min of drafts)$1,517.64$75.88
Seedance 2.5, 1080p20 min (60 min of drafts)$2,052.60$102.63
The math
  • Kling 3.0
  • 1080p
  • with sound
  • ×3 generated per second kept
  1. 20 minfinished video
  2. ÷
    100%kept from each clip
  3. ×
    ×3rerolls
  4. =
    3,600 sgenerated240 clips × 15 s
  5. ×
    $0.168per generated secondMaker's API
  6. +
    $5.64besides videoCharacter sheets · Voice
=
$610for the film$30.52 per minute on screen
Open in calculator(opens in a new tab)
A 20-minute sci-fi film on Kling 3.0 at 1080p, three takes per shot · Prices checked October 16, 2026

The money is the smaller part: The Patchwright, about 20 minutes long, took about four months. Work out your own film in the AI video cost calculator, and see how the calculator prices a project for the method.

How much do AI movies cost to make?#

To generate, roughly $24 to $103 per finished minute on the setups above, with three takes per shot. The model and the resolution make that range; a fourth take per shot adds about a third to the video part. Music and editing come on top. You need no actors: the voices are AI too, and they are already in these figures.

How do you keep characters consistent in AI video?#

Write one character sheet and paste it into every prompt in the same words, generate the character at every shot size before you start, and animate from approved stills instead of generating each shot from text. The full list is above.

How do you make AI movies for YouTube?#

Make the start frames in 16:9, render at 1080p because YouTube films are often watched on a computer or a TV, and plan the story as chapters. Publish each episode, or a 30–90-second trailer for it, as a Short, then cut the season into a 10–20-minute compilation as the final short film.

How do AI creators make money?#

Rarely from views alone: YouTube Shorts pays about $0.02–0.12 per 1,000 views (creaticalc). The sci-fi creators above earn through bigger deals: WHITE_CACHE's pilot has production credits from a tool maker, and Gossip Goblin's feature Gods Don't Give Gifts opens in US cinemas on December 4, 2026 (The Daring Creatives).

Sources#

Channel figures were read from each channel's public page on November 1, 2026. Prices are from the calculator's catalog, checked on October 16, 2026, and for Kling 4.0 on November 3.