Skye Vale (@skyevale_) header
Skye Vale (@skyevale_) profile picture

AI films in every flavor 🍳 testing every new model Now cooking: MIDNIGHT COOKS 🍜 90s anime, new eps every week Free prompts, always

the midnight kitchen
Joined May 2018
Photo by Skye Vale (@skyevale_)Photo by Skye Vale (@skyevale_)Photo by Skye Vale (@skyevale_)Photo by Skye Vale (@skyevale_)Photo by Skye Vale (@skyevale_)Photo by Skye Vale (@skyevale_)Photo by Skye Vale (@skyevale_)Photo by Skye Vale (@skyevale_)Photo by Skye Vale (@skyevale_)
Claude building games is today's theme, so here's mine 🎮 GPT Image 2.6 drew every sprite, stage and effect Claude (Opus 5.5) built them into a 90s arcade fighter MIDNIGHT COOKS · 真夜中の料理人
I wanted to test Fable 5.2's quality with one-shot prompts without human feedback. Everything in this game is from scratch: models, sounds, textures. ~$550 at API pricing, ~2.7M output and 445M input tokens, heavily cached. Took 8 hours! Play it here: claude.ai/artifact/BGY1SztrA…
4
1
6
189
Everyone on the street froze. Except the mannequin. 🧍‍♀️ my take on @Mavern_0x's prompt · Seedance 2.5 | @NolviaAI
Everyone on the platform froze. Only one of them wasn't human. Created by Seedance 2.5 on @makeugc Prompt:👇 STYLE: authentic vertical smartphone video filmed handheld by an ordinary commuter on a subway platform. Real phone footage, not cinematic: auto exposure reacting to the train headlights, slight digital noise in shadows, mild motion blur on fast movement, autofocus occasionally hunting, slightly crushed blacks, cold fluorescent color cast. Underneath this texture everything is fully real: real people, real skin, real steel, real gravel, real physics. FORMAT 15 seconds, vertical 9:16, real-time speed, one continuous uncut take, no slow motion, no speed ramps. Photorealistic live-action rescue drama with a dry comic ending. No text, captions, UI overlays, timestamp or music. REFERENCE Image #1 defines only the adult female humanoid robot. Preserve exactly: soft youthful adult face, brown eyes, natural brows, subtle makeup; long straight light brown hair with a center part falling past the shoulders; small stud earrings; slim adult build with long legs; fitted beige short-sleeved T-shirt with a small dark chest logo, hem ending just above the waist; high-waisted dark purple cycling shorts with a small front drawstring; white chunky running sneakers. Exposed mechanical waist between shirt hem and waistband: brushed silver mechanical blocks and actuators, small bright white LED lights, bundles of blue, yellow and black wires, metal brackets and fasteners, one blue cable running down her left hip. Her torso is always connected to her pelvis through this mechanism, same scale and construction from every angle. Do not copy the photo's background, crowd, lanyards, stanchions or floor. PHYSICS RULE The robot is far stronger and heavier than she looks. This is the only supernatural element and it is consistent: her landing sprays gravel and her sneakers sink into the ballast, she lifts an adult man easily, her waist mechanism visibly compresses under every load. Everything else obeys real physics: the man's weight, the train's mass and braking, falling bodies, gravel, dust and air pressure. CAMERA OPERATOR The operator is never visible, not even in reflections on train windows. He stands on the platform about three meters back from the edge, filming at chest height. The camera looks diagonally along the platform toward the tunnel: the track runs along the left side of the frame, the tunnel mouth is in the far left background, the train approaches from the background and passes along the left side of the frame, never toward the camera. Real handheld phone behavior: micro-shake, breathing sway, a slight tilt, subjects drifting off-center, late autofocus, exposure dimming as the headlights grow. He reacts physically: steps toward the edge to look down at the tracks, flinches back as the train enters, then steps forward again. These imperfections never hide the key moments: the fall, the jump, the lift, the man reaching the platform, her vault out, the train passing, the phone handover and both lines. CHARACTERS The Man: early thirties, average build, short dark hair, light stubble, grey hoodie, dark jeans, black sneakers, white wireless earbuds, holding a phone. Tired commuter face. Passengers: about ten adult commuters with backpacks and work bags, ordinary clothing, real faces. A woman in a beige coat and a heavy-set man in a dark jacket stand closest to the edge, near the Man. Robot: calm, focused, silent. Her face stays composed; effort shows only in her body and mechanism. No bulging muscles, no grimace. The Man's arc: distracted, shock and pain, terror, breathless gratitude, confused deadpan. Passengers' arc: idle, panic, shouting, stunned silence. LOCATION AND BLOCKING Evening underground metro station on an overhead-wire line: a catenary wire runs above the track, there is no electrified third rail on the ground. Tiled walls, round concrete columns, yellow tactile safety strip along the platform edge, cold fluorescent ceiling lights, dark tunnel mouth in the background. The platform edge is about 1.1 meters above the track bed. Track bed: two steel rails on concrete sleepers, grey gravel ballast, dust, a little litter. Fixed positions: the robot stands by a column about two meters from the edge, in the left-center of the frame. The Man stands on the yellow strip about two meters further along the platform toward the tunnel, clearly visible. The woman in the beige coat and the heavy-set man stand just behind him. TIMELINE 00:00-00:02 | WAITING Quiet platform. The robot stands by the column, waist LEDs glowing softly. The operator films her curiously, focus pulling onto her mechanism. The Man stands on the yellow strip, earbuds in, scrolling his phone. Faint tunnel rumble. 00:02-00:04 | THE FALL A passenger with a large backpack hurries past behind the Man; the backpack clips his shoulder. He stumbles, one foot steps off the edge and he falls onto the track bed, landing hard on his side on the gravel between the two rails. One earbud pops out onto the gravel. His phone skids and stops face down on the gravel just beside the far rail. He grabs his ankle, grimacing, tries to stand and collapses back. The woman in the beige coat screams "Oh my God!". Passengers rush to the yellow strip and stop at the edge. The operator steps forward and tilts down at the Man. At 00:03.5 the tunnel walls light up: two train headlights appear deep inside the tunnel. A loud horn sounds. 00:04-00:05.5 | THE JUMP The heavy-set man shouts "Get up! Get up!". The Man looks toward the headlights, frozen. The robot moves: two fast steps to the edge and she jumps down, landing on both feet beside the Man. Gravel sprays, her sneakers sink into the ballast, her waist mechanism compresses on impact and extends. The train is now visible leaving the tunnel, about seventy meters away, already braking hard: wheels screeching, sparks under the bogies. 00:05.5-00:08 | THE LIFT Preparation: she crouches and grips the Man under both arms, his hoodie bunching in her fists. Lift: in one smooth powerful motion she straightens up and raises him to the platform edge, her waist actuators compressing with a mechanical whine. His chest and arms land on the yellow strip. The woman in the beige coat and the heavy-set man grab his arms and hoodie and drag him fully onto the platform. All three fall backward onto the tiles. He is safe at 00:07.5. The robot turns and scoops his phone off the gravel with her left hand. 00:08-00:11 | THE ESCAPE The headlights flood the frame, exposure dims. The train is about twenty meters away, braking, still dangerous. Phone in her left hand, she plants her right palm on the platform edge and vaults up in one movement, knees tucked, landing in a crouch on the yellow strip at 00:08.8. At 00:09.2 the front of the train slides past behind her at moderate braking speed, half a meter from her heels. The air push lifts her hair forward and flaps her T-shirt; passengers' hair and coats move. The operator flinches back, frame shaking. The carriages roll past, slowing, and the train stops along the platform at 00:11 with one door stopping directly beside the robot and the Man. 00:11-00:13 | GRATITUDE Brakes release with a hiss. Silence on the platform. The Man sits on the tiles between the two passengers, breathing hard, holding his ankle. The robot stands, brushes gravel dust off her shorts once, steps to him and holds out his phone. The screen is completely shattered. The Man, breathless, in natural English: "...Thank you." Spoken 00:11.6-00:12.3. He takes the phone and stares at the cracked screen. 00:13-00:15 | THE ENDING The door beside them opens with a hiss and a chime at 00:13. The robot calmly steps into the carriage, unhurried, waist LEDs glowing, and takes a handrail, expression unreadable. The doors close behind her at 00:14. The Man looks from his shattered phone to the closed doors and says flatly, in natural English: "...That was my train." Spoken 00:14.0-00:14.9. The train begins to move. At 00:15 the recording stops abruptly, no fade. IMAGE DETAIL Real skin pores, stubble, sweat, flushed skin. Individual hair strands with natural inertia, reacting to the train's air push. Hoodie and jeans with gravel dust after the fall. Grimy tile grout, scuffed yellow strip, dust in the air lit by headlights, brake sparks under the wheels. Real metro carriages: scratched paint, dirty windows, warm interior light. Waist mechanism: brushed silver, white LEDs, colored wires, real reflections of fluorescent light, real occlusion shadows. Spiderweb cracks on the phone glass. No legible signs, ads or logos. AUDIO Smartphone microphone: compressed, loud sounds slightly clipped, handling noise. Station ambience: echoing hall, fluorescent hum, distant footsteps, tunnel rumble growing. Synced sounds: backpack bump and the Man's thud on gravel at 00:02.5; phone clatter on gravel; horn at 00:03.5 and again at 00:06; gravel crunch of her landing at 00:05; screeching wheels and brake squeal from 00:05 to 00:11; mechanical whine of her waist actuators during the lift; scrape of the Man dragged over the edge; her sneakers on tiles at 00:08.8; heavy rumble and air rush of the train passing from 00:09.2; brake release hiss at 00:11; door hiss and chime at 00:13; doors closing at 00:14; train starting to move. Voices: woman's scream "Oh my God!" at 00:02.8; heavy-set man's "Get up! Get up!" at 00:04.2; passengers' gasps and shouts during the lift; operator's sharp breath as the train passes; silence after the train stops. The only lines after that: the Man's "...Thank you." and "...That was my train." The robot is silent. No music. CONTINUITY Fixed order: waiting, fall, headlights, jump, lift, Man dragged onto platform, she picks up the phone, she vaults out, train passes and stops, phone handover, "Thank you", door opens, she boards, door closes, "That was my train", train starts, recording stops. The Man stays sitting on the platform floor from 00:07.5 to the end and never boards. The phone stays on the gravel until she picks it up, then stays in her left hand until she gives it to him. The earbud stays on the gravel. The train never touches anyone. Only one train. Gravel dust stays on her shorts and his hoodie. Consistent faces, wardrobe, light and station layout. NEGATIVE CONSTRAINTS No cinematic camera look, no 4K sharpness, no gimbal smoothness, no color grading. No cuts, replays or slow motion. The robot never teleports or flies; she jumps down and vaults up physically. The train never drives toward the camera and never stops instantly. No third rail, no electric arcs, no glow other than headlights and brake sparks. No glowing eyes, energy effects or holograms. No cartoon look, plastic skin, bulging muscles, extra limbs, fused fingers, sliding feet, bodies passing through objects or morphing. No legible text anywhere. Camera operator never visible. STYLE REMINDER: shaky, real smartphone footage of a real event on a subway platform.
4
113
ChatGPT Images 2.6 can make GIFs now, so I turned my anime series into a 90s arcade fighting game 🎮 every sprite, portrait, stage and effect: GPT Image 2.6 try it here: @NolviaAI built into a game by Claude Opus 5.5 MIDNIGHT COOKS · 真夜中の料理人
Images, upgraded—introducing ChatGPT Images 2.6. Now with GIFs and reference images. Plus, two new models are coming to the API with 50% faster image generation.
4
1
5
247
zombie apocalypse, but the night-shift ramen cook can vanish into steam 🍜 turns out they weren't hungry for brains. this is Kiri(霧) from MIDNIGHT COOKS, live-action my take on @ZentrixHQ's prompt · Midjourney + Gpt-image + Seedance 2.5
This AI video turned a zombie apocalypse into a superhero playground. Turns out, super strength makes a terrible day look fun. Created by Seedance 2.5 on @makeugc Prompt: TELEKINETIC AWAKENING — ZOMBIE GAS STATION A 30-second horizontal 16:9 photorealistic supernatural action scene. A young woman escapes a zombie-filled store at an abandoned wasteland gas station at dusk, her violet telekinetic power awakens, and she takes down EIGHTEEN zombies with seven moves: a mid-air freeze, a head-to-head collision, a barrel volley, a lifted car, a car-door shield and a shockwave. It ends with her walking alone down the empty road while the whole horde lies still behind her. MAIN SUBJECT The woman from Image1, an adult in her early twenties, East Asian, dark brown eyes. Hair: a choppy chin-length bob, hot pink at the roots fading to violet and bright blue at the tips, ONE small silver lightning-bolt clip on the left side. Outfit: a fitted pink-and-white sports crop top with white stripes; a pink pleated mini skirt with white inner pleats and two white stripes on the hem, with pink fitted shorts underneath; a pink backpack strapped tightly to her back with black buckle straps. Footwear: knee-high glossy black combat boots on chunky pink platform soles with pink straps, silver studs and metal rings. All fabric is dense, opaque and stays in place during every move. Athletic, toned build. Frightened at first, then cold and focused. Preserve her face, hair colour, outfit and boots EXACTLY as in Image1. ZOMBIES Adult men and women with grey, dusty skin, cloudy white eyes, matted hair and torn, faded, dusty clothes. Jerky, uneven movement, some lunging fast. No blood, no wounds. LOCATION An abandoned gas station on a desert road at dusk. Cracked asphalt lot about 40 metres wide. A small convenience store with shattered windows and a dark doorway on the left. TWO rusted fuel pumps 6 metres apart under a sagging metal canopy 5 metres high in the centre. FOUR rusted cars: CAR A, a grey sedan parked in the open lot 3 metres outside the canopy edge; CAR B, a brown hatchback near the store; CAR C, a rusted pickup beside the canopy; CAR D, a white van far out by the road. FOUR steel oil barrels stand in the open lot between the canopy and the road. Background: the empty road to the horizon and a ruined city skyline about 10 kilometres away under a heavy overcast sky. Wind blows sand and dust across the lot. CAMERA AND LOOK Photorealistic live-action action film, real stunt choreography. ONE light source: soft overcast dusk daylight with diffuse light, soft shadows, muted dusty palette. The only other light is the violet glow of her power reflecting on asphalt, glass and metal. Handheld shake, slight delayed autofocus, mild bloom, faint film noise. Shots last 1.5-2.5 seconds, cut after each completed move; the car lift is ONE continuous shot. No camera or crew in frame. No glossy beauty lighting, no heavy colour grading. TELEKINESIS EFFECTS Luminous violet energy with a deep purple edge. Thick slow purple smoke coils around her hands and forearms up to the elbow, thin electric arcs crackle between her fingers, tiny violet particles drift upward like embers. Anything she holds gets a thin glowing violet outline with short arcs and a faint air distortion, hangs with real weight and sheds dust. Each impact gives ONE sharp violet flash. Open palm to hold, fist to crush together, push to throw, palm on the ground for the shockwave. The energy stays on her hands and never recolours her hair or clothes. CONTINUOUS STORY 00:00-00:03.5 — BREAKOUT Start in motion: she bursts backward out of the dark store doorway, facing inside, as a zombie's hand grabs her wrist from the darkness. She twists free and falls onto the asphalt. Low wide shot: TWO zombies vault over the hood of CAR B toward her and land heavily. She scrambles back on her hands until her back hits the first fuel pump. 00:03.5-00:07 — FROZEN IN THE AIR (zombie 1) The zombie from the store follows her out and leaps at her face. She throws up ONE open palm; violet sparks explode into smoke and arcs around her forearm, and the zombie freezes in mid-air 1 metre from her face. Close-up: looking at it, her fear turning cold, she says quietly: "Not today." She flicks her wrist; the zombie flies 10 metres back into the store wall, cracks the stucco and drops. 00:07-00:10 — HEAD TO HEAD (zombies 2-3) She rises to her feet. The TWO zombies who vaulted CAR B rush her from left and right. She clenches her fist; both are jerked toward each other, collide head to head with a dull thud and fall. 00:10-00:15 — BARREL VOLLEY (zombies 4-7) She walks out to the edge of the canopy. FOUR zombies stagger toward her from the road. She spreads both arms: the FOUR barrels rise together and hang in a row at chest height, glowing violet. She pushes both palms forward in ONE motion; all four barrels fly at once and knock all FOUR zombies flat, clanging and rolling. She stays at the canopy edge, CAR A 3 metres to her right, breathing hard: a tight group of SIX zombies runs out from behind CAR D toward her, and ONE more crawls to the edge of the canopy roof above her. 00:15-00:20 — THE CAR (ONE continuous shot, zombies 8-13) Three-quarter wide shot keeping her boots, CAR A and the zombies in frame: she plants her feet and raises both hands, and CAR A lifts off the ground, glowing violet; the wheels hang loose, sand and rust pour from its underside, and the asphalt cracks under her boots. The car rises to 3 metres in the open air outside the canopy. She swings both arms in a flat arc across her front; the car sweeps through the group and knocks all SIX zombies off their feet, then lands on its roof in the open lot, rolls once and stops on its side, glass bursting out. 00:20-00:23.5 — DOOR SHIELD (zombie 14) The zombie on the canopy roof leaps down at her from 5 metres. In the same moment she jerks her hand toward CAR C; its driver's door rips off its hinges with a metal shriek and stops 1 metre in front of her, upright, as a glowing violet shield. The falling zombie slams into the door and bounces onto the asphalt; she pushes forward and the door drops flat on top of it, pinning it. In the background, FOUR zombies come out of the store doorway. 00:23.5-00:27.5 — SHOCKWAVE AND VORTEX (zombies 15-18) The FOUR zombies from the store surround her at arm's length. She drops into a crouch and slams one palm onto the asphalt: a low ring of violet light and air distortion bursts outward and throws all FOUR zombies 5 metres away onto their backs. As she stands, broken glass, sand and small debris rise around her into a glowing violet vortex at waist height. She opens her hand and everything drops at once. 00:27.5-00:30 — WALK AWAY Medium side shot: she turns her head toward the defeated horde, face in clear profile, and says flatly: "Stay down." Wide shot from behind: she takes a few steps away down the empty road past CAR A lying on its side, pink backpack and platform boots clearly visible, violet smoke fading from her fingers, the ruined skyline ahead. DIALOGUE Two short lines in English, calm and low, never to the camera: "Not today." at 00:05 to the frozen zombie, "Stay down." at 00:28 in profile to the defeated zombies. No other speech. SOUND Natural and event-driven: glass crunching at the doorway, zombie groans and snarls, her fast breathing, feet landing on the car hood, a deep rumble and electric crackle as the energy appears, a low hum while she holds objects, the thud against the store wall, the double thud of the collision, four barrels clanging, the car's metal groaning, sand pouring, the crash and glass bursting, the door ripping off its hinges and the zombie slamming into it, a heavy low boom for the shockwave, the rush of the vortex and glass falling, wind and her footsteps at the end. No music, no narration. ESSENTIAL CONTINUITY - Order: breakout, frozen zombie, collision, barrel volley, car, door shield, shockwave and vortex, walk away. - EIGHTEEN zombies total: 1 from the store, 2 car vaulters, 4 from the road (barrels), 6 from CAR D (car), 1 from the canopy roof (door), 4 from the store doorway (shockwave). Each is knocked down ONCE and stays down. No other zombies appear. - She never touches lifted objects or zombies with her hands. Lifted objects have real weight and gravity, no clipping; CAR A never hits the canopy, pillars or pumps. - The store wall stays cracked, the barrels stay scattered, CAR A stays on its side, the door from CAR C stays on the pinned zombie. - Violet energy only on her hands and forearms, on held objects, in impact flashes, in the shockwave and in the vortex. - Preserve her face, hair, outfit and footwear exactly as in Image1 throughout. No slow motion, teleportation, duplicated people, flying heroine, extra dialogue, talking to camera, changing outfits, changing hair colour, lost backpack, extra limbs, distorted hands, guns, blood, gore, explosions, fire, cartoon or anime style, captions, subtitles, readable signs, watermarks. #zombies #aivideo
6
1
6
399
happy dragon's day 🐉 Shigure 時雨 doesn't summon a dragon, she asks the rain to become one. MIDNIGHT COOKS · 真夜中の料理人
Say it with art: Happy Dragons Day 🐉 Good morning. Rise like fire, fly like legend. 🔥
2
6
70
POV: it's 3am at the midnight kitchen and you're feeding kitsune udon to actual kitsune 🦊 one of them read the menu. (it's just tofu. it's always been just tofu.) my take on @devgetsit's prompt · Seedance 2.5
POV: it's 3am in Miami and you're the guy feeding caviar to mermaids Tested Seedance 2.5 on @makeugc. Costs $1 to try it yourself, setup in the replies 👇 Two of them saw $300 caviar. One of them saw relatives. Every group has one who reads the label. Is it cannibalism if you're only half fish? Prompt: Amateur vertical iPhone video, 9:16, filmed at 2am at a Miami hotel pool with the phone's harsh built-in flash, shaky handheld, grainy, slight motion blur, auto-exposure flicker, candid real home video look, not cinematic, no music. The man filming is kneeling on the pool edge, camera low and close, about half a meter from their faces. Layout locked for the whole video: the three mermaids are in the water right in front of him, leaning on the same pool edge; a low flat silver tray with crushed ice sits on the edge between the RIGHT mermaid and the camera, at the right side of the frame, holding two big open round caviar tins, one full of glossy black caviar and one full of bright orange-red salmon roe, and a few small blinis topped with heaps of black caviar; behind them the dark pool water, the far side of the pool, palm trees and dim hotel lights. The creatures: three real mermaids leaning on the pool edge with their forearms, close to the lens, their shimmering fish tails visible under the water behind them. All three are beautiful glamorous Miami women with hugely overfilled, giant, glossy duck lips, so swollen and stiff they can barely open their mouths, clearly visible from the very first frame, long fake lashes, heavy contour makeup, long stiletto acrylic nails. Three clearly different women, not twins: LEFT: honey blonde with wet beach waves, bronzed skin, peach glossy lips, coral tail. CENTER: dark-skinned beauty with long sleek straight black hair, plum glossy lips, violet tail. RIGHT: fair-skinned girl with long straight strawberry-red hair and light freckles, cherry-red glossy lips, silver tail, quiet and serious. The man's hand: the only male hand in the video is the man's own hand from the camera side: a big tanned masculine hand with hairy wrist, thick fingers and short nails. Each mermaid has only her own two hands; no extra hands appear. Important: they speak slowly, slurred and muffled, barely moving their giant lips, as if their mouths can't open. Nobody laughs at any point. All three are alive and moving every second: no one ever freezes or stares blankly while another acts. They glance at each other, at the food and at the camera, shift their weight on the edge, flick their tails under the water, react to each other with real, different, asynchronous movements. The LEFT and CENTER ones are greedy like real animals around food. The RIGHT one never eats anything: she watches the caviar and the other two with growing worry. 0-2s: The flash hits all three faces at once, faces big in the frame. The CENTER one leans in closest and says slowly through barely moving lips, muffled: "Feed us." At the same time the LEFT one pushes her shoulder against the CENTER one to get closer to the lens, and the RIGHT one stares at the caviar tins on the tray next to her with a worried frown, her tail swishing under the water. 2-5s: The man's hand holds out a blini heaped with black caviar to the LEFT one. She snatches it and tries to push it into her mouth, but her swollen lips barely part, the blini folds and the black caviar smears across her lips and chin, little black eggs sticking to her peach gloss. She mumbles, muffled: "It doesn't fit." Meanwhile the CENTER one grabs at the man's now empty hand with her nails, and the man pulls his hand back out of the frame. The RIGHT one looks from the smeared caviar on the LEFT one's lips to the tins, more and more upset. 5-8s: The CENTER one reaches across, grabs the tin of orange-red salmon roe from the tray and tips it straight into her mouth like a drink; only the CENTER one touches this tin. Her stiff swollen lips can't take it all, the orange eggs pour over her lips and drip down her chin, her cheeks balloon, she chews hard. The LEFT one licks the black caviar off her own fingers and reaches for the tin too but can't get it; the CENTER one shoves her hand away and splashes water at her with her tail. The RIGHT one watches them in horror, then quietly pulls the tray with the black caviar tin closer to herself. 8-11s: The RIGHT one hugs the silver tray to her chest with both arms, looks down at the black caviar tin, not at the camera, and says to it slowly and seriously through her stiff lips, muffled: "I'll save you." She sinks under the water with the tray, ice cubes scattering on the surface, her silver tail flipping up and slapping the surface. 11-13s: The LEFT and CENTER ones turn their heads sharply to the bubbles where she disappeared, mouths still full, caviar smeared all over their lips and chins. They look at each other, then keep chewing, the CENTER one still holding the empty orange tin and scraping the last eggs out with her finger, the LEFT one licking her lips, both tails swishing under the water. Sound: pool filter hum, water splashes, wet chewing, muffled slurred voices, ice cubes clinking, one loud tail slap, waves hitting the edge, the man's quiet stunned breathing. No laughter. Faces stay sharp, close and lit by the flash the whole time.
6
1
10
546
Same baby chaos, but 90s anime 🍜 This is Kiri(霧), 20 years before MIDNIGHT COOKS. (watch till the last second) my take on @Zarnab_with_Ai's prompt · GPT-image 2.5 + Kling AI
Made with seedance 2.5 Prompt 👇 Style: High-quality 3D animated family comedy, adorable baby character, Pixar-inspired cinematic animation, ultra-detailed textures, expressive facial animations, warm lighting, realistic movement, and professional cinematic storytelling. Character: An adorable chubby baby girl with big sparkling brown eyes, rosy cheeks, short dark brown hair, wearing a cute green knitted frog hat with big frog eyes and a white baby romper featuring a small green frog design. Scene 1 – Baby in the Crib: A cute baby girl stands inside a beautiful wooden baby crib in a cozy nursery. She holds the wooden bars with both hands, looking innocent and curious. Warm golden sunlight enters through the window, creating a peaceful atmosphere. Scene 2 – Funny Baby Expressions: Show a close-up of the baby making adorable funny facial expressions, squeezing her eyes, puffing her cheeks, and smiling playfully while wearing her oversized frog hat. Scene 3 – Baby Escapes: The baby cleverly crawls out of her crib and moves across the soft carpet. She explores the nursery with excitement, crawling behind furniture and hiding playfully. Scene 4 – Cute Hiding Moment: The baby hides behind a large cozy armchair. Only her little face and frog hat are visible. She peeks over the chair with a mischievous smile, then hides again. Scene 5 – Mischievous Adventure: The baby attempts to climb a nearby white dresser, reaching toward its drawers with her tiny hands. Show her funny and determined expressions. Scene 6 – Unexpected Daddy Reaction: Suddenly, her father appears in the nursery and is shocked to discover the baby's mischievous adventure. Show his exaggerated surprised facial expression, wide-open eyes, and funny reaction. Scene 7 – Emotional Funny Ending: The father discovers the adorable baby sitting safely in a little chair. He looks at her in disbelief while she smiles innocently. End with a cute close-up selfie-style shot of the father and baby together. Visual Quality: Ultra-detailed 3D animation, realistic knitted fabric, soft baby skin, expressive eyes, natural body movements, cinematic depth of field, warm golden-hour lighting, smooth camera transitions, professional animation, 4K quality, vertical 9:16 aspect ratio. Important: Maintain consistent baby character design, identical frog hat, clothing, facial features, and nursery environment throughout every scene. Make the story funny, heartwarming, adorable, and highly engaging.
5
5
119
MIDNIGHT COOKS · ep.1 🍜 雨と湯気 \ Rain and Steam After 2am, cooks fight in closed kitchens. One rule: no blades. 🍜 Kiri vanishes into steam. She's hunting whoever burned down her family's ramen stall. ☔️ Shigure commands the rain. Then someone brings a sword... ▶ start here · ep.0 below · follow for ep.2
湯気と屋台 \ Steam and Stall 🍜 my take on @aimikoda's prompt · Midjourney + Seedance 2.5 | @NolviaAI
1
3
87
Your prompt said "avoid: anime"… so here's the anime version hehehe my take on @AIwithSynthia's prompt · Midjourney + Seedance 2.5 on @NolviaAI
AI just peaked here 🥵 Seedance 2.5 Prompt : Create an ultra-photorealistic Japanese supernatural martial arts battle at an ancient temple during heavy rain. A young Japanese female shinobi in a crimson-and-black outfit confronts a powerful male samurai in black armor wielding a massive katana. They exchange angry Japanese dialogue before the samurai launches a brutal attack. The woman dodges his sword strikes, counters with powerful punches, spinning kicks, flying attacks, and intense close-quarter martial arts. The samurai unleashes a dark energy shockwave, cracking the ground and spreading black flames. The woman awakens her supernatural power, her eyes glowing icy blue as ancient symbols appear on her arms. She performs mystical hand seals and summons a gigantic spectral black dragon with glowing blue eyes. The final battle erupts as the samurai attacks with his flaming katana while the dragon dives toward him. Epic sword clashes, explosive impacts, aerial combat, powerful kicks, and spectacular supernatural energy effects follow. The woman finally disarms him with a devastating spinning kick, while the dragon pins him to the ground. She calmly says: 「もう終わりよ。」 and walks away as the dragon dissolves into glowing blue particles. Visual Style: Hollywood-level live-action cinematography, realistic martial arts choreography, dramatic camera tracking, slow-motion impacts, heavy rain, volumetric lighting, detailed facial expressions, realistic physics, cinematic supernatural VFX, and perfect character consistency. Audio: Natural rain, footsteps, breathing, Japanese dialogue, sword clashes, impacts, explosions, and dragon roars. No music, narration, subtitles, or text overlays. Avoid: Anime, cartoon visuals, video-game graphics, cheap CGI, distorted anatomy, inconsistent faces, blood, and gore.
2
4
58
湯気と屋台 \ Steam and Stall 🍜 my take on @aimikoda's prompt · Midjourney + Seedance 2.5 | @NolviaAI
影と社 \ Shadow and Shrine Seedance 2.5 Prompt in the replies. 👇
3
1
10
367
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
2,367
7,262
5,753
55,940
10,053,546
plot twist: the guy filming trained the crow 🐦‍⬛ my take on @techhalla's prompt · GPT-image 2.5 + Seedance 2.5
i think she's in cahoots with the cat prompt 👇
1
4
297
Halloween 2003. dad's camcorder. count the kids. 👻 (made with Seedance · prompt 👇)
5
7
116
【Reference】@16f2776d-7488-44cb-aa1f-82b6413df46b Only reference for image quality, and the facial features & costumes of the two children. Standard-definition footage from a 2003 consumer home DV camcorder; camcorder fill light illuminates the foreground, cool dark blue background, heavy video noise, soft focus. Do NOT start the shot using the composition from @16f2776d-7488-44cb-aa1f-82b6413df46b. Composition and actions are free to interpret. Do NOT copy the ghost decorations hanging from the trees in @16f2776d-7488-44cb-aa1f-82b6413df46b. 【Overall Style】4:3 horizontal aspect ratio, 30 seconds. Maintain the 2003 consumer home DV camcorder look from @16f2776d-7488-44cb-aa1f-82b6413df46b throughout: standard definition, soft focus, prominent video noise, blown-out highlights, muddy shadows, cool blue twilight tint, foreground lit by the camcorder’s built-in fill light. Do NOT sharpen the image at any point, no cinematic quality, no cinematic lighting. Shot hand-held by an unseen father; subtle shaky movement, autofocus hunts occasionally, casual framing, authentic vintage VHS/DV tape aesthetic. One continuous take, no cuts, no subtitles, no legible text anywhere in frame. 【Visuals】 0–5s: Dusk, just after dark. On the front porch steps of a house, the witch girl and dinosaur boy from @16f2776d-7488-44cb-aa1f-82b6413df46b hold orange pumpkin candy buckets, wave at the camera and shout “Happy Halloween!” (audio should not be loud or crisp, with background hiss and ambient noise like a genuine DVD home recording). At the edge of frame, a mother’s hand reaches out to straighten the girl’s pointed hat. The camera pans briefly across the street; far away on the neighbor’s lawn stands a faint figure of a child draped in a white bedsheet with two black eye holes, completely still, very unnoticeable. The camera does not linger here. 5–12s: The camera follows the two children along the sidewalk. Along the curb are inflatable pumpkin decorations, trees draped with fake spider webs, and other costumed kids out trick-or-treating. The two kids run up to the first house and call out “Trick or treat!”. A smiling neighbor lady drops candy into their buckets. 12–20s: The camera follows them to a second house. Three children stand at the doorway: the witch girl, the dinosaur boy, and between them the child covered in a white bedsheet with two black eye holes, quietly holding an empty old pillowcase. Off-camera, the father laughs and asks: “Who's your friend?”. The witch girl turns back, confused, and replies: “I don't know… he's been following us.” 20–27s: Slow digital zoom push-in (digital zoom makes the image softer, grain/noise heavier) onto the bedsheet child. He stands with his back to the camera, then turns his head extremely slowly, his two black eye holes staring straight into the camcorder lens. 27–30s: The frame freezes with stutter, blocky digital pixelation, color corruption artifacts and horizontal scan interference lines, the image shakes, then cuts to black. 【Audio】Recorded with the built-in mono microphone of a 2003 home DV camcorder: muffled midrange-heavy tone, weak low end, rolled-off high frequencies, constant tape hiss plus faint mechanical whir of the camcorder motor; occasional wind pop noises hitting the mic. The father’s voice, close to the microphone, is loud and prone to distortion when he laughs. Kids’ shouts, the neighbor lady’s greeting, doorbell chimes and distant barking sound quiet and muffled. In quiet moments, auto-gain amplifies the background hiss. When the bedsheet child turns his head, surrounding voices fade away, leaving only amplified hiss plus a low, droning hum. As the video glitches, audio stutters, warps and stretches, mixed with harsh digital distortion, then cuts off abruptly. All dialogue is in English.
47
there's a whole town inside this pumpkin 🎃 and somewhere in that town, someone is carving a pumpkin… prompt 👇
4
11
382
A few thoughts on agentic video editing I’ve had over the last few days: 1. 99% of ads should be edited agentically now. If your editors are still editing in the timeline, they’re falling behind. I got a message from my senior VE saying that he’s putting out 5x edits/day now at the same quality he was before. 2. The conversation with video editors will be uncomfortable because the timeline is all they’ve ever known. Some will be hesitant and some will be fearful for their job. I think the best way to address this is head on. The tech is inevitable, but having taste is still going to be incredibly valuable to train and give feedback to agents. Those who have this will become more valuable, those who don’t will struggle. 3. Opus 5.5 can natively edit all of your ads without needing any MCPs or plugins. I’ve tried the alternatives and couldn’t find anything that was significantly better than doing everything inside of Claude. 4. Have your senior editor(s) build skills for each style of edit (e.g. mashups, UGC, AI animations…). You can build these by feeding 4-5 inspo edits to Claude and tell it to make a skill out of it. I prefer building lots of mini-skills to one big video editing skill, so that the context stays as relevant as possible. 5. Once you have a base skill (e.g. ugly ads edits), the best way to train it is through feedback loops. Have the skill re-edit the last 25 ads each senior edited then put them in an artifact, and have the senior VE give feedback on each. Keep doing this until the skill is refined and it understands what a good edit is. 6. If you do this, there’s no reason why you shouldn’t be able to make edits that need little/no feedback. Most of ours are approved after v1 or v2, sometimes v3. 7. Only A-player editors should be training the skills. 8. Re adoption in the team, the AI champion builds the v1. Then get out of the way. I’m not the best editor on my team, so my job is to build the harness that lets the best editors train the skills, not build the skills myself. 9. If you want to supercharge this process, have your top editors record their screen as they edit. Turn on Gmeet or loom and get them to narrate the reasoning behind their process. E.g. why they chose this cut, pacing or b-roll clip…this is all context that gets fed into your skills. 10. It takes time to build, train and refine these skills. You have to go through the dip to get out the other side. Budget time in for senior VEs to work on this - otherwise you’ll end up with skills that put out subpar edits and don’t save you much/any time. 11. Operate under the assumption that this is the worst video editing is ever going to be. Opus 5.5 can do most of our editing today…what will Opus 6/Fable 6 be able to do? The editing skills you build are only going to become more valuable as the tech evolves. I still don’t think people realise how big video editing being unblocked is. Things that I thought were 3-6 months away are now on the table. Will continue to share more as I learn. CRAZY, crazy times.
197
139
110
2,566
554,455
Tried a long shot through 1930s Shanghai with Seedance draft mode.. what does the plum blossom mean? prompt 👇
this means AI camera now can film any complicated shot.. prompt 👇
3
1
7
238
What if Batman was… compiled? 🦇💻 My nerd version: // 7 eras. 7 engines. 0 paintbrushes. 1939 — Silent Film 1966 — 80s Cel Anime 1989 — Woodcut 1992 — Ink Wash 2008 — Tech Keynote 2022 — Found Footage ??? — Hologram HUD Which one’s yours? 🤓
🤖 Made with AI
Toshi's Daily Challenge: Bat 🦇 Not the hero we asked for, but the one this challenge deserves. Show me your Bat.
1
8
205
Ross & Rachel, finally on the same page. 📖🦞 Made by Claude Opus 5.5 in Claude Code: 100% code, edited in CapCut, no video model. #Friends #WeWereOnABreak #PaperArt #opus55
1
2
3
119