Joe Lin

[Maciej Dobrodziej @M_Dobrodziej]

Prompt: https://t.co/7EJ9fJHgiK

Metoda w skrócie: klatka startowa = zdjęcie z narysowaną trasą, linia i cyfry znikają w pierwszej połowie sekundy, potem kamera wykonuje zaplanowany ruch.

[Kōda @aimikoda]

Seedance 2.5 Prompt:
A live-action thriller follows Sue leaving a luxury-hotel elevator and fighting bare-handed down the corridor as ten masked attackers try to stop her.

@[char ref] is the sole authority for Sue's complete appearance. Sue wears exactly its outfit, footwear, accessories, hairstyle, proportions, and materials throughout, with no redesign or substitution; ignore the white sheet, text, repeated views and poses.

The five-star corridor has brass elevator doors, burgundy carpet, warm sconces, suite doors, and alcoves. Exactly ten distinct attackers wear dark tailored suits and black cloth masks. All ten start in the corridor between the elevator and penthouse. No attacker starts inside the elevator; no bystanders.

Everyone remains unarmed. Sue uses grounded Wing Chun: centerline pressure, triangular footwork, pak sau, tan sau, bong sau, lap sau, chain punches, palm-heels, short elbows, and low stop kicks. Attackers actively box, kick, grab, body-lock, flank, and stagger pair attacks; Sue changes lanes and angles one into the other's path. Fallen attackers stay down permanently.

No dance occurs before the tenth attacker is fully down. Sue performs one party-girl dance only at the end.

Opening lock: Sue is the elevator's sole occupant. No attacker, limb, shadow, reflection, or silhouette appears inside the car. The first attacker waits out of frame in a corridor alcove three meters from the threshold. He appears only after Sue exits completely and the doors close, and he never crosses the threshold. Every fight occurs on the corridor carpet.

Use one continuous wide-lens Steadicam take. Begin facing the elevator and frame only Sue inside. Hold as she steps onto the carpet, revealing the empty car, then close the doors before revealing the first attacker. Lead Sue backward from the front along one reachable corridor path, arcing around contact and bodies. No cut, teleport, hidden transition, axis reset, slow motion, speed ramp, acrobatics, wirework, weapons, glowing trails, or energy effects.

0-2 seconds: The doors open on Sue alone. She exits unopposed, both feet reach the carpet, the empty elevator is clearly visible, and its doors close. No attacker is visible during this exit.

2-4 seconds: The first attacker now enters from the corridor alcove and attacks on the carpet; Sue uses pak sau, a straight palm, and a short elbow, then drops him beside the wall. Neither approaches the elevator threshold.

4-8 seconds: The second and third attackers converge, one punching and one grabbing; Sue lines them up, drops the second with lap sau and chain punches, then stops the third's renewed rush with a low kick and palm. Both stay down.

8-11 seconds: The fourth attacker charges from an alcove with a body lock; Sue turns him with bong sau and lap sau, finishes with short elbow-palm contact, and advances as he falls.

11-15 seconds: The fifth attacker engages Sue alone. The sixth stays three meters behind and outside engagement range, never striking, grabbing, or touching her. Sue drops the fifth with pak sau and compact centerline punches. Only after the fifth lies fully down does the sixth advance and begin a separate one-on-one body-lock attack; Sue turns him off-line with bong sau and lap sau, then drops him with a short elbow and palm. The two attacks never overlap and the men never engage Sue simultaneously.

15-18 seconds: The seventh attacker feints, blocks Sue's first counter, and returns a hook; Sue regains centerline with bong sau, lap sau, and chain punches, then drops him and advances.

18-23 seconds: The eighth and ninth coordinate front and rear pressure; Sue pivots the eighth into the ninth's path, drops him with elbow and palm, then intercepts the ninth's renewed rush with pak sau, a low kick, and a straight palm. Both stay down.

23-27 seconds: The tenth attacker blocks Sue's first chain punches, clinches, and drives her sideways into a suite door; Sue keeps her feet, works inside through tan sau, bong sau, and lap sau, then drops him with compact body punches and one palm-heel. All ten attackers are now down.

27-30 seconds: Sue performs the only dance, a brief confident shoulder shimmy, hip pop, and playful half-turn. The half-turn triggers a continuous camera move around her left side through a visible 180-degree arc. Without cutting, the camera settles directly behind Sue on the same corridor centerline, matches her forward direction, and follows her toward the penthouse. Fallen attackers remain fixed behind the moving camera.

Photoreal live action with realistic cloth, friction, impacts, and body weight. No dialogue and no BGM. Use elevator chime, doors, footsteps, suit cloth, breath, contact, wall impacts, and natural falls. Keep one Sue and the same ten attackers; preserve Sue's exact referenced outfit, identities, masks, positions, layout, and lighting with no duplication, replacement, disappearance, or respawning.

[Emily @IamEmily2050]

Almost one year old and Nano banana Pro still the best model for these types of images, Hopefully Nano banana Pro V2 soon 🤞

{
"aspect_ratio": "9:16",
"subject": "Adult East Asian woman taking a close indoor portrait in a simple dressing-room setting.",
"appearance": {
"face": "Soft oval face, large almond-shaped dark eyes, straight brows, small nose, full glossy lips, light blush, smooth natural skin.",
"hair": "Very long dark brown hair with a loose center part, softly layered, slightly tousled, with a few strands falling across one side of the face.",
"expression": "Calm, slightly pouty, direct gaze into the camera."
},
"outfit": {
"top": "Fitted silver-gray short-sleeve ribbed top with a low neckline and decorative front lacing."
},
"pose": "Upper-body selfie-like portrait, body turned slightly to one side, head tilted gently toward the camera, one shoulder slightly raised.",
"setting": {
"location": "Plain backstage or dressing-room interior.",
"background": "Beige walls, suspended ceiling with rectangular light panels, a warm floor lamp in the background, a mirror and a few indistinct room details."
},
"lighting": {
"main": "Soft overhead room light.",
"secondary": "Warm lamp glow in the background.",
"look": "Even light on the face with soft shadows under the jaw and hair."
},
"camera": {
"framing": "Vertical close portrait from upper chest upward.",
"angle": "Slightly below eye level, similar to a handheld phone photo.",
"focus": "Sharpest on the eyes, lips, and front hair strands; background slightly soft."
},
"style": {
"image_type": "Natural beauty portrait",
"mood": "Quiet, polished, casual, intimate"
},
"key_details": [
"Keep the long dark hair framing both sides of the face and torso.",
"Keep a few loose strands crossing the face.",
"Keep the soft glossy lips and subtle eye makeup.",
"Keep the simple indoor room visible behind her."
],
"avoid": [
"No app interface icons, counters, or screenshot overlays.",
"No extra accessories or jewelry.",
"No heavy glamour retouching or plastic skin.",
"No exaggerated anime features.",
"No studio backdrop or luxury environment."
]
}

[Joe Lin @Shaolin_Liu]

@PhyrexNi 极限拉扯，谈恋爱都没费这么大劲的

[Joe Lin @Shaolin_Liu]

@geekbb 最近几天要突击努力工作，因为梁老板在催我😂

[Kōda @aimikoda]

Tried creating a Pixar-style short.

Generated in Seedance 2.5 at 480p then upscaled 4× with Topaz.

Still working on prompting for 2.5. This video is text to video, no references used. You can check out the prompt in the replies.

Generated and upscaled on @astorie_ai

[Umesh @umesh_ai]

Prompt : a 15 second split-screen video showing the same real-time action from two different camera angles. The screen is divided vertically into two equal halves. On the left side, show a static, wide overhead view from a ceiling-mounted camera positioned in the back corner of the room. The full room layout is visible, including a sofa, coffee table, rug, bookshelves, and a tall shelf near the wall. A cat is perched on top of the tall shelf. At the 2-second mark, the cat suddenly jumps down, twists and flips naturally in mid-air above the furniture, then lands gracefully on the coffee table. On the right side, show the exact same action in perfect synchronization from an extremely low-angle camera placed on the floor near the base of the sofa. The camera is tilted sharply upward toward the top of the shelf, creating a dramatic worm’s-eye view. The cat should appear high above the camera before jumping downward almost directly toward and over the lens. Capture the underside of the cat, its paws, body rotation, fur movement, and the rapid change in scale as it descends. The cat passes close above the camera before continuing forward and landing on the coffee table. Keep the camera mostly static, with only a subtle upward adjustment if needed to maintain the cat in frame. Both sides must show the exact same event with identical timing, body position, movement, landing point, lighting, room arrangement, and realistic physics. The two angles must remain precisely synchronized throughout the entire shot.

[Kōda @aimikoda]

We finally have an affordable API for nearly all the generative media models we actually use.

Connect it to Codex, build your own platform or create a custom workflow around it. The Pika API Club brings 70+ models into one API, with member pricing up to 80% lower than other aggregators.

This video was made entirely through the Pika API Club using Seedance 2.0, MiniMax H3 and MiniMax Music 3.0. Three 15-second Seedance videos already make the $10 fee back.

I had mostly avoided API workflows because platforms usually offered better pricing and additional benefits. This might finally make API usage worth it for independent creators too.

Check Pika API Club here: https://t.co/oNDj66cAYg

[Pika @pika_labs]

Today, we’re proud to introduce the Pika API Club—a membership that gets you access to the leading gen media models at previously impossible prices.

We were shocked by the 3x markup by many API aggregation platforms. So we built our own, negotiated better prices, and kept our markup as low as possible to pass the savings on.

It’s time for AI to empower the next generation of builders and creators, not price them out.

[Kōda @aimikoda]

The Seedance 2.5 guides everyone has been gatekeeping.

Check both before writing your prompts. The way they define reference roles, structure actions into stages, handle continuity and especially the syntax used for audio and text is pretty interesting.

🔗 User Guide: https://t.co/NhTWns3UpO
🔗 Prompt Guide: https://t.co/x16yx1oGSw

[Kōda @aimikoda]

Let's create a music video with these characters.

Used the Midjourney images directly into Seedance 2.0 and added the track I created in Suno as a blackout video reference. You can find the prompt below.

[AI Will @FinanceYF5]

国内开源了一个模型，随便拍段视频，就能实时重建出3D场景。

一个摄像头，不用LiDAR，1万多帧都不会崩，单卡就能跑到20fps。跑分比传统优化类方法更强，无人机航拍、行车记录、室内漫游都测过。

而且完全开源。

[Geek @geekbb]

中文网文和长篇小说作者缺一个本地跑、不依赖网页对话框的 AI 写作工作台。这个工具把大纲、人设、正文生成和改稿串成一条流水线。

项目配置、世界观、角色卡、章节蓝图、正文生成、审稿修稿、知识库和导入参考小说这些功能都集成在一个 Electron 窗口里。

https://t.co/hUj4pxsWom

[Kōda @aimikoda]

Midjourney prompt: soccer player, vibrant colors --no text --ar 16:9 --sref 74117661 --profile ycjrvus --stylize 1000 --preview

Seedream 5.0 pro prompt for storyboard:
Create a 16:9 monochrome 2D rough-sketch board for a football gauntlet and keeper finish; lock staging, and do not render bracketed labels.

[HEADER]
Use crisp type and thin rule; show only:
"Claim Through Five"
"Five defenders. One keeper. One strike."

[BOARD]
4x4 P## grid; 14 equal 16:9 landscape panels (width 1.78x height), incomplete last row centered.
Headers use `P## / short tag / short beat`; one frozen beat each.

[STYLE]
Monochrome 2D rough pencil/ink on white paper: open outlines, sparse construction, simple perspective, featureless figures.
No fills, wash, hatching, shading, shadows, ambient occlusion, color, 3D/clay volume, texture, or finished illustration.

[REFERENCES]
Image A: lead build, hair outline, wardrobe silhouette only.

[CONTINUITY]
One lead, one ball, five outfielders, one goalkeeper; lead and ball move screen right to goal, marks attach only at contact.
Outfielders have distinct builds and jerseys; goalkeeper stays inside goal, no extras.

[RULES]
Only header lines and panel headers are visible; no panel text, arrows, icons, insets, extra panels, duplicates, ghost poses, or object-like marks.

[BEATS]
Draw one storyboard panel per visual BEAT:
BEATS:
P01 / low void / ignition: Ground-height low full; leaning lead fills the left third, ball at outside boot, short floor trail aimed screen right.
P02 / ground evidence / toe cut: Ground-height foot close-up; outside boot meets ball lower left, opening opponent's lunging leg fills right, small unfilled burst at contact.
P03 / canted pressure / body lie: Knee-height canted medium-full; lead shoulders tilt left, hips and ball angle right, same opponent leans away in the right third.
P04 / lateral slash / second gate: Waist-height profile wide; lead and ball cut screen right, compact second opponent planted foreground left, short mark behind ball.
P05 / ball witness / split line: Ball-height tight wide; ball sits between lean third opponent's feet at center, lead boot fills left edge, compressed arc touches its path.
P06 / overhead trap / angle break: Overhead wide; broad fourth opponent blocks center, lead hooks ball lower-right, their body vectors forming a right angle.
P07 / compressed impact / drag escape: Eye-height compressed medium; fourth opponent's leg crosses foreground, lead coils around ball right, one flare at sole contact.
P08 / low pursuit / final wall: Knee-height low wide; narrow final opponent fills the right third between lead and goal, ball rises diagonally from lower left.
P09 / impact proximity / last cut: Ground-height tight full; lead foot guides ball beyond the final opponent's reaching toe, thin flare attached inside the gap.
P10 / release wide / open lane: Eye-height wide; lead and ball occupy the right third facing goal.
P11 / frontal threat / keeper set: Eye-height wide; lead and ball fill lower left, goalkeeper centers in the distant goal.
P12 / canted close / vow lock: Waist-height canted medium-wide; planted foot and ball sit lower left, wound torso fills left, goalkeeper inside the right goal.
P13 / worm impact / keeper dive: Ground-height extreme low wide; boot compresses ball foreground left, goalkeeper reaches across distant goal, impact ring points screen right.
P14 / net witness / beaten reach: Goal-side eye-height medium wide; ball deforms net center-right beyond the goalkeeper, sparse burst attached at contact.

Seedance 2.0 Prompt:
Use @[storyboard ref] only for P## staging: match each panel's angle, scale, framing, blocking, positions, and direction. Ignore sketch artifacts; written beats, references, ENVIRONMENT, VISUAL MEDIUM, and CINEMATOGRAPHY control final identity, materials, color, motion, and effects.
Use @[midjourney image] as C1's sole appearance source; never transfer its likeness to C2-C7.

VISUAL MEDIUM: Cinematic 3D realism fused with painterly anime surface abstraction.
MOTION LANGUAGE: Impossible acceleration, elastic poses, violent smears, grounded foot-and-ball contact physics.
VISUAL EFFECTS: Orange, violet, blue, black ribbons; ink splinters, contact flares, ball wakes.
CINEMATOGRAPHY: High-key overhead light, saturated accents, ultra-wide depth, aggressive diagonals.
AUDIO: No music; breath, cleats, ball impacts, net snap. Internal Japanese voice, no lip-sync: P01 「お前たちがどれだけ望もうと、」; over P13 strike 「あのトロフィーは俺がもらう！」; action never pauses.
ENVIRONMENT: White pitch, charcoal lines, black steel goal, white net; overhead softbox.
RHYTHM + ESCALATION: Cuts accelerate through five duels and keeper, flowing through inner voice into impact.

BEATS:
P01: Ground-height low full; C1 fills the left third, outside boot driving the white-black leather football screen right toward goal, orange floor wake thinning behind.
P02: Ground-height foot close-up; C1 cuts the white-black leather football outside C2's Netherlands-orange lunge, a violet flare dying at lower-left contact.
P03: Knee-height canted medium-full; C1 throws shoulders left, hips and the white-black leather football cutting screen right toward goal as C2 tilts away.
P04: Waist-height profile wide; a lateral whip follows C1 past compact C3 in Japan-blue, the white-black leather football at his instep, blue ribbon floor-bound.
P05: Ball-height tight wide; camera races beside the white-black leather football through lean C4's Argentina sky-blue-and-white stance, C1's boot left, orange arc collapsing.
P06: Overhead wide; C5 in Brazil yellow-and-green seals center, C1 sole-hooks the white-black leather football lower-right, vector breaking screen right toward goal.
P07: Eye-height compressed medium; C5's tackle crosses foreground, C1 drag-turns around the white-black leather football in the right third, violet sole flare fading.
P08: Knee-height low wide; narrow C6 in France navy-and-red fills the right third as C1 and the white-black leather football accelerate diagonally screen right toward goal.
P09: Ground-height tight full; C1 carries the white-black leather football beyond C6's toe, orange-blue shock line flashing through the gap as C6 folds.
P10: Eye-height wide; C1 and the white-black leather football occupy the open right third facing goal, C2-C6 stagger defeated across distant left.
P11: Eye-height wide; C1 and the white-black leather football fill lower left, while C7 in matte-black goalkeeper kit and white gloves sets centrally inside goal.
P12: Waist-height canted medium-wide; C1 takes a running setup touch with the white-black leather football lower left, torso coiling without slowing, C7 centered inside the right goal.
P13: Ground-height extreme low wide; at full sprint C1 strikes the white-black leather football screen right toward goal as C7 commits, four-color splinters erupting at contact.
P14: Goal-side eye-height medium wide; the white-black leather football slams center-right into the white net beyond C7's stretched gloves, snapping it back as chromatic ribbons erupt and clear.

Suno Style: Dark industrial big beat with distorted breakbeats, crushing bass, snarling synth riffs, chopped female vocal fragments, chaotic rave fills, metallic impacts, tense pauses and massive explosive drops. Fast, filthy, confrontational and cinematic, with late-90s underground club energy and modern production weight, 142 BPM
Style influence: %100

[VoxelPlot @voxelplot]

Seedance 2.0 - Advanced Workflows Series

12. Cinematics with Unreal Engine and Claude Fable 5
Fable 5 is an absolute beast model that can take control of Unreal Engine and create videogames.

We can build scenes as a little video game with physics and assets and configure a playable character to create a cinematic scene.

In this example I created a scene of an astronaut lost in space floating between asteroids.

I created a zero gravity environment and made assets like cubes and spheres representing asteroids interact in this environment.

I recorded the gameplay and later place a camera to build the cinematic.

Workflow + Prompts 👇

[Kōda @aimikoda]

Midjourney: --sref 2343335488 --profile txgf1m1 --stylize 1000 --preview

GPT Image 2 prompt for storyboard:
Create a 16:9 high-sakuga-energy hand-drawn action cinematic storyboard sheet image.
[SUBJECT]
A polished modern-minimal production board for a fast-paced sci-fi motorcycle chase scene. The sheet should communicate extreme perspective, dramatic foreshortening, and tracking camera views clearly through camera angle, shot scale, pose, object state, screen direction, and spatial geography.
[HEADER]
Design an artistic production-board header with scene-aware typography, thin readable rules, clear hierarchy, generous spacing, and restrained graphic treatment outside panel interiors only.
The header must contain exactly these two quoted lines:
"CHASE OVERDRIVE"
"High-sakuga chase sequence under extreme perspective."
[BOARD STRUCTURE]
Use AUTO layout with 14 panels.
Use compact panel headers exactly in this format: `P## / shot tag / beat name`.
Draw one panel per BEATS entry, preserve P## order, and keep each panel to one clear visual action beat.
[VISUAL STYLE]
Panel interiors are silent ultra-clean blocking thumbnails: simple open-outline silhouettes, thin medium-light graphite linework, mostly white paper, broad negative space, and only the anchors needed to read pose, object contact, screen direction, and spatial result.
Open-outline silhouettes show only outer body-mass and limb contours, with no internal anatomy, face detail, clothing detail, texture, or character-design finish. Graphite linework means outline strokes only; no graphite toning, gray wash, tonal modeling, or shaded fill.
Keep all storyboard panel interiors monochrome. Use only non-color sheet-level graphic treatment outside panel interiors. If the final video has colored energy, aura, light, wardrobe, props, creatures, or environment details, describe those storyboard elements with color-neutral shape/function wording and keep all final color and palette wording only in `prompt_video.txt`.
Effects inside panels must read as abstract attached state marks, not physical props: short trails, contact bursts, halos, dust wedges, spray arcs, smoke plumes, or impact rings tied to a visible origin.
[REFERENCES]
Image A: lead girl silhouette, monster cap headwear, hair mass only.
Image B: futuristic motorcycle silhouette, low chassis, round headlight block.
Image C: spiky pig-like creature silhouette.
Image D: horned target-eye creature silhouette.
Image E: rabbit-like metal-leg creature silhouette.
If a character reference applies to one role, mark it exclusive to that role only. If the scene has other character silhouettes, keep those other silhouettes separate and unreferenced.
[CONTINUITY]
Keep natural role identity, character/entity count, prop state, effect origin, screen direction, fixed geography, and start-to-end spatial result consistent across panels.
Preserve important carried, dropped, broken, open, blocked, glowing, damaged, wet, marked, or missing states as the same role/object continuing across panels, not as repeated extra copies.
When both referenced and unreferenced character roles exist, keep them distinct; do not transfer a referenced face, hair block, wardrobe block, or body type to any other role.
Keep physical props and abstract effects separate in wording and placement; never let an effect mark become a held, carried, dropped, collected, or standalone object unless the story requires a real object.
Each storyboard panel is one frozen instant: use one pose/state per role/object and avoid before/after wording such as `then`, `after`, `before`, `first`, `next`, `later`, or `remains ... then`.
[COUNT LOCK]
Every count-sensitive panel must preserve exactly one lead girl silhouette, one futuristic motorcycle, one spiky pig-like creature, one horned target-eye creature, and one rabbit-like metal-leg creature as one instance each, except when a creature is left behind or out of frame as noted in specific beats. Never omit or duplicate active entities.
[TEXT RULES]
Visible text: only the two quoted header lines and compact panel headers. Do not render prompt section labels, role names, entity IDs, count-lock text, notes, arrows, callouts, or any other annotations inside panels.
[CONSTRAINTS]
Avoid logos, watermarks, overlays, extra panels, insets, finished illustration, dense detail, panel color, duplicate/ghost entities, before-after copies, visible role labels, visible entity IDs, and inconsistent counts.
Avoid effect marks that resemble props, icons, labels, arrows, stickers, UI, extra characters, or floating objects.
Draw one storyboard panel per visual BEAT:
BEATS:
P01 / low wide track / chase open: Wide low-angle view. A lead girl silhouette rides a futuristic motorcycle down a straight highway lane away from the camera. Three monstrous silhouettes (a spiky pig-like creature, a horned target-eye creature, and a rabbit-like metal-leg creature) chase closely behind, running in a cluster.
P02 / low rear evidence / close lunge: Low-angle rear view close to the road. The rear tires and exhausts of the futuristic motorcycle are heavily foreshortened in the foreground. The lead girl silhouette is crouched forward, looking back over her shoulder. Just behind the rear wheel, the spiky pig-like creature lunges forward with open jaws.
P03 / canted profile / near swipe: Canted side-angle view. The futuristic motorcycle is tilted sharply in a low-crouch drift. The horned target-eye creature swipes a large claw downward, barely missing the rear wheel. A short curved speed-line follows the claw path.
P04 / extreme close / rider focus: Extreme close-up of the lead girl silhouette. Her face is in profile, showing intense concentration behind goggles, her monster cap headwear framing her head. Neon reflection-lines cover her visor.
P05 / overhead void / hazard weave: High bird's-eye view looking straight down. The futuristic motorcycle weaves in an S-curve between simple blocky highway debris. The rabbit-like metal-leg creature leaps from the right frame edge toward the lane, body extended.
P06 / ground insert / ramp launch: Ultra-low angle looking up from the road surface. The front tire of the futuristic motorcycle is raised high off a broken road edge, mid-launch. A wedge of small outline debris erupts below the tires.
P07 / low silhouette / mid-air leap: Low-angle view looking straight up. The futuristic motorcycle is suspended in the air, silhouetted against a simple grid billboard. Below the airborne motorcycle, the spiky pig-like creature leaps upward, arms outstretched and jaws snapping toward the chassis.
P08 / canted tracking / landing slide: Sharp canted side view. The futuristic motorcycle lands on the road, front forks compressed. The rabbit-like metal-leg creature runs parallel on the right, metal legs posed in a running stride, lunging forward.
P09 / subjective POV / trap block: Subjective rider view looking over the motorcycle handlebars at the road ahead. The horned target-eye creature stands in the center of the lane, crouching low in a blocking stance. A simple wall block borders the background.
P10 / tight detail / brake lock: Close-up of the motorcycle's handlebars. The lead girl silhouette's gloved hand grips a brake lever tightly. Thin tension-lines radiate from the hand. The background is simple negative space.
P11 / ground tracking / screen drift: Low ground-level view. The futuristic motorcycle slides sideways in a sharp 90-degree drift toward the road shoulder. A large fan-shaped dust cloud rises from the rear tire, partially obscuring the horned target-eye creature behind the smoke.
P12 / flat graphic / impact break: Flat side-profile view. High-contrast solid silhouette of the futuristic motorcycle crashing through a simple barricade gate. Large jagged starburst outlines radiate from the impact point, shattering the gate into outline pieces.
P13 / low depth track / escape burst: Low-angle view looking down a steep downhill highway. The futuristic motorcycle, heavily foreshortened and small in the distance, speeds away toward the horizon. A straight dust trail marks its path.
P14 / overhead void / lost pursuit: High downward view from behind a broken barrier edge. The spiky pig-like creature, horned target-eye creature, and rabbit-like metal-leg creature stand at the barrier edge, looking down. In the far distance, the motorcycle headlight is a tiny dot.

Seedance 2.0 Prompt:
Use @[storyboard ref] only as layout anchor; ignore all sketch artifacts, text, extra limbs, and icons. Read each panel as a separate shot; preserve C#/object identities and do not merge poses.

Use @[char ref] for C1 appearance/likeness, @[bike ref] for S1 bike design, @[monster1 ref] for C2 pig-beast, @[monster2 ref] for C3 horned-beast, and @[monster3 ref] for C4 rabbit-beast likeness/materials; keep characters distinct.
VISUAL STYLE: High-sakuga action anime medium, extreme perspective, hand-drawn keyframe aesthetic. Low-key neon pink/yellow lighting, cool teal shadows. Fast kinetic motion smears, brief impact frames, dramatic foreshortening.
AUDIO: Kinetic chase sound design: engine hum, scraping claws, wind whooshes, tire screeches, impact thuds, brief vacuum silence.
ENVIRONMENT: Dark futuristic highway, concrete barriers, scattered debris, steel scaffolding.
EMOTIONAL GUIDANCE: Tense desperation, high-focus adrenaline, predatory aggression.
RHYTHM + ESCALATION: Fast building tempo peaking in mid-air escape, followed by a drift, barrier smash, and distant fade-out. L3 to L5 peak, resolving in L2 tension.

BEATS:
P01: Wide low-angle tracking shot. C1 rides S1 down a dark highway away from the camera. S1's front headlight points forward down the road, away from the camera, taillight facing the lens. C2, C3, C4 chase close behind in a cluster.
P02: Low-angle tracking shot from behind S1. S1's front wheel and headlight point forward down the road, while exhausts are heavily foreshortened in the foreground. C1 looks back. C2 lunges.
P03: Canted side track. S1 tilts into a low-crouch drift, its front wheel pointing forward-left down the road. C3 swipes a claw downward, missing S1's rear wheel by inches. Dynamic teal wind smear follows.
P04: Extreme close profile. C1's face shows focus, blonde hair whipping. Pink/yellow neon reflections dance on her goggles.
P05: High bird's-eye track. S1 weaves forward (front wheel leading) between concrete barriers. C4 leaps from the right wall toward the lane center.
P06: Ultra-low angle looking up. S1's front wheel lifts off a broken edge, launching forward and upward into the air. Sparks explode from the chassis undercarriage.
P07: Low-angle look up. Airborne S1, front headlight pointing forward-right, is silhouetted against a neon billboard. C2 leaps up, jaws snapping just below S1's chassis.
P08: Canted track. S1 lands forward on the asphalt, front wheel touching down first as shocks compress. C4 runs parallel on the right, lunging to tackle C1.
P09: Subjective rider POV looking forward over S1's handlebars. The highway is blocked by C3 crouching low in the lane center, forehead eye glowing red.
P10: Tight close-up. C1's gloved hand pulls the front brake lever back. Friction smoke billows from the handlebar assembly.
P11: Low ground track. S1 drifts sideways in a sharp 90-degree slide toward the road shoulder. Smoke and tire sparks erupt, blinding C3.
P12: High-contrast split-second impact frame. S1 completes the slide, crashing head-on (front wheel first) through the metal gate. Frame flashes into flat black-and-white.
P13: Low-angle track looking downhill. S1 accelerates forward down the steep road, its pink taillight fading into the distant valley.
P14: High downward angle from behind the broken gate. C2, C3, and C4 stand at the edge looking down in defeat as S1's headlight disappears into the city.

[Kay @justsosooknow]

现在红果短剧等各类短视频平台有多挣钱我就不说了，今天给大家介绍一款可以一站做短剧的Git仓库。

一个不会画画不会动画的人，靠一个开源工具做出了460万粉的AI动画频道，104条视频，最高一条5300万播放。这个工具就是Toonflow。

Toonflow是开源的AI短剧创作工作台，把小说或剧本直接变成动画短剧。集成了AI编剧、智能分镜、角色生成、视频合成，全流程AI化。支持7种语言界面，从写故事到出片，一个工具全搞定。

安装教程：
1. 去GitHub Releases页面下载安装包，Windows/Mac/Linux都有
2. macOS用户装完后去设置→隐私与安全性里放行，否则可能打不开
3. 启动后浏览器打开，首次登录账号admin密码admin123
4. 准备好LLM接口地址和视频生成服务接口，支持Sora、豆包视频等
5. Docker部署：git clone后yarn docker:local一键构建启动，访问localhost:10588/web/index.html

常见问题：
Mac打不开→隐私与安全性里点"仍要打开"
Docker部署后找不到页面→确认端口映射正确，访问路径是/web/index.html
需要自备AI服务接口，Toonflow本身不提供API key

https://t.co/BgUm9amP6u
想做AI短剧的可以收藏，一个工具从剧本到成片全搞定。

[simeon-sanai @Naiknelofar788]

Seedance 2.0 4K on Higgsfield

Prompt:

Scene: A seamless ultra-cinematic one-take shot starting from deep space and ending in an intimate café moment. The sequence emphasizes speed, scale, and immersive transition from cosmic to human scale.

Subject / Character: Final subject is @img1 — a young woman sitting in an open-air café named “LUMOS”, wearing denim shorts and a white shirt, casually eating a hamburger with a drink beside her.

Action Timeline (TOTAL: 15s):

0–4s (Space → Earth Approach):
Wide cinematic shot of Earth in deep space. Camera accelerates forward smoothly. Earth rapidly grows in frame. Subtle light streaks, atmospheric glow becomes visible. Motion builds gradually but feels grand.

4–8s (Atmospheric Entry → Fast Descent):
Camera pierces atmosphere with intense glow and motion blur. Clouds rush past. Continents and terrain sweep underneath at high speed. Strong sense of acceleration.

8–11s (City Dive → Street Flow):
Camera locks onto a city and dives sharply. Skyscrapers rise fast. Transition into street-level movement — fluid glide through streets, passing buildings, corners, and urban elements with dynamic motion.

11–12s (Café Target Lock):
Camera spots outdoor café with large “LUMOS” sign. Rapid but smooth deceleration begins. Focus tightens.

12–15s (Final Scene – Character):
Camera settles into a medium shot of @img1. She sits casually, eating a hamburger, drink on table. Natural motion (taking a bite, relaxed posture). Warm, calm contrast to previous high-speed sequence.

Camera:
One continuous shot, no cuts. Extreme speed ramping: slow → ultra-fast → controlled slowdown. Wide lens in space → natural cinematic lens at final shot. Smooth stabilization with slight handheld realism at the end.

Audio:
Cinematic rise from deep space ambience → intense whooshing during descent → city ambience → soft café sounds (light chatter, ambient noise).

Style:
Cinematic color grading: cool tones in space/descent → gradually warmer tones at street/café. High contrast, subtle film grain, volumetric lighting, atmospheric particles, ultra-detailed, 8K, photorealistic.

[Geek @geekbb]

pi + DeepSeek 画的，才发现这个技能不需要生图模型，是通过 LLM 将自然语言描述转为结构化 JSON → Node.js 渲染器用纯几何算法生成 SVG → 注入自包含 HTML。

https://t.co/ClTZDcWZdr