APIMODELSAPIMODELS

FLUX 3 视频提示词大全

收录 17 条 FLUX 3 Video 视频提示词,来自创作者在 X 和 Civitai 上发的帖子。每条都逐字全文收录、署名作者、链回原帖,卡片里直接播放它生成的视频。这些提示词都不是我们写的;作者也不一定是在 FLUX 3 Max 上跑的,FLUX 3 Max 是 apimodels.app 提供的那个 FLUX 3 Video 版本。

最近更新: · FLUX 3 Max API 价格与在线体验 · API 文档

一句话摘要

FLUX 3 Video 是 Black Forest Labs 的视频模型家族;apimodels.app 上的 FLUX 3 Max 是它经过后训练、专门压低延迟的一个版本,提示词遵循和画面质量都调得更好。一个模型名 flux-3-max 有三种用法,按你传的东西自动选:只给提示词是文生视频;给首帧(可以再加一张尾帧)是图生视频;给最多 10 张钉在具体时间点上的图,是关键帧生视频。时长 5 到 20 秒,720p 或 1080p,默认自带同步音轨;720p $0.221/秒、1080p $0.377/秒,失败不扣费。我们自己测过,720p 5 秒的片子提交后 9 秒就拿到了。它不收参考视频、参考音频,关键帧之外也只收一张图。本页收录 17 条别人公开发布的 FLUX 3 Video 提示词,附各自生成的视频。它们不全是在 FLUX 3 Max 上做的,同一条提示词放到 flux-3-max 上,结果可能不一样。

出处:Black Forest Labs FLUX 3(模型方官方站点)

悬停预览 · 点击有声

分屏女子与狗

Split-screen video. Two equal vertical halves. Both halves show the SAME event, at the SAME time, frame-synchronized, filmed by two different cameras. SCENE: A quiet street, daytime. A long tall hedge wall runs along the sidewalk, too tall to see over. There is one narrow gap in the hedge ahead. A woman in a red jacket walks alone along the sidewalk on the near side of the hedge. On the FAR side of the hedge is an open grass field, where a large golden dog runs. LEFT HALF — CAMERA A: ground-level tracking shot on the sidewalk, following the woman from behind at shoulder height. IMPORTANT: from this camera, only the woman, the sidewalk, and the hedge wall are visible. The field, the dog, and anything behind the hedge are NEVER visible in the left half. The street looks calm and empty. RIGHT HALF — CAMERA B: aerial top-down drone shot, directly above, moving with the woman. This view shows BOTH sides of the hedge at once: the woman on the sidewalk, the hedge as a thin green line, and the golden dog sprinting across the field on the other side, on a converging path toward the gap in the hedge. ACTION — identical timing in both halves: 0–8s: The woman walks calmly. LEFT: peaceful, nothing unusual. RIGHT: the dog races closer and closer to the gap, its path and the woman's path clearly about to meet. 8–11s: The woman reaches the gap. The dog bursts through it. LEFT: the dog appears suddenly from nowhere, a total surprise; the woman flinches. RIGHT: the meeting looks perfectly predictable, two paths joining. 11–15s: The dog jumps up joyfully; it is her own dog greeting her. She laughs, kneels, and hugs it. Both halves show this ending. RULES: Same woman, same dog, same timing in both halves. The dog is visible ONLY in the right half until 8s. No cuts, no other people, no cars.

来自 @umesh_ai看原帖
悬停预览 · 点击有声

安纳托利亚乡村录像

Amateur home-video footage shot on an old consumer handycam camcorder in a small traditional village in rural Central Anatolia, Turkey, on a sunny late afternoon. Five separate short scenes with hard cuts between them, as if a villager's relative is filming everyone with a handheld camcorder. Amateur camera style throughout: shaky handheld movement, sudden clumsy zoom-ins and zoom-outs, autofocus hunting and brief de-focus moments that snap back into focus, slight overexposure from the sun, slightly soft camcorder image quality, imperfect framing. Setting: stone and mud-brick houses, wooden doors, dusty lanes, a village coffeehouse, a tandır oven, chickens and a few sheep, poplar trees and golden hills in the background. Natural, authentic, humorous documentary feel, photorealistic. Scene 1 (0-4s): An old mustached man in a flat cap sitting on a wooden bench in front of the coffeehouse with a tea glass, the camera zooms in clumsily on him, he looks into the camera and says in Turkish with a strong, warm Anatolian village accent: "Ula gardaş, bu yeni yapay zeka videoyu bi göz açıp kapayınca yapıveriyo!" Scene 2 (4-8s): Cut to a middle-aged woman in a floral headscarf kneading dough next to a tandır oven, the focus briefly blurs then sharpens, she laughs and says: "Ben daha hamuru yoğurmadan, o videoyu bitirivemiş!" Scene 3 (8-12s): Cut to a young shepherd with a staff standing beside a few sheep on a dusty slope, the camera zooms out shakily, he says: "Goyunları ağıla sokana gadar on tane video çıkarıveriyo!" Scene 4 (12-16s): Cut to an elderly grandmother in a white headscarf sitting on her doorstep by a wooden door, the camera zooms in fast and refocuses, she raises her hands and says: "Bizim zamanımızda böyle hız mı vardı, maşallah!" Scene 5 (16-20s): Cut to a young man leaning on a red tractor, he grins, points at the camera and says: "Eee, ne duruyonuz? Siz de deneyivirin!" The camera wobbles and zooms in on his grin. Audio: clear Turkish speech with rural Anatolian dialect for each speaker, slight camcorder microphone handling noise, ambient village sounds, roosters, distant sheep bells, light wind. No subtitles, no on-screen text, no timestamp overlay.

来自 @ozansihay看原帖
悬停预览 · 点击有声

白色卡车沙漠飞跃

20s, 720p, 21:9. trophy truck in white jumps through various desert jumps, anamorphic lens, fast paced cut.

来自 @junwatu看原帖
悬停预览 · 点击有声

夜森林黄雨靴

Prompt (I2V) : Create a dark, photorealistic cinematic horror sequence using the reference image as the opening shot. The mood should be beautiful, eerie, mysterious, and quietly disturbing — grounded, naturalistic horror rather than gothic or fantasy. The scene should feel like something impossible is happening in a real forest at night. OPENING IMAGE Begin with two glossy yellow rain boots standing alone on a muddy, puddled forest path at night. The path is framed by wet grass and swallowed by darkness in the distance. The lighting is low, cold, and moody, with deep blue-black tones, faint reflections in the puddles, and just enough light to catch the shine of the boots and the wet ground. The camera starts low and close to the ground, positioned behind the boots. Everything is still. Rain falls softly. Tiny ripples move across the puddles. The forest is silent. After a brief pause, one boot slowly lifts on its own. Then the other follows. The boots begin walking away from the camera, moving forward down the path with their fronts facing into the darkness ahead, as if an invisible woman is already wearing them. The camera watches from behind as they take slow, deliberate steps farther away. Each footstep lands with natural weight, softly splashing through shallow water and pressing into the mud. The movement should feel graceful, realistic, and unsettling. THE FORMING FIGURE As the boots continue walking away from the camera, a female figure begins to form above them. She must already be clothed from the start of the transformation. A dark dress begins appearing immediately above the boots as a vague moving silhouette, already covering the form as it develops. The clothing becomes gradually more solid and detailed while she walks: wet fabric, soft folds, long sleeves, a simple natural drape. It should feel understated, realistic, and elegant — not gothic, ornate, or theatrical. As the camera slowly rises, more of the woman becomes visible within the already-formed clothing: her body, shoulders, neck, hair, and finally her face. The transformation should feel seamless and uncanny, as if she is condensing out of shadow, mist, and rain while continuing to walk. No glowing effects, no fantasy particles, no stylized magic. CAMERA MOVEMENT The sequence should play as one smooth, unbroken cinematic shot. At first, the camera follows from behind at a low angle, watching the boots walk away down the path. Then it slowly cranes upward, rising above and behind her while continuing to track her movement. As it ascends, it reveals more of her fully formed figure, with the yellow boots still visible beneath the hem of the dress. By the time the camera reaches a high overhead angle, the woman is fully formed and still walking away into the darkness. THE TURN From above, the camera slowly pushes in. As it gets closer, the woman seems to sense it. She slows slightly, then turns her head first, then her shoulders, and finally enough of her body to look back up toward the camera. The turn should feel calm, graceful, and deliberate. When she looks up, we see her face clearly. She is naturally beautiful and fully human, with wet hair, pale skin touched by cold night light, and calm, realistic features. Nothing grotesque or monstrous. But there is something faintly unnerving in her stillness and in the way she looks directly at the camera. As the camera continues to zoom in from above, she gives a soft, beautiful smile. The smile should look normal and human — subtle, elegant, and controlled — but also slightly unsettling the longer we look at it. It should feel warm on the surface, yet unreadable underneath. Her eyes remain fixed on the camera, calm and knowing. FINAL FEELING End on the image of her looking upward, smiling softly in the middle of the wet forest path at night. The yellow boots remain visible as an important anchor to the original image. The final feeling should be mysterious, beautiful, and quietly unnerving. SOUND AND MUSIC Use sparse atmospheric sound: soft rain, distant wind, subtle forest ambience, and the wet sound of rubber boots stepping into mud. As the boots begin moving on their own, introduce a very low dark drone. Add faint minimal strings, distant reversed textures, and soft eerie tonal layers. Keep it elegant and restrained. When she turns and looks up, reduce the sound slightly to create a sense of suspended stillness. As the camera zooms in and she smiles, let a delicate but unsettling musical note emerge beneath the rain. No dialogue. No scream. No harsh stinger. End on mystery. EMOTIONAL ARC stillness → curiosity → unease → fascination → quiet dread

来自 @umesh_ai看原帖
悬停预览 · 点击有声

咖啡馆窗边男子

Photorealistic 16mm film shot of a man in his early 40s sitting alone at a small café table by a window. Natural window light, muted warm colors with slight green-magenta shift. He wears a plain button-up shirt. He takes a slow sip of coffee, then glances outside with a thoughtful, restrained expression, subtle tension in the brow and a quiet look in his eyes. Realistic skin detail, natural micro-expressions, gentle breathing. Authentic 16mm film grain, soft contrast, slight gate weave, organic imperfections, cinematic shallow depth of field, natural motion.

来自 @zeng_wt看原帖
悬停预览 · 点击有声

慢镜头水球爆裂

Prompt 👇 One continuous 10-second shot. Photorealistic ultra slow motion macro, 16:9, against a plain dark blue background. A red water balloon hangs from a thin string in the center of the frame. The camera stays still. 0-3s: The balloon wobbles gently on its string. 3-5s: A hand enters from frame-right holding a thin silver needle between two fingers and touches the needle tip to the balloon. 5-9s: The rubber tears and snaps away in an instant, leaving a perfect balloon-shaped blob of water floating in mid-air, which slowly loses its shape and bursts apart into glittering droplets. 9-10s: The droplets fall out of frame. End state: the empty string swings in front of the blue background. Audio: a soft pop, then a slowed-down deep water splash and droplets pattering. Only natural sound effects, nobody speaks.

来自 @openclipsai看原帖
悬停预览 · 点击有声

竹林中的武士

日本の江戸時代。
竹林の中に一人の武士。笹の葉が常に舞い落ちている。 整った武士の服装で、ダークブルーが基調。 武士は男性で、年齢は30歳。無精髭あり。中肉中背。 落ち着いた表情。 カメラは
低い位置からリフトして人の目の高さ。
ゆっくりオービットパン。
 0-2秒:直立して呼吸を整える。 2-4秒:刀を鞘からまっすぐ抜き、両手に握り構える。八相の構え。
 4-7秒:北辰一刀流の殺陣の演舞。素早い動き。 最後の3秒:刀を時代劇と同じ動作で丁寧に鞘に納め、呼吸を整える。カメラが武士に近寄る。

 SONY VENICE 35mmレンズ。シネマティック。 
武士以外のキャラクターは存在しない。

来自 @Tentamasoba1016看原帖
悬停预览 · 点击有声

发光眼头盔人物

subject_definitions: <Picture 1> is {'Physical Actions': 'The subject is standing still, with no discernible movement.', 'Body Language': "The subject's posture is upright and commanding, with hands slightly bent at the elbows, suggesting readiness or alertness.", 'Physical Interactions': 'The subject is not interacting with any other objects or characters in the image.', 'Facial Expressions': "The subject's face is covered by a helmet with glowing orange eyes, which convey a sense of intensity or focus."} detailed_description: [Shot 1] 0-2s - The subject stands still, adopting the posture shown in <Picture 1>, with hands slightly bent at the elbows, conveying readiness and alertness. [Shot 2] 2-5s - The subject remains in the same position, maintaining the same posture and readiness, with the glowing orange eyes within the helmet continuing to convey intensity and focus. overall_soundscape: Quiet indoor room tone with subtle fabric rustling.

来自 4lth3rego看原帖
悬停预览 · 点击有声

女子吃巧克力特写

1]## HYPERREALISTIC CHOCOLATE UGC — 10 SECONDS **REFERENCE:** Use the provided image #Image1 as the exact first frame. Preserve the woman's facial identity, blue eyes, blonde hair, natural skin texture, white clothing, jewelry, dark chocolate, white ceramic bowl, blue woven chair, beige cushion and original warm indoor lighting. **CREATIVE DIRECTION** An extremely believable, spontaneous smartphone video of a young woman eating chocolate and casually reacting to its taste. Feels like a friend is recording her from close range. Playful, slightly messy, intimate and genuinely unscripted. The camera is handheld and noticeably alive, never artificially locked. ### ACTION, EXPRESSIONS & TIMING **0:00–0:02 — THE BITE** Beginning from the exact reference pose, she bites into the chocolate naturally. Her front teeth make contact, her jaw closes with believable resistance, and a small piece breaks away. The remaining chocolate stays between her fingers. Her eyes briefly narrow as she bites. She withdraws the chocolate a little from her lips, with a tiny instinctive downward glance. The handheld camera drifts slightly left and downward, then gently corrects. A couple of irregular micro-shakes from the person holding the phone. **0:02–0:04 — THE REAL REACTION** She chews naturally with closed lips, approximately two or three uneven chewing movements. Her eyes wander briefly away from the camera, then return. A subtle eyebrow lift suggests she likes the taste. A tiny smile appears while she's still chewing, slightly raising one cheek. She leans backward a few centimeters into the chair. The bowl shifts naturally with her hand. The camera makes a small spontaneous upward correction, briefly softening focus before settling again. **0:04–0:06 — THE MOMENT BEFORE SPEAKING** She finishes chewing and swallows naturally. A very small amused exhale escapes through her nose. She glances down at the chocolate, then back toward the person filming. Her lips part as she starts talking, with a slight smile that she doesn't deliberately hold. **0:06–0:10 — THE DIALOGUE** She talks naturally while looking mostly at the camera, occasionally glancing down at the chocolate. Her eyebrows shift subtly, her head tilts just a fraction, and her mouth forms an asymmetric smile. A tiny, genuine breathy laugh interrupts her sentence. She gently raises the chocolate a few centimeters as if considering another bite, but does not actually bite again. ### EXACT DIALOGUE "Wait... that's actually so good. Like, I need another piece." ### VOICE PERFORMANCE Extremely natural young adult American female voice. Soft-medium volume, slightly low and relaxed, with realistic vocal texture and casual American pronunciation. - **"Wait..."** Said quietly with mild surprise, almost under her breath. - **"That's actually so good."** Warm, spontaneous, with a tiny smile audible in the voice. Slight extra emphasis on "so." - **"Like,"** A quick, casual filler with a small breathy laugh. - **"I need another piece."** Slightly faster, playful and understated, as if she's casually admitting something funny. Not every word should be perfectly articulated. Use naturally connected syllables, tiny breaths, irregular pauses, subtle pitch changes and unpolished speech timing. She must finish swallowing before speaking. No dialogue with an unrealistically full mouth. ### HANDHELD CAMERA — IMPORTANT The smartphone is physically handheld by another person seated or standing immediately in front of her. The framing must feel organically unstable throughout: - Small unpredictable horizontal and vertical movements from natural hand tremor. - Occasional slightly stronger shake, particularly around 0:02 and 0:07. - Very gentle forward/backward drift as the camera holder shifts their weight. - Imperfect recentering when she leans back. - Subtle rolling-shutter behavior and momentary autofocus breathing. - Natural changes in perspective and background parallax. - A tiny accidental clockwise tilt, naturally corrected without a mechanical movement. **Never make the camera perfectly static.** However, shaking must stay within believable smartphone-handling limits, not violent or deliberately cinematic. ### PHYSICAL REALISM Natural jaw mechanics, realistic chocolate fracture, subtle swallowing, believable lip compression, tiny eye-focus changes, occasional irregular blinking, visible skin pores, realistic asymmetrical facial muscle movement, and slight hair displacement as she shifts. Accurate finger contact with the chocolate and stable grip on the bowl. Both objects have consistent physical weight. Preserve realistic shadows and occlusions. ### SOUND Intimate smartphone microphone recording. Very faint indoor room ambience, subtle chocolate-bite sound, small natural chewing noises, realistic breath, slight clothing rustle and a soft spontaneous laugh. The recording should sound like an ordinary phone video, not studio voiceover. No background music. ### STRICT NEGATIVES No beauty filters, no robotic facial movements, no artificial smiles, no exaggerated enjoyment, no perfectly synchronized blinking, no unnatural chewing, no chocolate melting or morphing, no floating crumbs, no distorted hands, no disappearing fingers, no bowl deformation, no frozen background, no smooth cinematic gimbal movement, no constant mechanical camera shake, no artificial voice delivery, no subtitles, no cuts. **FINAL RESULT:** A genuinely convincing 10-second handheld chocolate UGC video. Slightly shaky, naturally imperfect, dynamically framed, with believable eating mechanics, subtle facial reactions and spontaneous conversational dialogue. It should feel like a real casual moment accidentally captured on someone's phone. 2]## HYPERREALISTIC STREET-STYLE UGC — 10 SECONDS **FIRST FRAME & VISUAL CONTINUITY** Use the provided image #Image1 as the exact opening frame. Preserve the man's facial identity, natural Afro-textured hair, facial hair, skin texture, silver earring, gold necklaces, grey graphic T-shirt, hand anatomy, red takeaway cup, striped straw, and printed branding. Preserve the original sunny urban street, architecture, vehicles, and natural daylight. The video must feel like a spontaneous smartphone recording of a real guy reacting to a drink on a sunny afternoon. ### PERFORMANCE & MICRO-EXPRESSIONS **0:00–0:02 — THE SIP** He continues sipping through the straw exactly as shown in the first frame. A tiny natural hollowing of the cheeks, slight jaw tension, subtle throat movement. His eyes maintain casual contact with the camera, slightly narrowed from the sunlight. He finishes sipping, releases the straw naturally with his lips, and swallows. No exaggerated reaction. **0:02–0:04 — THE REACTION** His eyes briefly move toward the cup. One eyebrow rises almost imperceptibly. His mouth forms a tiny surprised half-smile. A very quiet nasal exhale, almost a suppressed laugh. He lowers the cup approximately 5–8 centimeters while keeping it comfortably in his hand. He starts speaking as though genuinely surprised by how much he likes the drink. **0:04–0:08 — THE CONVERSATION** He looks back into the camera. His head makes a small, involuntary nod while speaking. One corner of his mouth lifts naturally. Slight eyebrow movement and subtle changes in eye focus accompany his words. His facial muscles remain relaxed. No dramatic expressions. **0:08–0:10 — THE ENDING** He finishes his sentence with a small amused smile, glances briefly down at the drink, then returns his gaze to the camera. A tiny satisfied nod. The expression fades naturally rather than freezing into a perfect smile. ### EXACT SPOKEN DIALOGUE "Okay, wait... that's actually so good. I thought it'd be way too sweet, but nah. I could drink this all day." ### VOICE & SPEECH DIRECTION Authentic young adult American male voice, medium-low register, slightly textured, warm, relaxed and conversational. The delivery must sound like completely unscripted everyday speech. - **"Okay, wait..."** Slightly surprised, spoken quietly after swallowing. A tiny hesitation between the words. - **"That's actually so good."** Genuine understated appreciation. Slight natural pitch rise on "actually," with an almost involuntary smile. - **"I thought it'd be way too sweet"** Faster, casually connected speech, as though explaining his initial assumption. - **"But nah."** Short, lower-pitched, lightly amused, with a tiny head movement. - **"I could drink this all day."** Relaxed and confident, ending with a small smile rather than an advertising-style punchline. Realistic American conversational pronunciation, natural contractions, slight variations in vocal intensity, authentic breaths, and subtle vocal texture. Avoid rhythmic, perfectly enunciated scripted speech. ### PHYSICAL & AUDIO REALISM Precise lip synchronization with convincing jaw, tongue, cheek and neck-muscle movement. Natural swallowing before speech. Realistic facial asymmetry, tiny irregular blinks, sunlight-induced squinting, and spontaneous eye movements. The cup has realistic weight and inertia. Fingers maintain believable contact and grip pressure. The straw moves only as physically caused by his lips or the movement of the cup. Necklaces respond subtly to body movement. Use authentic nearby street ambience: faint traffic, distant pedestrians and a soft urban background. His voice is clear and close to the phone microphone, with realistic outdoor acoustic characteristics. ### CAMERA Single continuous handheld smartphone recording. Maintain the original framing, perspective and lighting. Very subtle natural handheld instability, without deliberate pans, zooms, cinematic camera moves, or cuts. Preserve the slightly harsh direct sunlight, authentic skin highlights, real shadows, fabric texture and believable background depth. ### STRICT NEGATIVES No robotic lip-sync, exaggerated eyebrow raising, forced laughter, theatrical expressions, unnecessary body swaying, random hand gestures, artificial smooth skin, facial identity changes, altered cup branding, distorted fingers, floating straw, unnatural teeth, scripted influencer enthusiasm, music, subtitles, or transitions. **FINAL RESULT:** A convincing, completely natural street-style UGC moment, where a real person casually takes a sip, discovers he likes the drink, and tells the camera with subtle, spontaneous human reactions. dont keep the camers stuck.. move it or shake it naturally somtimes 3]**HYPER-REALISTIC UGC — LITTLE GIRL OUTFIT CHECK** **Duration:** 12 seconds. Single continuous handheld smartphone recording. Authentic home-video realism. **FIRST FRAME:** Use the supplied image exactly. Preserve the child's facial identity, natural skin texture, hairstyle, blue and white hair clips, orange-tinted sunglasses, pink bear dress, body proportions, handbag, wooden flooring and surrounding environment. **PERFORMANCE:** A young girl is being casually filmed by a parent holding a smartphone slightly above her eye level. She initially looks down with the calm, slightly serious expression visible in the reference. After hearing the person filming, she gradually lifts her chin, looks toward the phone through her sunglasses and gives a tiny, proud smile. She speaks with natural childhood enthusiasm, imperfect timing and small pauses. She briefly looks at her dress, lightly sways her handbag, shifts her weight between her feet and gives a tiny shoulder shrug. Her confidence feels spontaneous and adorable, not rehearsed. **DIALOGUE — Natural young American girl's voice:** "Okay, so... these are my favorite sunglasses. And look at my dress! It has a little bear on it. I picked it myself!" **VOICE & FACIAL PERFORMANCE:** Authentic young child's voice, natural American English pronunciation. Slightly soft beginning, playful emphasis on "favorite sunglasses," a tiny excited pitch rise at "look at my dress," and a proud, cheeky smile after "myself." Natural pauses, little breaths, subtle mouth movements, believable jaw articulation and precise lip synchronization. No adult-sounding voice, artificial baby voice, excessive cuteness or exaggerated expressions. **CAMERA:** Parent-recorded vertical smartphone footage. Maintain the original high-angle perspective. Gentle handheld shaking, imperfect framing, tiny accidental camera adjustments and slight reframing as she lifts her head. Never perfectly stabilized or cinematic. **REALISM:** Natural indoor daylight. Small fabric movements, realistic hair physics, subtle blinking behind tinted lenses, natural balance adjustments and accurate handbag movement. Very faint indoor room ambience. No background music, captions, beauty filters, cuts, artificial zooms or dramatic acting. **GOAL:** A genuinely candid video of a little girl proudly showing off her outfit to a family member. It should feel like a real spontaneous childhood moment, not an AI-generated advertisement.

来自 @RareParkStudios看原帖
悬停预览 · 点击有声

玻璃瓶逆放重组

In a surreal but physically clean reverse-motion shot, scattered shards of a green glass bottle slide and leap from a concrete floor, reassembling into one intact bottle standing upright. Every shard contributes to the final bottle; dust lifts back into cracks. Static camera, reversed glass sounds, no hands.

来自 fal.ai看原帖
悬停预览 · 点击有声

火车窗边卷发飞扬

Profile close-up of a woman with long curly hair standing beside an open train window. A passing tunnel briefly blocks the light as airflow pushes individual curls backward; the hair settles naturally when the train slows. Stable face, realistic strand motion, rhythmic rail audio.

来自 fal.ai看原帖
悬停预览 · 点击有声

老人大笑特写

tight shot of an elderly man's face laughing hard at a joke, deep wrinkles creasing, eyes squeezing shut, then wiping a tear from the corner of his eye

来自 fal.ai看原帖
悬停预览 · 点击有声

钢琴手指快速演奏

a pianist's hands playing a fast passage across the keys, fingers crossing over each other, both hands visible and independent

来自 fal.ai看原帖
悬停预览 · 点击有声

曼谷街头炒面摊

street food vendor in bangkok flipping noodles in a wok over roaring flames at night

来自 fal.ai看原帖
悬停预览 · 点击有声

狗狗探头兜风

a golden retriever sticking its head out a car window on the highway, ears flapping in the wind

来自 fal.ai看原帖
悬停预览 · 点击有声

定格动画纸片水手

Handmade stop-motion scene of a tiny paper sailor raising a blue fabric sail on a cardboard boat. The cloth catches a gust, billows to the right, and pulls the boat forward across a rippled cellophane sea. Visible handcrafted texture, consistent puppet proportions, papery rustles and gentle wooden creaks.

来自 fal.ai看原帖
悬停预览 · 点击有声

猫咪走过窗台

Cat walks across the windowsill, tail swaying. It pauses to look outside, then continues and hops down. Soft paw steps and a gentle meow.

来自 fal.ai看原帖

怎么写好 FLUX 3 视频提示词

  1. 1用定时关键帧把节拍钉住. 本页有好几条提示词在文字里给动作排时间,0 到 2 秒做一个动作,2 到 4 秒做下一个。在 FLUX 3 Max 上,你还可以把图片钉到这些时间点上:最多传 10 个关键帧,每个写上秒数(time)或者 24 fps 下的帧号(frame_index),提示词只管描述两张图之间怎么动。
  2. 2发请求前先把帧号算好. frame_index 按 24 fps 数帧,范围是 0 到时长 × 24,同一个帧号不能用两次;写 time(秒)的话我们会帮你换算。4 秒处的关键帧是第 96 帧,10 秒片子的最后一帧是第 240 帧。原帖要是按别的帧率编号,先换成秒再换回来。
  3. 3只在乎开头结尾,就用首尾帧. 如果你只想定住镜头从哪开始、落在哪,传 first_frame_url 和 last_frame_url,提示词专心写中间这段怎么走。关键帧不能和首帧、尾帧放进同一个请求,二选一。
  4. 4说清楚要什么声音,不要就关掉. 声音默认会生成,提示词本来就该写你想听到的声音,和主体、动作、镜头、光线一样重要。写明环境音,台词放进引号。如果你打算自己配乐,传 generate_audio: false,别在提示词里写「没有声音」。
  5. 5时长、分辨率、画幅放进请求参数. 本页有些提示词第一行就把这些设置写上了。在 API 上它们是参数:duration(5 到 20 的整数秒)、resolution(720p 或 1080p)、aspect_ratio(auto、21:9、2:1、16:9、4:3、1:1、3:4、9:16)。带首帧时 auto 跟着图片走,只有文字时是 16:9。
  6. 6多张参考图的提示词,改成关键帧或首帧. 有些帖子要靠好几张参考图或一段参考视频。FLUX 3 Max 在关键帧之外只收一张图,参考视频、参考音频或多出来的图都会返回 400。把参考图改成定时关键帧,或者挑最重要的那张当首帧,其余的用文字描述。
  7. 7打草稿用 reasoning_level low. reasoning_level 默认是 high,low 更快。先用 low、720p、5 秒试提示词($1.105),定下来的那条再用默认的 high 或 1080p 出。

提示词公式

设置放进请求参数(5 到 20 秒、720p 或 1080p、画幅)+ 主体和场景 + 按秒排好的动作 + 镜头运动 + 光线和胶片质感 + 声音和引号里的台词 + 用首尾帧,或最多 10 个定时关键帧钉住画面

例(本库收录的原文,原样引用,来自 fal.ai 的 FLUX 3 页面):In a surreal but physically clean reverse-motion shot, scattered shards of a green glass bottle slide and leap from a concrete floor, reassembling into one intact bottle standing upright. Every shard contributes to the final bottle; dust lifts back into cracks. Static camera, reversed glass sounds, no hands.

常见问题

FLUX 3 Max 是什么?和 FLUX 3 Video 什么关系?

FLUX 3 Max 是 FLUX 3 Video(Black Forest Labs 的视频模型家族)经过后训练、专门压低延迟的一个版本,提示词遵循和画面质量都调得更好。在 apimodels.app 上它的模型名是 flux-3-max:文生视频、首尾帧图生视频、最多 10 个定时关键帧,5 到 20 秒,720p 或 1080p,默认带同步音轨。厂商自己测的是 10 秒视频约 11 秒出片;我们自己测,720p 5 秒的片子提交后 9 秒就拿到了。

这些提示词是谁写的?

是 X 和 Civitai 上的创作者写的,我们一个字都没写、也没改。每张卡片都逐字全文引用提示词、写明作者、链回原帖,视频直接从作者发布的地方播放。提示词和视频归原作者所有;我们自己写的说明和整理的数据按 CC BY 4.0 授权。要更正署名或撤下某一条,发邮件给我们,或者在 GitHub 仓库提 issue。

这些提示词是在 FLUX 3 Max 上跑的吗?

不一定:作者们用的是自己手边能用的 FLUX 3 Video,只有一部分说明了用的是 FLUX 3 Max。FLUX 3 Max 是后训练过的版本,同一条提示词在 flux-3-max 上跑出来,可能和卡片上的视频不一样。卡片上的视频是作者当时拿到的结果,不代表 API 一定能给你同样的东西。

在哪能跑这些提示词?多少钱?

在 apimodels.app 上,模型名 flux-3-max,接口 POST /v1/video/generations,按你请求的秒数计费。720p $0.221/秒:5 秒 $1.105,10 秒 $2.21,20 秒 $4.42;1080p $0.377/秒:5 秒 $1.885,10 秒 $3.77。只有成功的才扣费,失败的会退回。

本页的提示词放到 API 上失败,为什么?

不合规则的请求在创建时就返回 400,不会建任务,也不扣费。对 FLUX 3 Max 来说包括:传了参考视频、参考音频,或者关键帧之外传了不止一张图;关键帧和首帧、尾帧混用;关键帧超过 10 个,或者 frame_index 超出 0 到时长 × 24 的范围、有重复;时长不在 5 到 20 的整数秒之内。上游还有内容安全审核,被拒的生成不扣费。

FLUX 3 的图片提示词能拿来跑 FLUX 3 Max 吗?

能跑,但图片提示词只描述一个静止的瞬间,动作、镜头和声音就只能交给模型自己猜。更好的做法是先有一张定稿的图,把它当 first_frame_url 传进去,再在提示词里写它怎么动。FLUX 3 的图片提示词在本站另有一页。

这些提示词和视频能商用吗?

提示词和视频归原作者所有,商用前先看原帖怎么说。你自己生成的视频,按你用来生成的那个服务的条款执行。

在 apimodels.app 上跑这些提示词

FLUX 3 Max 已经上线,和另外 180+ 个图像、视频、语音、大模型共用一个 API、一个 key,按量计费、注册送 $0.10 试用额度。

其它提示词库