跳到主要内容
Kyeo AI
AI 视频模型
Google
Gemini Omni

Gemini Omni Video

把文字构想、参考图或视频片段作为短片的起点。先确定主体与镜头变化，再根据输入方式选择参数；视频输入时，所选时长不决定成片时长。

使用这个模型查看积分价格返回模型广场
输入方式
文字、参考图或视频片段，可组合图片与视频
分辨率
720p / 1080p / 4k
预扣积分
45–180
快速了解
  • 分辨率可选 720p、1080p 或 4K。
  • 无视频输入时最多可用 7 张参考图；加入视频后最多可同时使用 5 张。
  • 视频输入可截取源片段，输出时长由模型决定。

相关查询: Gemini Omni Video 视频生成 · Gemini Omni Video 积分预扣 · Gemini Omni Video 可调参数

示例

Cinematic scene

Generate a 30-second horizontal 16:9 realistic fitness lifestyle influencer vlog from the provided first frame. Style:
Realistic daily fitness influencer vlog, natural indoor lighting, casual social media filming style, warm private gym atmosphere. The video should feel like a smooth everyday workout vlog filmed by an influencer, with natural movement, clean pacing, and coherent camera transitions. Character:
One main character only: Maya, a young adult female fitness creator with brown hair in a high ponytail, wearing a taupe long-sleeve athletic top, matching leggings, white socks, and light grey training shoes. Keep her face, hairstyle, outfit, body type, and identity consistent throughout the video. Scene:
A quiet private indoor gym with dark grey walls, black rubber floor, warm ceiling lights, a workout bench, clear plastic water bottle, small towel, gym bag, small dumbbells, mirror area, chair, and wall-mounted drinking fountain. Keep the environment consistent. Objects should stay in believable positions. Camera style:
Use two smooth camera perspectives:
1. Selfie camera perspective: Maya holds the camera or phone at arm’s length, showing her face and upper body.
2. Third-person camera perspective: stable camera placed near the bench or mirror, showing her full body and workout actions. Camera transitions:
Transitions must feel smooth and intentional, like a polished influencer vlog edit.
Use gentle cuts, soft handheld reframing, slow push-ins, and small pans.
Each shot should last long enough to feel natural.
Avoid rapid cutting, sudden jumps, extreme zooms, spinning camera, or chaotic handheld movement. Action flow:
All actions should connect naturally:
Maya places her bag, adjusts her hair, introduces the workout, picks up a dumbbell, does a short set, rests, drinks water, stretches, checks the mirror, and ends with a relaxed smile.
Movements should be casual, continuous, and realistic, not stiff or overposed. Action timing:
0s - 4s:
Third-person view. Maya stands beside the workout bench and calmly places her gym bag or towel on the bench. She adjusts the position of the water bottle slightly. Camera is stable, medium-wide, showing the bench and gym room. 4s - 7s:
Smooth cut to selfie camera. Maya holds the camera at arm’s length, looks into the lens, gives a small natural smile, and lightly adjusts her ponytail. The camera moves gently with her hand but remains stable and readable. 7s - 10s:
Selfie camera slowly tilts down and transitions to a third-person bench view. Maya lowers the camera or the edit cuts softly to the bench angle. She reaches for a small dumbbell on the floor. The movement should feel connected, not abrupt. 10s - 15s:
Third-person workout shot. Maya places one hand on the bench and performs a short controlled dumbbell row set, about 3 to 4 smooth repetitions. Keep the motion steady and anatomically correct. Hair and fabric move subtly with each rep. 15s - 17s:
She finishes the set, gently places the dumbbell back on the floor, and sits on the bench. Camera stays stable with a slight push-in, capturing the transition from exercise to rest. 17s - 20s:
Close detail shot. Her hand reaches for the clear water bottle, opens it, and brings it up naturally. The camera follows slowly, without shaking. 20s - 22s:
Selfie camera perspective. Maya takes a small drink, exhales, and looks into the lens with a relaxed expression. Her breathing feels natural after the set. 22s - 25s:
Smooth cut to third-person view near the mirror. Maya stands, rolls her shoulders, stretches one arm across her chest, then switches sides. Camera gently pans to keep her centered. 25s - 27s:
Handheld lifestyle detail. Maya picks up the towel, wipes her hands, and checks her ponytail in the mirror. The action is small, casual, and continuous. 27s - 30s:

Cinematic scene

Use the attached character reference (Image1) as the silhouette base. Create a pastel neo-noir teen-assassin anime title sequence, cut like a music video with hard, punchy edits (no crossfades). Style the character as a flat black silhouette against solid baby-pink gradients (blush pink, bubblegum, rose quartz, cotton-candy white), rendered in bold pop-art compositions with retro shōjo-manga panel breaks, chrome-and-neon accent lines, extreme close-ups on eyes/lips/weapon, and wide skyline or rooftop establishing shots. Alternate fast, kinetic action bursts (sprint, blade draw, spin-kick, hair whip) with hard freeze-frame beats — each freeze holding just long enough to read as a "poster moment." Use motion graphics throughout: speed-lines, halftone dot shading, comic-style impact bursts ("POW"-style but in soft pastel palettes), chromatic glitch flickers, and light streaks trailing off silhouette edges on transitions. Bold typography: chunky, rounded bubble-serif or Y2K-manga-style lettering (like sticker-text or magazine cutouts) slamming into frame on each hard cut, sometimes overlapping the silhouette, sometimes punched into negative space. Title cards should feel like collectible trading-card / vinyl-poster inserts — thick outlines, drop shadows, slight jitter/shake on impact. Finish with the character frozen in full silhouette against a solid baby-pink background as the title logo stamps in with a soft chrome shimmer and a light glitter-dust settle effect. Apply a light VHS/Y2K digicam texture throughout — subtle scan lines, soft chromatic aberration, occasional analog glitch/tracking-line flicker, and faint sparkle/bloom — for a nostalgic early-2000s music-video finish (instead of film grain/dust, since there's no rugged western texture here). BGM: hard-hitting electronic / drum-heavy future bass with aggressive drops, risers, sub hits and glitch fills locked to every cut. Sneaker impacts, whooshes, glass shatters, window-slam hits and typography slams are rhythmic sound-design elements. Peak at CUT 12 and end with a cold electronic logo stinger. Premium AAA quality, anime-inspired cinematic rendering, stylish and explosive, strong graphic-design identity, consistent character design, exactly 13 cuts.

Cinematic scene

【风格】剑道Vlog蒙太奇（Kendo Dojo Vlog Montage），真实手机竖屏实拍质感（Photorealistic Handheld），8K超清，高帧率，日式道场暖调室内光，多机位卡点快剪（Multi-Angle Beat-Synced Jump Cuts），皮肤有真实毛孔和光泽、布料可见织纹和垂坠感，无AI塑料感，竖屏9:16 【时长】15秒 【场景】传统日式剑道馆（道場），包含抛光木地板练习区和防具装备存放区，始终同一间道场 【角色】主角 @图片1（全片核心人物，面部严格锁定参考图，六个镜头之间不漂移）。教练（身穿黑色短袖T恤+深藏青色袴的男性，仅在镜头1和镜头4中作为配角出现） 【转场机制】全片Vlog蒙太奇快剪风格，镜头之间全部硬切（Hard Cut / Jump Cut），每一刀踩在BGM鼓点上。明确禁止：淡入淡出、叠化、黑场、闪白、模糊、任何动画过渡。 [00:00-02:30] 镜头1：并肩挥刀·道场初亮相（Practice Floor / Medium Shot） 场景：道场练习区，下半段深色木质竖条纹护墙板，上半段米白色墙面。左侧墙上悬挂一幅竖式墨色书法条幅，右侧背景可见木格障子推拉门（门后透出柔和暖光），门旁嵌一个小型木质展示架摆着铁壶茶具。暖黄色顶灯打出柔和侧光，抛光木地板有淡淡哑光反射。 主角居画面左侧，教练在右侧稍后方。 穿着：主角身穿深藏青色刺子织剣道着（kendogi，交叉V领、胸前黑色系带打结、面料厚实有明显刺子织菱形凹凸纹理），下穿深藏青色袴（hakama，五道竖褶线条笔直、裙摆宽松垂坠到脚背），赤足踩在木地板上，右手腕戴一只浅色珠串手链。 动作：两人各持一把居合练习刀（iaito，金属刀身有冷光反射、皮绳缠柄），主角双手握刀在身前横向展示，随后双臂缓缓向上举起——刀身从水平提至头顶正上方（上段构え / Jodan-no-kamae），刀刃朝前、刀尖微向后倾。教练同步举刀。 表情：专注冷静，目光平视前方，嘴唇微抿。**硬切** [02:30-04:30] 镜头2：竹刀怼脸·装备架前亮相（Equipment Room / Low Angle Close-up） 场景切换至装备存放区。背景是整面墙的木格方形储物架，每格里塞满黑色剣道防具（深色皮革和靛蓝棉布缝制的面、小手、胴），暖黄灯光从架顶打下形成上暖下暗的层次光。左侧可见木格障子门一角，墙上另一幅书法条幅。 机位：略低角度仰拍（Low Angle），镜头在主角腰部以下向上拍。画面只有主角一人。 动作：主角双手各握一把竹刀（shinai，四片竹条编束、皮革刀锷、白色先革包头），两把竹刀交叉斜举朝镜头——竹刀尖端几乎怼到镜头前，占据画面上半部。 表情：微微歪头，目光从竹刀尖端移向镜头，眼神带一丝冷淡的挑衅——"要不要来试试"。**硬切** [04:30-06:30] 镜头3：穿甲回眸·障子门逆光（Armor Up / Backlit Silhouette） 场景：道场内木格障子推拉门正前方，门后透进大面积柔和白光形成强逆光。 穿着变化：主角在剣道着和袴外面加穿了胴甲（do / 剣道胸甲）——深藏青色漆边框+棕褐色鹿皮革面板，胸甲用深蓝棉绳交叉系扎在背后，下方垂片在腰间微微晃动。 动作：主角背对镜头站在障子门前，身体缓缓向右转——从纯背影转到四分之三侧身，最后回眸直视镜头。逆光在轮廓和胴甲边缘勾出一圈暖金色边缘光（Rim Light）。 表情：回眸瞬间嘴唇微启，下巴微抬，眼神沉静带锋芒——穿上铠甲后的笃定。**硬切** [06:30-09:30] 镜头4：师徒对打·竹刀交击（Sparring / Dynamic Profile Shot） 场景：回到练习区木地板，背景墙上书法条幅和障子门可见（与镜头1同一空间）。 机位：侧面中景（Profile Medium Shot），略低角度仰拍，两人对峙占据画面左右。轻微手持晃动增强临场感。 穿着：主角穿剣道着+袴+胴甲（与镜头3一致），双手握竹刀。教练穿黑色T恤+袴，戴深色皮革小手（kote / 护手），双手握竹刀。两人均未戴面（men），面部清晰可见。 动作序列（3秒内快节奏连续动作）： ① 两人中段构え（Chudan-no-kamae）对峙，竹刀尖端交叉轻触对碰； ② 教练率先劈下面打ち（头部劈击）——主角向右侧身闪避，竹刀从下往上格挡，两刀相交发出清脆的"啪"一声竹响； ③ 主角借格挡反弹力顺势向前踏込み（fumikomi），右脚重重踩地"咚"一声，同时竹刀从上方果断劈下——一记干脆利落的面打ち，竹刀划破空气发出锐利的破风声呼啸。 表情：对打全程目光凌厉专注，击中瞬间不自觉轻喊一声短促气合（kiai）。**硬切** [09:30-12:00] 镜头5：席地抚刀·松弛反差（Cool Down / Low Angle Seated） 场景：装备存放区，主角盘腿坐在储物架前的木地板上，身后整面墙的黑色防具储物架作为背景，暖光从上方洒下。画面只有主角一人。 穿着：剣道着+袴（胴甲已卸下），坐姿让袴的宽大裙摆在地板上自然散开。 动作：一把竹刀横放在膝盖上，双手自然搭在竹刀两端。她微微仰头向右后方甩了一下头发，然后缓缓转头面向镜头——闭上眼睛约一秒，慢慢睁开，目光柔和但带着一丝倦怠的慵懒。 整体氛围从前面的凌厉对打突然切换到极致松弛——反差感。**硬切** [12:00-15:00] 镜头6：抱刀终幕·冷眼定格（Final Pose / Medium Close-up） 场景：装备存放区，储物架正前方。暖黄灯光从侧面打来，面部一半明一半暗（Split Lighting）。 动作：主角站在储物架前，双臂交叉环抱一把带鞘居合刀（黑色漆面刀鞘、深色皮绳缠柄）紧贴胸前，刀柄露出右肩上方。 表情：下巴微收，目光从侧面缓缓转向镜头正前方，眼神冷而定——不笑、不挑衅，安静地注视镜头。 最后0.5秒画面微微降速（Slight Slow-mo），定格在这个抱刀注视的姿态上，自然结束。 【音效】 节奏明快的日式电子混搭BGM贯穿全片（Japanese Lo-fi Hip-hop / Trap Beat），鼓点清晰有力，每一次硬切严格踩在重拍上。叠加真实拟音：镜头1赤足在木地板移动的沉闷步声+刀身划过空气的轻微呼啸；镜头2竹刀竹片轻微震颤的嗡鸣；镜头3胴甲皮革和棉绳摩擦的窸窣声；镜头4竹刀相交"啪"的脆响+踏込み踩地板"咚"的重击+破风声+短促气合喊声；镜头5头发甩动带起的轻柔气流声；镜头6最后1秒BGM鼓点抽空，只剩道场空调低频底噪和远处隐约的环境音。 【一致性锁定】 1. 全片竖屏9:16，所有镜头保持轻微手持呼吸感晃动，保留Vlog实拍质感。 2. 主角面部严格锁定 @图片1，六个镜头之间不出现任何面部漂移。 3. 主角基础穿着始终为深藏青色刺子织剣道着+深藏青色袴+右腕浅色珠串手链；镜头3-4加穿棕褐色鹿皮胴甲，镜头5-6卸甲回到基础穿着。 4. 道场环境统一：木地板色调、护墙板材质、储物架样式、障子门位置在各镜头中保持一致。 5. 只用硬切，不许出现任何转场特效、模糊、叠化、黑场、闪白。 6. 画面内不出现任何品牌logo、可辨认文字、字幕、水印。书法条幅和防具上的文字不清晰显示、保持模糊或暗部遮挡。 7. 不出现镜面反射或水面倒影中的人像。 8. 教练仅在镜头1和镜头4出现，其余镜头只有主角一人。

Cinematic scene

Create an original, commercial-safe live-action comedy video with a strict runtime of exactly 15.0 seconds, vertical 9:16, 1080x1920, 24 fps. Natural neutral-American English only. STYLE: Rapid-fire intellectual roommate comedy where a completely ordinary household mystery is treated like a major scientific investigation. Original warm American multi-camera studio sitcom aesthetic. No existing TV IP, characters, actors, sets, catchphrases, memes, logos, music, or recognizable scenes. HERO ANIMAL: A photorealistic plush gray British-shorthair-type cat dominates the opening, occupying at least 40% of the frame. The cat wears an oversized mustard-and-navy plaid shirt, clear round glasses, and a backward forest-green baseball cap. It sits upright at a cluttered laundry-analysis workstation containing three beige CRT monitors, a spinning washing machine visible behind it, magnifying glass, measuring tape, tangled cables, laundry baskets, clothespins, notebooks with no readable text, and blinking generic indicator lights. The cat investigates clothing with natural feline paws and speaks in a low, dry American voice. Subtle feline muzzle movement only. Never human hands, fingers, or lips. HUMAN: An original American woman in her late 20s with shoulder-length curly hair, blue cardigan, cream shirt, jeans, and sneakers. Practical, skeptical, understated delivery. SET: Original apartment laundry room connected to a small hallway. Warm floor lamps, wooden shelves, detergent bottles with no readable branding, folded clothes, storage boxes, patterned rug, and a red sofa partially visible through the doorway. Warm tungsten studio lighting, subtle analog color and film grain. CAMERA: Extreme hero close-up, locked wide, medium two-shot, reaction close-up. Clean hard cuts. No handheld dialogue, fisheye, or dramatic camera movement. TIMELINE: 0:00-0:01.0 — Extreme close-up of cat's glasses reflecting CRT laundry charts while its paws rapidly type. Original one-second sitcom sting using drums, handclap, bass, and two guitar chords. Add keyboard clicks and washing-machine hum. 0:01.0-0:03.5 — Woman enters holding one lonely sock. WOMAN: “Did you find my other sock?” 0:03.5-0:04.8 — Cat continues typing. CAT: “Not yet.” 0:04.8-0:08.1 — Woman points at a laundry basket. WOMAN: “You've searched for two hours.” 0:08.1-0:08.9 — Light audience laugh. 0:08.9-0:11.7 — Cat adjusts glasses and studies the monitor. CAT: “The sock isn't missing. It's relocating.” 0:11.7 — Biggest audience laugh. 0:13.6 — Short original two-chord end sting. 0:14.0-0:15.0 — Hold on cat's intensely serious face as the woman slowly looks at the single sock. Clean upper-right negative space for advertising copy added in post. Generate no text. AUDIO: No music under dialogue. Clean centered voices. Washing-machine hum, keyboard clacks, tiny electronic beeps, subtle room tone. Laughter never covers dialogue. NEGATIVE: Existing sitcom IP, copied apartment, famous characters, real-person likeness, copyrighted music, recognizable meme, readable text, branding, watermark, captions, cartoon cat, human cat hands, human lips, extra limbs, identity drift, wrong-speaker mouth movement, exaggerated accent, handheld camera, fisheye, harsh HDR, overlapping speech, continuous laughter.

Cinematic scene

Thirty second continuous handheld take, cinematic realism, Japanese dawn market documentary register. A grandfather teaches his granddaughter to gut fish, and neither of them says the thing they mean. SUBJECT A: TAKEDA ISAO Man, seventy one, short and thick through the shoulders, deeply weathered brown skin, sun cracked neck, liver spots on the backs of his hands, white stubble, cropped grey hair under a navy knit cap, the tip of the left index finger missing at the first joint. Blue rubber apron over a padded navy work jacket, grey towel knotted at the throat, black rubber boots, thin blue gloves gone slack at the fingers. SUBJECT B: TAKEDA MIO Girl, seventeen, thin wrists, pale skin gone red at the cheeks and nose from cold, chapped lips, black hair in a low knot with strands stuck to her temple, a school hair elastic on one wrist, a plaster on her left thumb. Borrowed white rubber apron too big on her, school tracksuit underneath, boots a size too large, bare ungloved hands going pink. SETTING: A wet concrete fish market stall at five in the morning. A steel gutting table, crushed ice trays of mackerel, stacked blue plastic crates, a bare hanging bulb over the table and cold fluorescent tubes deeper in the hall, a green hose running unattended across the floor, a wall clock, an open loading bay to a blue black street where a small truck idles with its headlights on. STYLE: Photoreal, unglamorous, documentary. Breath fog, wet skin, fish scale stuck to a forearm, blood pink water thinning in the runoff, condensation on steel. Three light registers in one take. Hard warm tungsten from the bulb over the table, flat green cold fluorescent from the hall behind, and pale blue predawn spilling in from the loading bay. 35mm grain, soft halation on the bulb, desaturated cold grade with warm skin. CAMERA: One unbroken handheld take, no cuts. Close and low, working across the table between them. Long lens, very shallow depth of field. Hands and knife enter and leave the foreground, crates and hanging plastic strip curtains occlude the frame. Focus hunts between the fish, her face, and his hands, misses once, recovers. Never locked off. Dialogue language: Japanese, low and clipped, northern rural inflection. Sparse. Long silences between lines. Sentences left unfinished. 0 to 6 seconds Extreme close on the steel table. ISAO's gloved hands, missing fingertip visible, pin a mackerel, open it in three fast strokes, and sweep the guts into a bucket. Camera drifts up off the fish and racks to MIO across the table, out of focus, breathing on her bare hands. Isao, not looking up: {手袋。} She shakes her head once. <ice shifting, a knife on steel, a hose running> 6 to 12 seconds MIO takes a fish, sets it wrong, and cuts too shallow. Camera swings around the table into a low angle over her shoulder. Isao's hand comes into frame and stops hers. ISAO: {角度。刃を寝かせて。} MIO, tight: {やってる。} ISAO: {やってない。} 12 to 18 seconds Tight two shot, both faces in the same frame, cold blue loading bay light behind them, breath fogging. Her second cut tears the belly open badly. She keeps going anyway. MIO: {お母さんは、帰ってこないよ。} Isao says nothing. He takes the ruined fish, sets it in the crate, and slides her a fresh one. <a truck idling, crates knocked, gulls outside> 18 to 24 seconds ISAO wipes his blade on the towel at his neck and relays the fish in front of her, squaring it with two fingers. Camera whips to the hall as a crate crashes somewhere off frame, blurs, recovers on his face, jaw set, eyes down. ISAO, quiet: {五時までに、終わらせる。} MIO: {うん。} 24 to 30 seconds MIO cuts again. Cleaner. She does not look up for approval and he does not give it. Camera settles low across the table, her hands sharp in the foreground, ISAO a soft blur behind her turning back to his own pile. Pale blue dawn strengthens in the loading bay. Neither of them stops working. AUDIO: Diegetic only. A running hose, a knife through fish and onto steel, crushed ice shifting, rubber boots on wet concrete, breath, a truck idling outside, gulls, a distant radio in the hall, plastic crates stacked. No background music. No score. No subtitles. Last frame: MIO alone in the near foreground, sharp, both hands on the fish, blade halfway through, ISAO a soft grey shape behind her in the blue light of the open loading bay. Camera at rest, still handheld and unsteady. No on screen text, no captions, no title cards, no logo, no watermark. Negative Prompt: No cuts, no edits, no dissolves, no cutaways. No smooth gimbal or dolly motion. No music, no score, no subtitles, no on screen text. No glamour lighting, no beauty retouching, no smooth skin, no makeup. No smiling, no hugging, no reconciliation gesture. No additional people at the table. No slow motion, no speed ramps. No flickering, no morphing faces, no changing wardrobe. The missing fingertip and the too large boots stay consistent throughout.

Cinematic scene

CAMERA: Authentic MiniDV/DV 16mm camcorder handheld footage recorded entirely by CHASE herself. Every shot is filmed in selfie mode or first-person POV with natural hand shake, imperfect framing, delayed autofocus, occasional overexposure, clumsy zooms, slight focus hunting, accidental head cropping, and realistic camera mistakes. The camera always feels hand-held and personal, never stabilized or cinematic, and the camcorder itself never appears on screen. LOOK: Genuine early-2000s DV tape aesthetic with soft image quality, subtle motion blur, faint tape grain, light digital noise, blooming highlights, muted contrast, warm indoor white balance, realistic skin tones, slight chromatic aberration, and gentle exposure flicker, creating the feeling of authentic home-video footage rather than a polished production. STYLE: A quiet late-night dance vlog with a relaxed, intimate atmosphere. CHASE has just finished a long rehearsal and is physically exhausted but genuinely happy with her progress. Her breathing is natural, her laughs are spontaneous, and every moment feels unscripted and personal, as if she's casually updating fans before heading home. Character: CHASE is a Korean idol in her early 20s with long straight black hair tied into a slightly messy ponytail, a few loose strands framing her face, glowing skin with a light sheen of sweat, expressive eyes, and a slim athletic build. She wears an oversized long-sleeve athletic practice shirt, loose grey sweatpants, and white sneakers with no jewelry. Setting: A spacious dance practice studio late at night featuring wooden floors, a full mirror wall, a Bluetooth speaker in the corner, a water bottle, towel, and dance bag resting against the mirror, while the hallway outside is dark and quiet, emphasizing that everyone else has already gone home. Storyboard: (~2s, camera propped against the mirror, medium shot) CHASE walks into frame breathing heavily, wipes sweat from her forehead with the back of her hand, smiles naturally, and says, "Finally finished... I don't even know what time it is anymore." (~2s, handheld selfie) She picks up the camera, slowly pans across the empty studio and mirror wall before pointing it back at herself with a small smile. "Everyone already went home..." she says before quietly laughing. (~2s, handheld walking shot) She walks toward the wall, grabs her water bottle, takes a long refreshing drink, closes her eyes briefly, exhales, and smiles. "That tastes amazing right now." (~1.5s, macro insert with shallow depth of field) Close-up of condensation on the water bottle as her fingers wipe away the droplets while only her soft breathing is heard. (~2s, medium shot with camera propped against the mirror) She steps back into frame and performs a short, sharp dance combination with impressive energy before immediately laughing and bending forward while catching her breath. (~2s, handheld close-up selfie) Her cheeks are flushed, hair slightly messy, and she's smiling as she says, "Still messed up the ending... Tomorrow again." (~3s, arm's-length selfie while walking toward the exit) She drapes a towel over her shoulder, reaches to switch off the studio lights behind her, gives the camera one warm smile and a small wave, then softly says, "Good night... see you tomorrow," before reaching toward the lens to stop the recording as the footage ends with a subtle MiniDV tape glitch.

一句话了解

这个模型怎么样？

无视频输入时，可选 4、6、8、10 秒；使用视频输入时，输出时长由模型决定，所选时长不生效。

模型定位
可从文字描述开始生成短片，也可加入参考图或视频片段。参考素材用于表达创作方向，动作、文字与主体的一致性仍需检查。
适合谁用
构思概念短片、尝试参考图动态化，以及围绕现有视频片段创作新版本。
本站使用说明
提交前结合本页的输入限制和工作台估算确认方案；任务受理后先预扣积分，符合结算条件后再退还差额。

关键参数一览

根据这些信息，判断模型是否适合你的任务。

厂商
Google
模型家族
Gemini Omni
提示词上限
20,000 字符

它还叫什么

同一个模型在公开资料中可能有不同名称，这里整理相关称呼，方便查找和比较。

Gemini Omni Video

用户常见问题

从任务、输入条件和结果要求出发，整理选择模型时需要确认的问题。

Gemini Omni Video 视频生成
Gemini Omni Video 积分预扣
Gemini Omni Video 可调参数
Gemini Omni Video 视频输入限制

选型指南

根据输入素材、任务要求和积分成本，逐项确认模型是否适合。

1
从文字开始

已有明确的短片构想时，可先围绕一个主要动作写出主体、场景和镜头运动。

2
已有参考图

需要参考构图或外观时，说明每张图中希望保留的部分，并写清允许发生的变化。

3
已有视频片段

先选取与构想相关的片段，再描述希望发生的变化。裁剪窗口最长 10 秒，不能用所选时长控制输出。

适用情况

依据公开资料与当前站内设置，说明模型适用的任务及需要注意的限制。

先看预扣，再看结算

提交前的积分数是估算。任务受理后预扣，成功后仍须核实状态和实际用量；未满足结算条件时，预扣会继续保留，等待核对。

确认素材与输出边界

自定义音频与角色资源暂不开放。参考素材不能保证主体或文字保持不变，成片仍需检查后再使用。

能力详解

文字生成短片

用文字描述主体、动作、场景和镜头。未输入视频时，可选的时长为 4、6、8、10 秒。

加入参考图

无视频输入时最多可用 7 张参考图。提示词应交代图片之间的关系，以及希望保留或改变的内容。

从视频片段生成新版本

可输入 1 段视频，并配合最多 5 张参考图。先设置片段起止位置；模型自行决定输出时长，所选时长不会生效。

适合做什么

概念短片草稿

把想法写成短片描述，围绕一个主要动作安排主体和镜头，用于检查创作方向。

参考图动态化

为现有图片尝试动作或镜头变化，在提示词中说明希望保留的外观与构图。

视频片段再创作

围绕已有片段尝试新的表现，先检查生成结果中的动作连续性与主体变化，再决定是否调整素材重做。

提示词技巧

把动作与镜头分开写

先描述一个主要动作，再说明镜头如何移动，避免把互相冲突的指令挤在一起。

说明参考图的用途

逐张交代图片中的主体、构图或氛围各有什么用途，并说明哪些部分允许变化。

把修改目标写清楚

使用视频片段时，直接写明希望改变和保留的内容；调整所选时长不能控制这类请求的成片时长。

为什么选它

可从文字、参考图或视频片段开始创作，并按输入方式配置素材。
可选择横向 16:9 或竖向 9:16 画面，分辨率提供 720p、1080p 和 4K 选项。
预扣档位可按是否输入视频、所选分辨率及时长查阅，便于提交前比较方案。

使用前要知道

源视频最长 30 秒，裁剪窗口最长 10 秒；上传成功后，提交时仍会校验输入限制。
使用视频输入时，输出时长由模型决定，所选时长不会生效。
自定义音频与角色资源暂不开放；动作、文字和主体一致性需要逐次检查。

可调参数

根据这些信息，判断模型是否适合你的任务。

分辨率
可选
参数类型: 下拉选择
720p
1080p
4k
视频时长
可选
参数类型: 下拉选择
4
6
8
10
宽高比
可选
参数类型: 下拉选择
16:9
9:16
随机种子
可选
参数类型: 数字输入
视频起点（秒）
可选
参数类型: 数字输入
视频终点（秒）
可选
参数类型: 数字输入

积分消耗

45–180 积分预扣

未输入视频时，720p 和 1080p 的 4 / 6 / 8 / 10 秒请求分别预扣 45 / 60 / 75 / 90 积分；4K 分别预扣 105 / 120 / 135 / 150 积分。输入视频时，720p 和 1080p 预扣 120 积分，4K 预扣 180 积分。提交前显示估算，任务受理后预扣。任务成功后，还须确认任务状态与有效实际用量；用量向上取整后不超过预扣，才按此结算并退还差额。用量缺失、无效、超出预扣或状态未核实时，继续保留预扣，等待核对。失败任务的费用请查看积分明细。

使用建议

准备多次生成时，建议先用少量素材测试，核对输入设置和结果是否适合任务，再决定是否继续。每次生成均按工作台显示的计费规则扣点。

常见问题

相关模型

按你的任务比较这些模型的输入要求、功能和当前状态，再查看对应详情。

Seedance 2.5

用 Seedance 2.5 规划较长镜头，先选择文字、首帧与首尾帧或多模态参考场景。请求时长可设为最长 30 秒；默认 5 秒，选择自动时长则按 30 秒预留预计积分。准备素材时，数量、单段时长与累计时长需分别核对。

AI 视频模型
预计预扣 102–5,700 积分
查看详情立即使用

MiniMax H3

MiniMax H3 可从文字、首尾帧或多模态参考开始，选择 768P 或 2K 及 4–15 秒请求时长。首尾帧场景允许只给首帧或尾帧；参考场景中的音频须搭配图片或视频。参考视频时长和第 6 张起的图片会影响预计预扣。

AI 视频模型
预计预扣 32–406 积分
查看详情立即使用

HappyHorse 1.1

先按图片数量准备输入：不附图片时用文字，1 张用于单图，2–9 张用于多图参考。再核对图片条件、提示文字、分辨率和请求时长，查看预计预扣。上传通过后仍需检查当前提交提示；生成后检查画面是否合用，并核对费用是否确认。如果结果含声音，再检查声音是否合用。

AI 视频模型
预计预扣 68–435 积分
查看详情立即使用

信息来源

页面内容依据公开模型资料与 Kyeo AI 的模型信息整理。能否在线使用，请以当前模型状态为准。

来源说明

本站支持文字、参考图和视频片段输入，图片与视频可组合使用。自定义音频与角色资源暂不开放；生成后仍需检查动作、文字和主体一致性。

内容复核日期: 2026-09-15
本站配置与接口资料
平台

输入配额、参数选项和预扣档位依据本站配置整理；视频输入的时长行为依据该模型接口资料。可选值与估算不保证具体成片效果，提交时请核对工作台信息。