多参考生成
单次生成最多可使用 50 个参考素材:30 张图片、10 个视频和 10 个音频文件。结合人物图像、产品图片、动作参考、场景参考与音频,在复杂视频中保持主体、动作、环境和视觉风格更高的一致性。
使用 Seedance 2.5 创作 AI 视频的完整指南。
Seedance 2.5 是字节跳动 Seed 团队推出的新一代 AI 视频模型,支持 30 秒视频生成、50 个参考输入、更强的一致性以及先进的编辑能力。
Seedance 2.5 是字节跳动 Seed 团队开发的新一代 AI 视频生成模型,于 2026 年 6 月 23 日发布,并于 2026 年 8 月 7 日在 SeeGen AI 上线。
作为 Seedance 2.0 的继任者,它在视频时长、参考控制、一致性和编辑能力方面均有提升。
Seedance 2.5 带来更长的视频生成、更丰富的多模态参考、精准的视频编辑、独立音频参考,以及更广泛的语言支持。
单次生成最多可使用 50 个参考素材:30 张图片、10 个视频和 10 个音频文件。结合人物图像、产品图片、动作参考、场景参考与音频,在复杂视频中保持主体、动作、环境和视觉风格更高的一致性。
单次生成最长可达 30 秒视频,是 Seedance 2.0 最长 15 秒的两倍。更长的时长为多步动作、产品演示、对话、镜头运动和完整故事情节提供了更大空间,无需拆分为多个片段。
可对现有视频的特定部分进行编辑,而无需重新生成整个场景。Seedance 2.5 能够替换物体、更改视觉细节、调整人物或修改场景局部,同时保留原始构图、光照、动作、音频和时间线连续性等要素。
可将音频作为独立参考使用,无需图片或视频参考。最多可上传 10 个音频文件,每个文件时长 2–30 秒,音频总时长最长 30 秒,用于引导节奏、对白、氛围、节拍和视听同步。
支持生成10多种语言的视频,包括中文、英语、西班牙语、印尼语、马来语、泰语、阿拉伯语、葡萄牙语、越南语、日语和韩语。这让 Seedance 2.5 更适合本地化广告、多语言叙事、产品视频以及面向国际受众制作的内容。
可向前或向后延展现有视频,在原片段之外继续展开场景。Seedance 2.5 能够保留主体、光线、动作和视觉风格等关键元素,让你更轻松地在已有素材基础上构建更长的镜头序列。
一个好的 Seedance 2.5 提示词应清晰描述你想看到的画面、主体应如何运动,以及最终视频应具有怎样的风格。
主体
动作
环境
光线
镜头运动
视觉风格
画质
限制条件
主体和动作是必填项,其他元素可根据需要添加。如果提示词较长,可加入简单的时间线,让动作和镜头运动更清晰。
Create a 30-second ultra-realistic candid home-video of a young Korean woman spending a quiet late morning in a residential Korean neighborhood.
SUBJECT:
Young Korean woman, early 20s, natural skin, minimal makeup, black wavy hair in a messy side ponytail with wispy bangs. Faded charcoal sleeveless crop top, loose light-wash jeans, black canvas sneakers, black cord necklace. Keep her face, body, hairstyle, and clothing consistent throughout.
SETTING:
Quiet Korean residential alleys with low-rise homes, terraces, potted plants, laundry lines, bicycles, utility poles, overhead wires, and mature trees. No shops, cafés, crowds, or commercial activity.
STYLE:
Early-2000s consumer DV camcorder footage. Imperfect handheld shake, awkward framing, autofocus hunting, exposure shifts, slight motion blur, faded colors, soft contrast, compression noise, and occasional reframing. No stabilization, cinematic camera moves, or modern color grading. It should feel genuinely recorded, not AI-generated.
TIMELINE:
00:00–00:05 — She sits outside her house adjusting her ponytail, smiling casually as the camera struggles to focus.
00:05–00:10 — She walks into a narrow alley, crouches to pet and feeds a stray cat. Focus shifts between her and the cat.
00:10–00:15 — She hangs laundry in a small front yard. Fabric moves in the breeze as exposure changes naturally.
00:15–00:20 — She sits on a terrace holding a ceramic coffee cup, watching the neighborhood and brushing hair behind her ear.
00:20–00:25 — Side profile. Someone off-camera greets her. She turns, smiles, waves, and naturally says, "Annyeong."
00:25–00:30 — She walks down a tree-lined lane with her coffee, notices the camera, gives a small smile, looks away, and keeps walking. Abrupt cut to black mid-motion.
AUDIO:
Natural location sound only: birds, distant motorcycles, wind, leaves, footsteps, faint neighborhood chatter, cat sounds, and laundry movement. Natural Korean speech only. No music, narration, or cinematic sound effects.
GOAL:
Make it feel like a forgotten early-2000s personal home video: intimate, spontaneous, imperfect, warm, and believable. Prioritize realistic motion, natural expressions, environmental detail, and identity consistency.探索真实的 Seedance 2.5 案例、可直接使用的提示词,以及适用于不同视频工作流的创意用法。
Create a 20-second, 16:9 photoreal live-action sci-fi sequence with SEVEN slow, deliberate shots. STYLE: Large-format anamorphic cinema. Heavy fog, slow falling snow, pale ice-blue, bone-white, and wet slate-grey palette with almost no saturation. The ONLY accent color is molten gold. Flat overcast light, no sun or hard shadows. Fine film grain, restrained grade, high-budget sci-fi epic. Original character and creature design only. CHARACTER LOCK: One slight lone figure in a floor-length bone-white felt coat with a deep fur-lined hood. Frost on shoulders and hem, ash-white wet hair, pale skin, frost on eyelashes, dark inner layers, worn gloves. No armor, weapons, insignia, jewelry, piercings, or forehead markings. Eyes stay closed until shot 4, then glow molten gold. COLOSSUS LOCK: An immense kneeling stone figure half-buried in ice. Only the bowed head, one shoulder, and one open palm are visible. Weathered granite, erosion, moss, and ice. Smooth ruined face, eyes closed, no horns, teeth, or demonic features. Thin dormant gold veins run through the cracks. WORLD: A frozen drowned city with ruined towers, snapped bridges, industrial pipes, and cables fading into white fog. No people, text, or signage. SHOTS: 0–3s — Extreme wide: frozen city, giant bowed head and open palm rising from the ice. Snow falls steadily. 3–6s — Wide slow push-in: the tiny figure stands on the colossus's palm, smaller than one fingernail. 6–9s — Close portrait: eyes closed, frost on lashes, visible breath, soft white fog behind. 9–12s — Extreme close-up: eyes open and glow deep molten gold, the first and only strong color. 12–15s — Wide low angle: the figure stands on a hovering dark stone slab above the city, coat blowing in the wind, colossus behind. 15–17.5s — Medium: the figure slowly raises one gloved hand toward the city below. 17.5–20s — Extreme wide: molten-gold veins spread through the ice and colossus as towers, bridges, and cables silently rise into the fog. CUTS: Hard cuts at 3s, 6s, 9s, 12s, 15s, and 17.5s. No fades, dissolves, speed ramps, slow motion, or opening freeze. Continuous subtle motion in every shot: snow, fog, breath, and fabric. AUDIO: Low wind, groaning ice, snow on fabric, sub-bass swell when the eyes open, rising low hum as the gold spreads, deep grinding as the city rises. No voice, narration, dialogue, or music. NEGATIVE: Anime, manga, stylized animation, ninja imagery, patterned cloaks, ringed irises, orange/spiky hair, weapons, gore, crowds, warm sunlight, blue sky, saturated colors except gold, lens flares, text, logos, watermark, cartoon, 3D render, video-game cinematic look, slow motion, speed ramping.
@Image1 as the only protagonist. Strictly preserve the protagonist's face, hairstyle, skin tone, age, and visual identity throughout. Create a 30-second cinematic emotional confrontation. This is not an action scene or gunfight. The entire scene focuses on the protagonist's painful internal struggle before pulling the trigger. SETTING: A quiet, dark, oppressive interior such as an abandoned room, hallway, warehouse, or nighttime space. Keep the background simple and unobtrusive. Moody cinematic lighting with clear facial contrast so the protagonist's eyes, tears, and subtle expressions remain visible. CAMERA: Over-the-shoulder shot from behind the other person. Never show their face; only a blurred shoulder, back of the head, or silhouette in the foreground. The camera faces the protagonist directly, with the gun aimed toward the camera. Stable cinematic handheld with subtle breathing and a very slow push-in. Keep focus on the protagonist's face, eyes, trembling hand, and gun. PERFORMANCE: The protagonist keeps the gun raised throughout. He is angry, hurt, resentful, heartbroken, and forcing himself to act, but clearly cannot bring himself to shoot. His eyes are red and filled with tears, breathing grows heavier, lips tremble slightly, jaw stays tense, and his hand subtly shakes. Several times his finger tightens as if he is about to pull the trigger, but he stops at the last moment. His expression should constantly shift between anger, pain, attachment, resentment, sorrow, and reluctance. He is not cold-blooded and not simply afraid — the feeling is: "I have to do this, but I cannot make myself do it." Keep the acting restrained and realistic. No screaming, dramatic crying, or exaggerated movement. Tears gather but do not fully fall apart. The emotional power comes from someone barely holding themselves together. ENDING: End unresolved. He is still aiming the gun, still unable to pull the trigger, breathing unevenly with tear-filled eyes. Hold on this suspended high-tension moment. STYLE: Cinematic emotional close-up, restrained performance, subtle facial acting, trembling hands, tear-filled eyes, high-tension silence, dramatic over-the-shoulder composition. Minimal cuts and minimal physical movement. Do not turn it into an action movie. AUDIO: Real environmental sound only: faint room tone, distant reverb, uneven breathing, soft clothing movement, and subtle hand/body movement. No music, dialogue, narration, subtitles, text, or UI.
Create a 30-second tutorial video showing how to set up and use a capsule coffee machine. 0–2s — Title card: "Seedance Capsule Coffee Machine Setup Tutorial." 2–5s — Step 1: Install the water tank. Use @image1. Slightly high rear angle. Align the tank with the back slot and push down until it clicks. VO: "First, install the water tank. Align it with the back slot and push until it clicks." 5–9s — Step 2: Install the drip tray. Use @image2. Front close-up. Slide the tray into the bottom rails until fully seated. VO: "Next, install the drip tray by sliding it into the bottom rails." 9–13s — Step 3: Install the used-capsule box. Use @image3. Close-up from a low angle. Push the box into the cavity beneath the drip tray. VO: "Then insert the capsule collection box. Used capsules will drop here automatically." 13–18s — Step 4: Fill the water tank. Use @image4. Side close-up. Open the lid, fill with clean water to the MAX line, then close it. VO: "Fill the tank with clean water, but do not exceed the MAX line." 18–25s — Step 5: Power on. Use @image5. Front medium shot. Plug in the machine and press the power button. Show the indicator blinking, then turning steady. VO: "Press the power button. When the blinking light turns steady, the machine is ready." 25–30s — Step 6: First rinse. Use @image6. Front-side view. Without inserting a capsule, press the brew button and let hot water rinse the system. VO: "For the first rinse, do not insert a capsule. Press brew, and once rinsing is complete, the machine is ready to use."
Create a bright, colorful commercial for fruity cookies in four flavors: strawberry, apple, grape, and orange. Use @image1 for the strawberry flavor reference. STYLE: Clean, premium, high-energy product ad with geometric cookie arrangements, vivid fruit colors, fast rhythmic editing, and strong musical timing. SEQUENCE: - Open with fruits rapidly orbiting around the hero cookie, inspired by @video1. - Different cookie flavors spiral toward the camera, switching on the beat, inspired by @video2. - Cookie arrays pan left and right with fast flavor changes, inspired by @video3. - Add vertical up-and-down movement like a precise machine, inspired by @video4. - Climax: a cookie snaps in half in slow motion, filling bursts out and crumbs scatter, inspired by @video5. - End with all four flavors lined up neatly, fruits bouncing in sync, and the text "Fresh on Seedance, made for viral vision" appearing word by word, inspired by @video6. Keep the visuals colorful, delicious, youthful, energetic, and highly shareable.
Create a 30-second single-take handheld boyfriend travel vlog in Tokyo. Use the woman from @Image1 as the main character. Keep her face, hairstyle, body proportions, and overall appearance fully consistent throughout. STYLE: Authentic personal phone or small-camera vlog footage. Natural handheld shake, imperfect framing, autofocus shifts, slight motion blur, and realistic exposure changes. Casual, spontaneous, and non-cinematic. She does not pose and sometimes forgets the camera is there. TIMELINE: 0–5s — Morning in a small Tokyo apartment. She fixes her hair near the bed, notices the camera, smiles, laughs, and playfully tells her boyfriend to stop filming. 5–10s — They walk through a quiet Tokyo neighborhood. She enters a convenience store, looks at drinks and snacks, then turns and asks which one she should choose. 10–18s — At a small local restaurant, she eats ramen with chopsticks, reacts naturally to the hot food, and laughs while the boyfriend laughs behind the camera. 18–25s — They explore a lively Tokyo neighborhood. She browses shops, takes photos, and occasionally looks back at the camera while crowds pass naturally. 25–30s — Tokyo at night. She walks through illuminated streets, turns back and smiles, then sits by a train window watching city lights pass outside. REQUIREMENTS: Natural expressions, realistic movement, authentic Tokyo ambience, stable identity, and continuous handheld vlog feeling. No commercial style, dramatic posing, text, logos, CGI look, artificial transitions, or identity changes.
Cinematic hip-hop / rap music video, photoreal quality, high-end tone, seaside setting. Build the frame from @image1: a band performs at a golden sand beach with crashing waves — a lead vocalist gripping a mic on a stand in the wet sand, one guitarist left, one right, a drummer at the back; a vast coastline behind, rolling waves, a warm golden-hour sun shimmering on the water, sea mist in the air. The lead in a red tracksuit raps to camera — lips and jaw precisely synced to every word, head punching to the beat. Bright, punchy, fast, confident rap. HARD CUT on the beat, each switch a double contrast (shot size and type change together). Lyrics (the lead sings 'hello' in each language in turn, precisely lip-synced): English "Hello", Chinese "你好", Japanese "こんにちは", Korean "안녕하세요", Portuguese "Olá", Thai "สวัสดี", Spanish "Hola", Arabic "مرحبا". 8 hard-cut shots (low-angle wide establishing; close-up rap to camera; macro guitar-string insert; 3/4 prowling orbit; lateral track at shore; drummer tilt-up; tight push on the lead; heroic full-band push-in), one language per shot. White balance 4000K, teal-and-amber grade, 35mm, shallow depth of field, film grain, sea mist, golden-hour flare. Premium feel, precise lip-sync, no subtitles, no text overlays, hard cuts only, total 20 seconds.
Seedance 2.5 提供更长的视频、更多参考输入和更强的编辑能力,而 Seedance 2.0 仍是更快、更实惠的短视频选择。
| 功能 | Seedance 2.5 | Seedance 2.0 |
|---|---|---|
| 状态 | 已在 SeeGen AI 上线 | 已在 SeeGen AI 上线 |
| 主要定位 | 更长视频与进阶创意控制 | 质量、控制与成本的平衡 |
| 适用场景 | 短剧、广告、音乐视频和复杂叙事 | 日常 AI 视频、广告和创意项目 |
| 视频质量 | 最佳整体质量与场景连贯性 | 稳定可靠的视频质量 |
| 角色一致性 | 更强的角色、面部和场景一致性 | 适用于大多数标准场景的良好一致性 |
| 运动控制 | 先进的运动、镜头和基于参考的控制 | 自然运动,具备扎实的提示词控制 |
| 视频时长 | 最长 30 秒 | 最长 15 秒 |
| 多模态参考 | 最多 30 张图片、10 个视频和 10 个音频参考 | 最多 9 张图片、3 个视频和 3 个音频参考 |
| 独立音频输入 | 支持 | 不支持 |
| 视频编辑 | 精确的视频编辑与参考生视频工作流 | 支持视频编辑、延展和参考工作流 |
| 定价 | 高级功能对应的高端定价 | 标准定价 |
| API 访问 | SeeGen AI 已支持 | SeeGen AI 已支持 |
| 现在就需要的最佳之选 | 选择它以获得最大控制力、更长视频和复杂项目支持 | 选择它以获得质量、功能与价格的平衡 |
生成前请上传参考图片、视频或音频。真人素材可能需要审核,公众人物、受版权/知识产权保护的角色或受限内容可能会被拒绝。
选择参考素材,撰写描述所需视频的提示词,选择时长和其他设置,然后点击生成以提交任务。
预览生成的视频并下载到设备,复制分享链接,或将作品分享到 SeeGen AI Discord 社区及其他社交平台。
借助多参考输入、音频引导和高级编辑功能,把你的创意变成更长、更连贯的 AI 视频。一站式搞定!