跳转到内容

Seedance 2.0 Reference to Video

集成

模型 id bytedance/seedance-2.0/reference-to-video
# Run the model; wait up to 20 s for the finished result in the same call
curl -X POST "https://api.doitong.ru/v1/run/bytedance/seedance-2.0/reference-to-video?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Duration: 10s | Strict 16:9 Landscape | Native 4K UHD | 24fps. Genre: Cinematic Slice-of-Life Fantasy Comedy / Elemental Magic. Target: Ultra-realistic premium live-action fantasy cinema. Use @ image1 as the sole identity reference; preserve exactly as shown, ignoring only background, pose, framing, and text. Exactly one Character A, with no redesign, duplication, morphing, identity drift, outfit change, or hairstyle change. In an old witch’s kitchen, Character A attempts to make one cup of tea using elemental magic, but the spell becomes unexpectedly powerful. The scene begins with a table-level macro shot of a still ceramic cup of tea while Character A’s tired, unimpressed face remains softly blurred behind it. Character A reaches for the wand and raises it in profile. A thin orange-blue magical filament travels continuously from the wand tip through open space into the center of the tea, visibly making contact with the liquid and creating a subtle glowing ripple and warm reflection inside the cup; only after contact does the tea begin rotating. The stirring starts gently, then accelerates unnaturally fast, creating a small controlled whirlpool and a few droplets rising from the surface without splashing wildly. Character A watches with slightly widened eyes, then makes one quick reverse wand gesture to cancel the spell. The tea immediately becomes still. Character A lowers the wand onto the table, lifts the cup carefully, and takes one small sip. There is absolutely no disgust before tasting. After a brief beat, the excessive heat registers: Character A’s eyes narrow, brows pinch, nose wrinkles slightly, and the lips tighten into a restrained, exhausted grimace. Character A leans forward and spits the same mouthful of tea back into the same ceramic cup, keeping the lower face and cup visible together; show one small restrained spray of liquid leaving the mouth and landing back inside the cup, with no vomiting or excessive saliva. Character A then stares at the cup with quiet offended disbelief while lowering it toward the table, ending during the movement with no smile, camera look, victory pose, freeze frame, or fade. Camera language is cinematic macro depth with rack focus, tight profile, wand insert, readable over-the-shoulder magical contact, overhead tea-whirlpool detail, extreme eye close-up, and tight three-quarter reaction, maintaining consistent screen direction and avoiding chaotic camera shake or teleportation. Audio consists of quiet witch-kitchen ambience, wand ignition, a subtle magical hum, liquid swirling, cancellation sound, a small sip, restrained “ptuh,” soft liquid landing, and a disappointed exhale. Music is cozy, mischievous slice-of-life comedy using acoustic guitar, pizzicato strings, muted piano, playful clarinet, and brushed percussion, with no vocals; build gently during the magical stirring, briefly thin the score as the heat registers, then add one understated plucked accent when the tea lands back in the cup. Keep music below the sound effects and avoid boing sounds, circus music, meme sounds, or laugh tracks. Maintain strict causality: FILAMENT CONTACT → TEA STIRS → SPELL ESCALATES → SPELL CANCELLED → TEA STILL → SIP → HEAT REGISTERS → SAME MOUTHFUL SPAT BACK INTO SAME CUP → DEADPAN REGRET. Priority locks are identity, visible magical contact, continuous tea-state consistency, sip before reaction, the same mouthful visibly returning to the same cup, stable facial anatomy, restrained acting, realistic liquid physics, cinematic close-ups, and readable comedic timing. Negative prompt: duplicate character, duplicate cup, duplicate wand, duplicate tea, missing magical contact, magic hitting table or hand, reaction before sipping, new liquid appearing, whole cup spat out, vomiting, excessive saliva, broken mouth anatomy, premature stirring, instant violent whirlpool, large magical explosion, injury, identity drift, exaggerated acting, cartoon physics, anime, cheap CGI, plastic skin, discontinuity, frozen frames, subtitles, text, logo, watermark.",
    "images": [
      "https://content.doitong.com/dev-api/2026/10/21/req_c3dk4hgrd2tzlsymbg986itv/0-6Vz76mAQT0PWAMVCZaZAY_KTBfm8A8.png"
    ],
    "resolution": "480p",
    "duration": "5",
    "aspect_ratio": "16:9",
    "generate_audio": true
  }'

# Still "queued" or "running"? Poll with the id from the response:
curl "https://api.doitong.ru/v1/requests/REQUEST_ID?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY"

此请求: $0.705 705 API 积分 · ≈ ₽75.84 在此页面运行

获取 API 密钥
ByteDance 参考图生视频 原生音频4K 输出参考输入4–15 秒最高 4k #audio#references#consistency

根据最多 9 张图像、3 个视频和 3 个音频片段的名称引用,生成带声音的 4 到 15 秒片段。

示例

输入
输出 · ~219 秒

提示词Duration: 10s | Strict 16:9 Landscape | Native 4K UHD | 24fps. Genre: Cinematic Slice-of-Life Fantasy Comedy / Elemental Magic. Target: Ultra-realistic premium live-action fantasy cinema. Use @ image1 as the sole identity reference; preserve exactly as shown, ignoring only background, pose, framing, and text. Exactly one Character A, with no redesign, duplication, morphing, identity drift, outfit change, or hairstyle change. In an old witch’s kitchen, Character A attempts to make one cup of tea using elemental magic, but the spell becomes unexpectedly powerful. The scene begins with a table-level macro shot of a still ceramic cup of tea while Character A’s tired, unimpressed face remains softly blurred behind it. Character A reaches for the wand and raises it in profile. A thin orange-blue magical filament travels continuously from the wand tip through open space into the center of the tea, visibly making contact with the liquid and creating a subtle glowing ripple and warm reflection inside the cup; only after contact does the tea begin rotating. The stirring starts gently, then accelerates unnaturally fast, creating a small controlled whirlpool and a few droplets rising from the surface without splashing wildly. Character A watches with slightly widened eyes, then makes one quick reverse wand gesture to cancel the spell. The tea immediately becomes still. Character A lowers the wand onto the table, lifts the cup carefully, and takes one small sip. There is absolutely no disgust before tasting. After a brief beat, the excessive heat registers: Character A’s eyes narrow, brows pinch, nose wrinkles slightly, and the lips tighten into a restrained, exhausted grimace. Character A leans forward and spits the same mouthful of tea back into the same ceramic cup, keeping the lower face and cup visible together; show one small restrained spray of liquid leaving the mouth and landing back inside the cup, with no vomiting or excessive saliva. Character A then stares at the cup with quiet offended disbelief while lowering it toward the table, ending during the movement with no smile, camera look, victory pose, freeze frame, or fade. Camera language is cinematic macro depth with rack focus, tight profile, wand insert, readable over-the-shoulder magical contact, overhead tea-whirlpool detail, extreme eye close-up, and tight three-quarter reaction, maintaining consistent screen direction and avoiding chaotic camera shake or teleportation. Audio consists of quiet witch-kitchen ambience, wand ignition, a subtle magical hum, liquid swirling, cancellation sound, a small sip, restrained “ptuh,” soft liquid landing, and a disappointed exhale. Music is cozy, mischievous slice-of-life comedy using acoustic guitar, pizzicato strings, muted piano, playful clarinet, and brushed percussion, with no vocals; build gently during the magical stirring, briefly thin the score as the heat registers, then add one understated plucked accent when the tea lands back in the cup. Keep music below the sound effects and avoid boing sounds, circus music, meme sounds, or laugh tracks. Maintain strict causality: FILAMENT CONTACT → TEA STIRS → SPELL ESCALATES → SPELL CANCELLED → TEA STILL → SIP → HEAT REGISTERS → SAME MOUTHFUL SPAT BACK INTO SAME CUP → DEADPAN REGRET. Priority locks are identity, visible magical contact, continuous tea-state consistency, sip before reaction, the same mouthful visibly returning to the same cup, stable facial anatomy, restrained acting, realistic liquid physics, cinematic close-ups, and readable comedic timing. Negative prompt: duplicate character, duplicate cup, duplicate wand, duplicate tea, missing magical contact, magic hitting table or hand, reaction before sipping, new liquid appearing, whole cup spat out, vomiting, excessive saliva, broken mouth anatomy, premature stirring, instant violent whirlpool, large magical explosion, injury, identity drift, exaggerated acting, cartoon physics, anime, cheap CGI, plastic skin, discontinuity, frozen frames, subtitles, text, logo, watermark.

5 s480p16:9generate_audio: true$0.705 · 705 API 积分

X 上 @MahiraEhan 的提示词

构建请求

控件根据模型架构生成。更改控件时,代码和价格会同步更新。

•必填

指导视频生成的文本提示词。

image_urls
拖放文件,或 · 最多 9 个

用于视频引导的图像参考,最多 9 个图像,且所有模态合计最多 12 个文件。每个图像最多 30 MB,格式必须为 JPEG、PNG 或 WebP。在提示词中使用 @Image1、@Image2 及后续编号标签标识这些参考图像。

video_urls
拖放文件,或 · 最多 3 个

用于生成的参考视频,以 MP4 或 MOV 文件提供,并在提示词中使用 @Video1、@Video2 及后续编号标签标识。每个视频的分辨率必须约为 480p (640x640) 至约 720p (834x1112)。最多 3 个视频,总大小小于 50 MB,总时长为 2–15 秒。

duration 5 s

视频时长选择:`auto` 根据提示词交由模型决定,明确指定的时长必须为 4–15 秒。

resolution

输出分辨率:480p 优先速度,720p 提供均衡效果,1080p 提供高质量,4k 提供最高质量。

aspect_ratio

输出宽高比:16:9 为横向,9:16 为纵向,1:1 为正方形,21:9 为电影级超宽屏。选择 `auto` 由模型选择宽高比。

高级 audios · bitrate_mode · codec
audio_urls
拖放文件,或 · 最多 3 个

用于视频引导的音频参考至少需要一个配套的图像或视频参考。提供 MP3 或 WAV 文件,每个最多 15 MB,最多 3 个文件,总时长上限为 15 秒。在提示词中使用 @Audio1、@Audio2 及后续编号标签标识。

bitrate_mode 默认 standard

选择输出编码比特率:'standard' 保留默认设置,'high' 要求模型提高编码质量,但会生成更大的文件。

codec 默认 auto

视频编解码器选择:`auto` 保持默认编解码器行为,`H264` 选择 H.264,`H265` 选择 H.265。

POST /run/bytedance/seedance-2.0/reference-to-video?wait=20 报价 → $0.705
# Run the model; wait up to 20 s for the finished result in the same call
curl -X POST "https://api.doitong.ru/v1/run/bytedance/seedance-2.0/reference-to-video?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "prompt": "Duration: 10s | Strict 16:9 Landscape | Native 4K UHD | 24fps. Genre: Cinematic Slice-of-Life Fantasy Comedy / Elemental Magic. Target: Ultra-realistic premium live-action fantasy cinema. Use @ image1 as the sole identity reference; preserve exactly as shown, ignoring only background, pose, framing, and text. Exactly one Character A, with no redesign, duplication, morphing, identity drift, outfit change, or hairstyle change. In an old witch’s kitchen, Character A attempts to make one cup of tea using elemental magic, but the spell becomes unexpectedly powerful. The scene begins with a table-level macro shot of a still ceramic cup of tea while Character A’s tired, unimpressed face remains softly blurred behind it. Character A reaches for the wand and raises it in profile. A thin orange-blue magical filament travels continuously from the wand tip through open space into the center of the tea, visibly making contact with the liquid and creating a subtle glowing ripple and warm reflection inside the cup; only after contact does the tea begin rotating. The stirring starts gently, then accelerates unnaturally fast, creating a small controlled whirlpool and a few droplets rising from the surface without splashing wildly. Character A watches with slightly widened eyes, then makes one quick reverse wand gesture to cancel the spell. The tea immediately becomes still. Character A lowers the wand onto the table, lifts the cup carefully, and takes one small sip. There is absolutely no disgust before tasting. After a brief beat, the excessive heat registers: Character A’s eyes narrow, brows pinch, nose wrinkles slightly, and the lips tighten into a restrained, exhausted grimace. Character A leans forward and spits the same mouthful of tea back into the same ceramic cup, keeping the lower face and cup visible together; show one small restrained spray of liquid leaving the mouth and landing back inside the cup, with no vomiting or excessive saliva. Character A then stares at the cup with quiet offended disbelief while lowering it toward the table, ending during the movement with no smile, camera look, victory pose, freeze frame, or fade. Camera language is cinematic macro depth with rack focus, tight profile, wand insert, readable over-the-shoulder magical contact, overhead tea-whirlpool detail, extreme eye close-up, and tight three-quarter reaction, maintaining consistent screen direction and avoiding chaotic camera shake or teleportation. Audio consists of quiet witch-kitchen ambience, wand ignition, a subtle magical hum, liquid swirling, cancellation sound, a small sip, restrained “ptuh,” soft liquid landing, and a disappointed exhale. Music is cozy, mischievous slice-of-life comedy using acoustic guitar, pizzicato strings, muted piano, playful clarinet, and brushed percussion, with no vocals; build gently during the magical stirring, briefly thin the score as the heat registers, then add one understated plucked accent when the tea lands back in the cup. Keep music below the sound effects and avoid boing sounds, circus music, meme sounds, or laugh tracks. Maintain strict causality: FILAMENT CONTACT → TEA STIRS → SPELL ESCALATES → SPELL CANCELLED → TEA STILL → SIP → HEAT REGISTERS → SAME MOUTHFUL SPAT BACK INTO SAME CUP → DEADPAN REGRET. Priority locks are identity, visible magical contact, continuous tea-state consistency, sip before reaction, the same mouthful visibly returning to the same cup, stable facial anatomy, restrained acting, realistic liquid physics, cinematic close-ups, and readable comedic timing. Negative prompt: duplicate character, duplicate cup, duplicate wand, duplicate tea, missing magical contact, magic hitting table or hand, reaction before sipping, new liquid appearing, whole cup spat out, vomiting, excessive saliva, broken mouth anatomy, premature stirring, instant violent whirlpool, large magical explosion, injury, identity drift, exaggerated acting, cartoon physics, anime, cheap CGI, plastic skin, discontinuity, frozen frames, subtitles, text, logo, watermark.",
    "images": [
      "https://content.doitong.com/dev-api/2026/10/21/req_c3dk4hgrd2tzlsymbg986itv/0-6Vz76mAQT0PWAMVCZaZAY_KTBfm8A8.png"
    ],
    "resolution": "480p",
    "duration": "5",
    "aspect_ratio": "16:9",
    "generate_audio": true
  }'

# Still "queued" or "running"? Poll with the id from the response:
curl "https://api.doitong.ru/v1/requests/REQUEST_ID?wait=20" \
  -H "Authorization: Bearer $DOITONG_API_KEY"

运行此请求

你的密钥会保留在此浏览器中:仅发送到 Doitong API;只有勾选“记住”后才会保存到 localStorage。 获取 API 密钥

在 Doitong 上制作

Doitong 社区使用 Seedance 2.0 制作的公开作品。一键复用提示词。

显示的公开作品:16
A close-up portrait of a striking cyberpunk woman with sleek dark hair partially wet and falling ... 视频

A close-up portrait of a striking cyberpunk woman with sleek dark hair partially wet and falling around her face. She’s wearing glossy, high-tech neon visor goggles with a vivid color gradient — bright yellow, orange, and pink reflecting glowing city lights. Her expression is confident and...

Ultra-realistic scene featuring miniature construction workers in orange vests and hard hats, act... 视频

Ultra-realistic scene featuring miniature construction workers in orange vests and hard hats, actively working on a giant [pizza]. The tiny workers are using small tools, cranes, and machines to [assemble]. The [pizza] is placed on a metal surface or kitchen countertop. Steam or particles rise...

Two sandboarders racing down massive golden sand dunes, carving aggressively in wide S-curves, he... 视频

Two sandboarders racing down massive golden sand dunes, carving aggressively in wide S-curves, heading directly toward the camera in a low, dramatic angle. They are captured in mid-motion, kicking up thick trails of sand behind their boards. Both riders wear full protective sandboarding suits in...

A high-fashion cinematic video scene in slow motion, set on a wide white desert with fine rippled... 视频

A high-fashion cinematic video scene in slow motion, set on a wide white desert with fine rippled sand under a deep blue sky. A tall Black male model and a woman walk together in full-body view, directly toward the camera. The male model wears a dramatic flowing red trench coat, matching tailored...

A fashionable young woman with black hair, walking confidently down a modern city street. She is ... 视频

A fashionable young woman with black hair, walking confidently down a modern city street. She is wearing stylish everyday fashion — a black turtleneck top, oversized gray blazer, and wide-leg black trousers with chunky black boots. Her outfit is monochrome, mostly black and gray, with minimalist...

Scene: On a large metal grill surface, miniature construction workers in orange safety vests and ... 视频

Scene: On a large metal grill surface, miniature construction workers in orange safety vests and hard hats are preparing a giant steak. The raw meat, marbled and richly textured, rests across the hot grill. Some workers are spreading seasoning using tiny wheelbarrows filled with salt and pepper,...

Subject: Young woman with closed eyes Context: Surrounded by goldfish swimming in the air Action:... 视频

Subject: Young woman with closed eyes Context: Surrounded by goldfish swimming in the air Action: Portraiture in a dreamy aesthetic Style: Ethereal, artistic, reminiscent of fine art photography Camera motion (optional): Composition (optional): Shallow depth of field Ambiance (optional): Soft...

photorealistic, live-action film, real actress, ARRI Alexa look, desaturated cold tones, soft nat... 视频

photorealistic, live-action film, real actress, ARRI Alexa look, desaturated cold tones, soft natural lighting, subtle dust in air, volumetric haze, consistent city destruction young fragile woman, pale skin, long dark ash-blonde hair, slightly messy, dirty torn gray nightgown, barefoot, natural...

关于 Seedance 2.0 Reference to Video

ByteDance 的多参考视频模型:上传人物、产品、地点、动作或镜头参考,以及语音或音乐片段,然后在提示词中以 @Image1、@Video1、@Audio1 等形式调用。它能在不同剪辑间保持身份一致,并生成同步音频。至少需要一个图像或视频参考,所有参考文件总数不得超过 12 个。输出为 480p 至 4K。

参数

发送规范名称或原生名称之一,不要同时发送两者。 规范名称

名称类型 默认值允许值 描述
prompt必填
类型string
指导视频生成的文本提示词。
image_urls可选 images
类型url[] · image
≤ 9 个项目
用于视频引导的图像参考,最多 9 个图像,且所有模态合计最多 12 个文件。每个图像最多 30 MB,格式必须为 JPEG、PNG 或 WebP。在提示词中使用 @Image1、@Image2 及后续编号标签标识这些参考图像。
video_urls可选 videos
类型url[] · video
≤ 3 个项目
用于生成的参考视频,以 MP4 或 MOV 文件提供,并在提示词中使用 @Video1、@Video2 及后续编号标签标识。每个视频的分辨率必须约为 480p (640x640) 至约 720p (834x1112)。最多 3 个视频,总大小小于 50 MB,总时长为 2–15 秒。
audio_urls可选 audios
类型url[] · audio
≤ 3 个项目
用于视频引导的音频参考至少需要一个配套的图像或视频参考。提供 MP3 或 WAV 文件,每个最多 15 MB,最多 3 个文件,总时长上限为 15 秒。在提示词中使用 @Audio1、@Audio2 及后续编号标签标识。
resolution可选
类型enum
默认值720p
480p720p1080p4k
输出分辨率:480p 优先速度,720p 提供均衡效果,1080p 提供高质量,4k 提供最高质量。
duration可选
类型enum
默认值auto
auto456789101112131415
视频时长选择:`auto` 根据提示词交由模型决定,明确指定的时长必须为 4–15 秒。
aspect_ratio可选
类型enum
默认值auto
auto21:916:94:31:13:49:16
输出宽高比:16:9 为横向,9:16 为纵向,1:1 为正方形,21:9 为电影级超宽屏。选择 `auto` 由模型选择宽高比。
generate_audio可选
类型boolean
默认值true
启用此选项可为视频生成同步音频。生成的音频包括与嘴部动作匹配的语音、环境氛围声和音效。
bitrate_mode可选
类型enum
默认值standard
standardhigh
选择输出编码比特率:'standard' 保留默认设置,'high' 要求模型提高编码质量,但会生成更大的文件。
codec可选
类型enum
默认值auto
autoH264H265
视频编解码器选择:`auto` 保持默认编解码器行为,`H264` 选择 H.264,`H265` 选择 H.265。
必填 规范名称,也可接受

文件参数接受公开的 https:// URL、最大 10 MB 的 data: URI,或上传文件的 file_url。 文件

输出

GET /requests/req_8k2m4q7x1z9c3v5b6n0p2r4t
{
  "id": "req_8k2m4q7x1z9c3v5b6n0p2r4t",
  "object": "request",
  "model": "bytedance/seedance-2.0/reference-to-video",
  "status": "succeeded",
  "created_at": "2026-10-20T12:00:00Z",
  "started_at": "2026-10-20T12:00:02Z",
  "completed_at": "2026-10-20T12:00:09Z",
  "queue_position": null,
  "cost": {
    "credits": 705,
    "usd": "0.705",
    "charged": 705,
    "refunded": 0
  },
  "input": {
    "prompt": "@Image1 The banana manager walks through the office and high-fives the carrot, upbeat comedy tone",
    "images": [
      "https://content.doitong.com/images/gp/image_20260610003801_541d9719_316882.png"
    ],
    "resolution": "720p",
    "duration": 5
  },
  "output": {
    "video": {
      "url": "https://content.doitong.com/dev-api/2026/10/20/req_8k2m4q7x1z9c3v5b6n0p2r4t/0-output.mp4",
      "content_type": "video/mp4",
      "file_name": "0-output.mp4",
      "file_size": 1843200
    },
    "seed": 42
  },
  "outputs": [
    {
      "type": "video",
      "url": "https://content.doitong.com/dev-api/2026/10/20/req_8k2m4q7x1z9c3v5b6n0p2r4t/0-output.mp4",
      "content_type": "video/mp4",
      "file_size": 1843200,
      "width": 1280,
      "height": 720,
      "duration": 5,
      "stored": true
    }
  ],
  "error": null,
  "metrics": {
    "inference_seconds": 6.8
  },
  "urls": {
    "self": "https://api.doitong.ru/v1/requests/req_8k2m4q7x1z9c3v5b6n0p2r4t",
    "cancel": "https://api.doitong.ru/v1/requests/req_8k2m4q7x1z9c3v5b6n0p2r4t/cancel"
  }
}
你的文件,由你选择每个 outputs[].url 都是存储中的副本:可永久保存(按 GB-月付费),也可免费保存 7 天,由账户设置或 ?storage= 决定。
每个模型的结构相同图像、视频和音频均包含 type、url、content_type、width、height 和 duration。output 是模型自身的结果对象。
输出字段
video
object {url, content_type, file_name, file_size}生成得到的视频文件。
seed
integer生成期间应用的种子值。

更多模型:参考图生视频

对于 AI 代理:此页面的 Markdown 版本位于 https://doitong.com/zh/developers/models/bytedance/seedance-2.0/reference-to-video.md. 所有模型和指南: https://doitong.com/developers/llms.txt