# Seedance 2.5 Flash API: from $0.06/s, 30s + Audio | KovAPI

> Seedance 2.5 Flash is the video model ByteDance released in November 2026, the Flash version of Seedance 2.5. Send a prompt, optionally with up to 10 reference images, 5 reference videos and 5 reference audio clips, and get a 4-30 s clip at 480P or 720P with dialogue and sound effects generated with the picture. On KovAPI it is `seedance-2.5-flash`: 720P $0.12/s, 480P $0.06/s, and a 30-second 720P clip costs $3.60. In our November tests 480P took about 2 minutes and 720P 4-8 minutes. There is no 1080P, and images are references, never a first frame.

Web page: https://kovapi.com/en/model/seedance-2.5-flash · Updated: 2026-11-19
Prices, limits and timings are live values from when this file was generated.

## Fact sheet

- Model ID: `seedance-2.5-flash`
- Endpoint: `POST https://kovapi.com/v1/videos/generations` → `GET https://kovapi.com/v1/videos/jobs/{id}`
- Length: 4–30 s, whole seconds
- Resolution: 480P / 720P
- Aspect ratio: 16:9 / 9:16 / 1:1 / 4:3 / 3:4
- References: images ≤10 · videos ≤5 (≤30 s in total) · audio ≤5
- Prompt: ≤ 11,000 characters
- Per second: 480P $0.06/s · 720P $0.12/s
- Per second with a reference video: 480P $0.035/s · 720P $0.07/s
- Results kept: 7 days

## Thirty seconds with sound, from images, video and audio together

- **Up to 30 seconds, shots on a timeline**: Up to 30 s per clip, with shots placed on whole-second timestamps. The guardian robot clip uses a nine-panel storyboard and three design sheets to cover a farewell, a rocket launch, an explosion and the old woman's cry in 30 seconds, with English lines in sync.
- **Eight references, six characters in one scene**: Up to 10 reference images, each with one job: a person, a set, a prop. The family drama uses six character turnarounds and two set images, numbered in order of first appearance, and the ticket window, the counting of the money and the children's cheer all keep the right faces.
- **Let a blockout animation set the motion and camera**: Upload a simple untextured 3D blockout and the model follows its motion and camera while painting in the scene and character from the prompt. This one used a six-second blockout and one line of render notes; the 30-second 3D animation adds nine key frames for each stretch. Reference-video seconds are billed too.
- **Pick the voice; lines and effects come with it**: Every clip has a soundtrack. A reference audio clip sets a character's voice: this one uses a line of Granny Zhou cut from a Seedance 2.5 clip, and the new lines Flash wrote for her keep that voice. Reference audio must come with a reference image or video and is not billed.
- **A product photo plus text cards make a talking ad**: The product photo fixes the look, the selling points go in as text images, and the presenter and voice-over are written in the prompt: one 16-second vertical ad per request. Flash got most of the big captions right but miswrote two characters in one vertical line; if on-screen text must be exact, see the comparison with Seedance 2.5 below.

## Modes and their API fields

All modes share `model` (`seedance-2.5-flash`), `prompt`, `duration`, `resolution` and `size`. Images, videos and audio can go in one request; each kind is numbered in upload order and named in the prompt as `@Image1`, `@Video1`, `@Audio1`.

| Mode | Send | You get | Limits |
| --- | --- | --- | --- |
| Text to video | Only `prompt` and the settings, no references. | Picture, motion and sound from the text; write dialogue, effects and music into the prompt. | Only the general limits (see the specs). |
| Multi-image reference to video | `images` in order; the prompt says what each image is for: a person, a product, a set, a style, a storyboard or a key frame. | New footage that follows the references' look, framing or storyboard; several characters can appear together. | Up to 10 images, 300-6000 px a side, aspect 0.4-2.5. Every image is a reference, never a first frame. |
| Reference video to video | `video_references` with video links (`images` may come too); the prompt says whether to follow the motion, the camera, the pacing or a blockout. | The reference's motion and camera with new characters, sets and style; a 3D blockout can be rendered into finished footage. | Up to 5 videos, 30 s in total; reference-video seconds are billed (see pricing). |
| Reference audio | `audio_reference` with audio links plus a reference image or video; the prompt says whether to follow the voice, the lines or the music. | A character speaks new lines in the reference voice, or the picture follows the reference music's rhythm. | Up to 5 clips, not billed; must come with a reference image or video. |

## What you can make with Seedance 2.5 Flash

- **Product photos to ad videos**: The product photo fixes the look, selling points go in as text images, the presenter and voice-over go in the prompt: one request, one complete ad; billed per second, so making several is cheap.
- **Short drama and dialogue scenes**: Upload character turnarounds and set images separately and give each shot its size, camera and one line of dialogue: good for drama samples and story previs.
- **Game and animation promos**: Feed the logo, characters, props and UI as ordered key frames for a one-take opener or an event teaser.
- **3D blockouts straight to film**: Block out motion and camera with simple shapes, then let the model render the full picture from your prompt: no detailed modelling or rendering.
- **Draft in Flash, finish in 2.5**: Run the same prompt and references through Flash a few times, pick the direction you like, then render the final with `seedance-2.5` (1080P or stricter shots) by changing only `model`.
- **Keep a character's voice across a series**: Cut one line of a character from the last episode as reference audio and add their image: the next episode keeps the voice and the look.

## Pricing

| Resolution | Seedance 2.5 Flash | With a reference video | ByteDance list price |
| --- | --- | --- | --- |
| 480P | $0.06/s | $0.035/s | $0.0672 / with a reference video $0.0413 |
| 720P | $0.12/s | $0.07/s | $0.1512 / with a reference video $0.0929 |

- Without a reference video: the per-second rate of the resolution x output seconds. Reference images and audio are free. Example: 720P for 10 s = 10 x $0.12.
- With a reference video: (reference-video seconds + output seconds) x the "with reference video" rate; every reference video counts. Example: a 10 s 720P clip with a 6 s reference = 16 x $0.07.
- The price is held at submit and refunded in full if the job fails; automatic retries of one job settle once.

## One official prompt: Seedance 2.5 Flash and Seedance 2.5 side by side

Hover to play the clips together; on a phone tap "Play together". Prompt, references, length and ratio are identical, each clip was rendered on its own, so the pictures differ: compare motion, faithfulness and detail. Under each clip is what that clip costs at today's price.

### Text only: a panda cub rolls down a hill (8 s, 16:9)

On a plain text prompt the two are close: Flash 720P rolls more playfully, Seedance 2.5 shows a little more fur detail; Flash 480P keeps the framing and motion but turns soft when enlarged -- fine for a first look.

- 8s · 16:9
- **Seedance 2.5 Flash 720P** ($0.96 at current prices): https://img.kovapi.com/i/mp-7913126fdea63ecd512cdc21.mp4 — Flash 720P: the roll, the sit-up and the look to camera are all there, at about three quarters of the 2.5 rate.
- **Seedance 2.5 720P** ($1.264 at current prices): https://img.kovapi.com/i/mp-ed231c10ec7c24e490b6cceb.mp4 — Seedance 2.5 720P: calmer motion, slightly more fur and light detail.
- **Seedance 2.5 Flash 480P** ($0.48 at current prices): https://img.kovapi.com/i/mp-845124c7cefabca29f45e306.mp4 — Flash 480P: half the 720P price, good for drafts in bulk.

Original prompt:

```text
写实自然纪录片风格，电影级真实光影，温暖的午后, 在森林草坡上, 一只圆滚滚的熊猫幼崽从坡上滚下来.

熊猫黑白毛发蓬松真实，体型小而胖，动作笨拙可爱。场景为绿色森林斜坡，地面覆盖青草、苔藓、三叶草、泥土、小石块、枯枝和少量小黄花，虚化背景中有高大树干与树林。镜头为低机位中远景，轻微手持感，整体基本固定，始终保持熊猫在画面中。

0s-3s：一只熊猫幼崽趴在绿色草坡上，身体圆滚滚, 它刚开始顺着斜坡慢慢侧滚，动作笨拙，草叶被身体轻轻压弯。微风拂过, 阳光从左上方穿过树林，形成斑驳光影。
3s-8s：熊猫滚到画面右下方，动作逐渐停止，从侧躺变成趴卧。它的圆脸朝向镜头，前爪压在草地上，熊猫趴在前景草丛中，身体调整到舒服的姿势，头部小幅抬起又放低, 发出轻微哼唧声。

低机位，轻微手持感，轻微跟随熊猫向右下方移动。景深自然，前景草叶轻微虚化，熊猫主体清晰，背景树林柔和虚化。自然环境音，风声、熊猫滚动时轻微柔软的扑通声, 整体温暖、真实、自然。
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
Realistic nature-documentary style, cinematic natural light, a warm afternoon: on a grassy forest slope, a chubby panda cub rolls down the hill.

The cub's black-and-white fur is fluffy and real; it is small and plump, clumsy and cute. The slope is green, covered with grass, moss, clover, soil, small stones, dry twigs and a few small yellow flowers, with tall trunks and woods in the blurred background. Low-angle medium-long shot with a slight handheld feel, mostly static, always keeping the panda in frame.

0s-3s: the cub lies on the green slope, round as a ball, and starts to roll slowly sideways down it, clumsily, gently pressing the grass flat. A breeze passes; sunlight comes through the trees from the upper left in dappled patches.
3s-8s: it rolls to the lower right of the frame and comes to a stop, turning from its side onto its belly. Its round face turns to the camera, front paws on the grass; lying in the foreground grass it settles in, lifts its head a little and lowers it again, with a soft grunt.

Low angle, slight handheld feel, gently following the panda to the lower right. Natural depth of field: foreground grass slightly soft, the panda sharp, the woods softly blurred. Natural ambience: wind and the soft thuds of the rolling panda; warm, real and natural throughout.
```

### A talking product ad: one product photo plus three text cards (16 s, 3:4)

Both drew the tyre from the photo and followed the voice-over and shots in the prompt. The difference is the on-screen text: Seedance 2.5 got all four selling-point captions right; Flash wrote one vertical line wrong (two characters) and its reading order is a little off.

- 16s · 3:4
- **Seedance 2.5 Flash 720P** ($1.92 at current prices): https://img.kovapi.com/i/mp-2dcaacdcc4f2451c5f1b45aa.mp4 — Flash: presenter, tyre and voice-over right; one vertical caption wrong.
- **Seedance 2.5 720P** ($2.528 at current prices): https://img.kovapi.com/i/mp-87927e076c7f1080f771dab5.mp4 — Seedance 2.5: all four captions right, cleaner layout.

Original prompt:

```text
为图1中的商品生成视频，由中国中年自然肤色黑色寸头日常户外风男模特出镜，手持商品进行动态展示与解说。镜头 1，镜头从中景缓慢推进至近景。阳光明媚的广阔户外公路旁，一位中国中年、自然肤色、黑色寸头、穿着日常户外冲锋衣的男模特，一手扶着立在地上的大品牌轮胎，面向镜头自信地进行解说。 男主播：“大品牌轮胎，伴你安心出行。” 背景音乐：节奏轻快的户外运动风纯音乐。 画面正下方显示粗体黑字“大品牌轮胎 伴你安心出行” 参考图2。镜头 2，镜头特写轮胎正面花纹。男模特的手指顺着轮胎上深深的纵向排水沟槽滑动，随后手掌有力地拍打胎面，动态展示轮胎清晰的纹理和扎实的质感。 男主播：“强效排水，稳健抓地。” 背景音乐持续，伴随轻微的摩擦音效。 画面左侧弹出黑体大字“强效排水 稳健抓地” 参考图3。镜头 3，镜头平移转至轮胎侧面特写。男模特用手掌按压轮胎侧壁展示其优良的韧性与弹性，随后手指指向侧面凸起的品牌标识区域。 男主播：“静音舒适，坚韧耐用。” 背景音乐持续。 画面右侧弹出黑体大字“静音舒适 坚韧耐用” 参考图4。镜头 4，镜头拉远回到中景。男模特双手扶住轮胎边缘，轻松地将其向前滚动半圈，随后站定，微笑着向镜头竖起大拇指，背景是充满活力的自然风光。 画外音：“你的安心之选。” 音乐高潮处有力收尾。 画面中央浮现黑体大字“伴你安心出行” 参考图2
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
Make a video of the product in image 1: a Chinese middle-aged male model with natural skin, short black hair and everyday outdoor clothes presents and explains it on camera, product in hand. Shot 1: the camera slowly pushes from a medium shot to a close shot. Beside a wide outdoor road on a sunny day, the model, in an everyday outdoor shell jacket, rests one hand on a big-brand tire standing on the ground and speaks to the camera with confidence. Presenter: "大品牌轮胎，伴你安心出行。" (A big-brand tire for worry-free trips.) Music: light, upbeat outdoor-sports instrumental. Bold black text at the bottom centre: "大品牌轮胎 伴你安心出行" — see image 2. Shot 2: close-up of the tread. His fingers run along the deep lengthwise drainage grooves, then his palm slaps the tread firmly, showing the clear pattern and solid build. Presenter: "强效排水，稳健抓地。" (Strong drainage, steady grip.) Music continues with a light rubbing sound. Big black text pops up on the left: "强效排水 稳健抓地" — see image 3. Shot 3: the camera pans to a close-up of the sidewall. He presses the sidewall with his palm to show its toughness and spring, then points at the raised brand area. Presenter: "静音舒适，坚韧耐用。" (Quiet and comfortable, tough and lasting.) Music continues. Big black text pops up on the right: "静音舒适 坚韧耐用" — see image 4. Shot 4: the camera pulls back to a medium shot. He holds the tire's edge with both hands, rolls it forward half a turn with ease, stands still and gives the camera a smiling thumbs-up against lively scenery. Voice-over: "你的安心之选。" (Your worry-free choice.) The music ends on a strong climax. Big black text appears in the centre: "伴你安心出行" — see image 2.
```

### Six key frames in one take: pixel-art martial arts (15 s, 9:16)

The more key frames that must be matched one by one, the clearer the gap: Seedance 2.5 hit the logo, the pixel scatter, the martial-arts manual, the hero and the event screen in order; Flash skipped the manual and put in a close-up of the hero instead, keeping the rest in order.

- 15s · 9:16
- **Seedance 2.5 Flash 720P** ($1.80 at current prices): https://img.kovapi.com/i/mp-2ee2d1750be8ebe3789d15e6.mp4 — Flash: one take and the pixel style are right; one key frame unused.
- **Seedance 2.5 720P** ($2.37 at current prices): https://img.kovapi.com/i/mp-d190ceb69b8b097b61b2e3b3.mp4 — Seedance 2.5: all six key frames matched.

Original prompt:

```text
根据 @图片1 - @图片6 制作一镜到底的像素武侠主题竖版视频，背景音乐使用国风 8bit 武侠风格音乐。全片统一浅蓝底色，像素风格统一，画面干净通透。
镜头 1：静止展示 @图片 1 的「江湖风云」水墨风格 logo，背景为统一浅蓝底色，画面保持约 1 秒静止。
镜头 2：@图片 1 中的文字区域消失后，@图片 2 的武侠男性角色像素脸部特写从画面底部滑入；角色眨眨眼看向镜头，随后快速向下位移离场；角色离场后，原本 logo 处变为 @图片 3 的蓝色「武功秘籍」像素书。
镜头 3：紧接 @图片 4，像素武侠小人角色从画面下方用力向上跃出，顶起上方的蓝色菱形问号标，问号标上方弹出深蓝色粗体字「今日闯江湖！」；角色落地后摆出 @图片 4 的站姿 pose，随后抬手打招呼，接着做出预备跑动作，转身向画面右侧奔跑（奔跑姿态参考 @图片 5），镜头跟随角色向右移动，角色从画面右侧跳出画面。
镜头 4：@图片 6 的 UI 界面从画面右侧位移入画，像素武侠角色从画面右上角跳入，落在「三月廿七日」大字的右下方，张开双臂摆出热情展示的定格姿势，最终定格在该画面。
整体像素武侠美术风格，色调统一浅蓝底，运镜连贯顺滑呈现一镜到底的连续位移与跟随，元素过渡自然、角色动作衔接流畅，画面无卡顿、无闪烁；文字与 UI 清晰稳定。
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
Using @Image1 - @Image6, make a one-take vertical video on a pixel-art wuxia theme, with Chinese-style 8-bit wuxia music. The same light-blue background throughout, one pixel style, clean and clear.
Shot 1: a still of the ink-style "江湖风云" logo from @Image 1 on the light-blue background, held for about 1 second.
Shot 2: the text of @Image 1 disappears and the pixel face close-up of the male wuxia hero from @Image 2 slides in from the bottom; he blinks, looks at the camera and quickly drops out of frame; the logo's place becomes the blue pixel book "武功秘籍" (martial-arts manual) from @Image 3.
Shot 3: straight into @Image 4: the little pixel hero leaps up from the bottom and bumps the blue diamond question-mark block above, which pops up the dark-blue bold words "今日闯江湖！" (into the jianghu today!); he lands in the standing pose of @Image 4, waves, crouches to run, turns and runs to the right (running pose as in @Image 5); the camera follows him right and he jumps out of frame on the right.
Shot 4: the UI screen of @Image 6 slides in from the right; the hero jumps in from the upper right, lands to the lower right of the big date "三月廿七日" (the 27th of the third month), throws his arms open in an eager presenting pose and freezes there.
Pixel wuxia art throughout on a uniform light-blue background, smooth continuous moves and follows in one take, natural transitions between elements, fluid character motion, no stutter or flicker; text and UI sharp and stable.
```

## Examples: prompts and parameters

### Guardian robot: a nine-panel storyboard shot as 30 seconds

- Made with KovAPI · Multi-image reference · 30s · 720P · 16:9
- $3.60 at current prices
- Output: https://img.kovapi.com/i/mp-4308bb84eb3bff4b6c0e2f69.mp4
- Image 1: https://img.kovapi.com/i/mp-467106a4c8d7cb83c354cc07.png
- Image 2: https://img.kovapi.com/i/mp-cf25b8d66e978ef5855cda31.png
- Image 3: https://img.kovapi.com/i/mp-d8e099cda9738ccccfa1bb5a.png
- Image 4: https://img.kovapi.com/i/mp-e01a7bd8d09d396d64099c9b.png

A nine-panel line storyboard sets the shot structure, a launch-site photo the palette, two design sheets the looks; 30 s cover the farewell, the launch and the explosion with English lines in sync. Flash gave the old woman more close-ups; the robot appears only at the start. The prompt and references come from the Volcengine Ark [Seedance 2.5 prompt guide](https://docs.byteplus.com/en/docs/ModelArk/seedance-2-5-prompt-guide) (Flash takes the same prompts as 2.5); KovAPI rendered it with Seedance 2.5 Flash at the original settings.

Original prompt:

```text
image1: 九宫格分镜参考，用于整体镜头结构、景别与运镜节奏。
image2: 火箭发射场黄昏草原实拍参考，用于环境构图、暖金夕照与冷暮蓝的写实彩色实拍质感基准。
image3: 主体 1 (守护机器人) 角色外观参考。
image4: 主体 2 (老奶奶) 角色外观参考。
【主体设定】
主体 1 (守护机器人): 参考 image3，近未来做旧复古机器人，做旧蓝绿色金属机身、斑驳锈蚀，圆顶头，两只发光的红色圆形机眼，细天线，细长关节四肢；体型高大，约为人类两倍高。
主体 2 (老奶奶): 参考 image4，瘦小年迈女性，银发挽成低髻，皱纹深刻，身穿明黄金色及地长裙、缀金蓝刺绣胸襟，神情不舍；身高只及机器人胸口。
环境 (黄昏草原・发射场): 参考 image2，近未来黄昏草原，暮色天空由暖金渐入冷蓝，远处地平线一座发射台矗立白色火箭、蒸汽升腾；及膝野草随风起伏，广袤空旷。
【整体风格】
真人实拍彩色电影正片，写实照片级质感，全程真实彩色画面；彩色 35mm 电影胶片质感，细腻真实胶片颗粒，浓郁饱满的电影级调色，IMAX 大画幅质感；手持摄影，呼吸感摇晃，浅景深大光圈，前景持续飘散的草叶、火星与灰烬，微微倾斜荷兰角，暖金夕照与冷暮蓝、爆炸暖橙强烈对撞，16:9 横屏。近未来温情灾难片氛围，静谧、悲壮、守护与不舍。
【严格排除】黑白、单色、灰度、去色；手绘、素描、线稿、插画、漫画、动画；分镜稿 / 故事板、草图；移轴微缩、玩偶感、塑料 CG、油腻过曝 CG。
【分镜头】(9 镜 ≈ 30s)
镜头 1 (0-3s): 超远景低机位仰视，黄昏草原两人背影仰望远方火箭。台词 (机器人): "I'm right here. I won't let go."
镜头 2 (3-6s): 正面中景手持，机器人搀扶老奶奶。
镜头 3 (6-10s): 面部特写，老奶奶眼含不舍。台词 (老奶奶): "Fly safe, my child. Come back to me."
镜头 4 (10-14s): 超远景仰摇，火箭升空拖曳浓白烟迹。台词 (老奶奶): "There he goes... there he goes."
镜头 5 (14-18s): 超远景，火箭在半空爆炸碎裂。台词 (老奶奶): "No... no, no—"
镜头 6 (18-22s): 面部大特写，老奶奶瞳孔骤缩泪滑落。台词 (老奶奶): "...he was almost there."
镜头 7 (22-25s): 特写转近景，老奶奶崩溃痛哭。台词 (老奶奶): "Bring him back! Please— bring him back!"
镜头 8 (25-28s): 超低视角近乎垂直仰拍，机器人环抱老奶奶罩成保护穹顶。台词 (机器人): "Don't look up. I've got you."
镜头 9 (28-30s): 超远景背影，两人紧紧相拥剪影。台词 (机器人): "I'm still here. I'll stay... as long as you need."
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
image1: nine-panel storyboard reference, for the overall shot structure, shot sizes and camera rhythm.
image2: live-action reference of a rocket launch site on a grassland at dusk, the baseline for composition and the realistic colour of warm golden sunset against cold dusk blue.
image3: appearance reference for subject 1 (the guardian robot).
image4: appearance reference for subject 2 (the old woman).
[Subjects]
Subject 1 (guardian robot): see image3. A weathered retro robot of the near future: worn blue-green metal body with patchy rust, a domed head, two glowing round red eyes, a thin antenna, long thin jointed limbs; very tall, about twice a person's height.
Subject 2 (old woman): see image4. A small, frail elderly woman, silver hair in a low bun, deep wrinkles, a floor-length bright golden dress with a gold-and-blue embroidered bodice, reluctant to part; she reaches only to the robot's chest.
Setting (dusk grassland, launch site): see image2. A near-future grassland at dusk, the sky shifting from warm gold to cold blue; on the far horizon a launch pad holds a white rocket with steam rising; knee-high grass sways in the wind, wide and empty.
[Overall style]
Live-action colour feature film, photorealistic, real colour throughout; colour 35mm film look with fine, real grain, rich cinematic grading, IMAX large-format feel; handheld with a breathing sway, shallow focus at a wide aperture, grass, sparks and ash drifting through the foreground, a slight Dutch tilt, warm gold sunset clashing with cold dusk blue and the warm orange of an explosion, 16:9 landscape. A tender near-future disaster film: quiet, tragic, protective, reluctant to part.
[Strictly exclude] black and white, monochrome, greyscale, desaturation; hand drawing, sketch, line art, illustration, comics, animation; storyboard drafts, rough sketches; tilt-shift miniature, toy look, plastic CG, greasy overexposed CG.
[Shots] (9 shots ≈ 30s)
Shot 1 (0-3s): extreme long shot from a low angle looking up; the two seen from behind on the dusk grassland, gazing at the distant rocket. Line (robot): "I'm right here. I won't let go."
Shot 2 (3-6s): frontal medium shot, handheld; the robot supports the old woman.
Shot 3 (6-10s): close-up of her face, eyes full of reluctance. Line (old woman): "Fly safe, my child. Come back to me."
Shot 4 (10-14s): extreme long shot tilting up; the rocket climbs on a thick trail of white smoke. Line (old woman): "There he goes... there he goes."
Shot 5 (14-18s): extreme long shot; the rocket explodes and breaks apart in mid-air. Line (old woman): "No... no, no—"
Shot 6 (18-22s): extreme close-up; her pupils contract and a tear runs down. Line (old woman): "...he was almost there."
Shot 7 (22-25s): close-up widening to a close shot; she breaks down sobbing. Line (old woman): "Bring him back! Please— bring him back!"
Shot 8 (25-28s): extreme low angle, almost vertical, looking up; the robot wraps its arms around her like a protective dome. Line (robot): "Don't look up. I've got you."
Shot 9 (28-30s): extreme long shot from behind; the two hold each other tight in silhouette. Line (robot): "I'm still here. I'll stay... as long as you need."
```

### Tyre ad: a product photo plus three text cards

- Made with KovAPI · Multi-image reference · 16s · 720P · 3:4
- $1.92 at current prices
- Output: https://img.kovapi.com/i/mp-2dcaacdcc4f2451c5f1b45aa.mp4
- Image 1: https://img.kovapi.com/i/mp-6d18a3f328848a8401c1d37c.png
- Image 2: https://img.kovapi.com/i/mp-e80747cc976af420c832baff.png
- Image 3: https://img.kovapi.com/i/mp-b7de1cac42d0cf89e63906d7.png
- Image 4: https://img.kovapi.com/i/mp-5b00ca03b6560e4b93f7390f.png

The product photo fixes the tyre, three text images keep the big captions right, and four shots carry a Chinese voice-over. Flash got most captions right but wrote one vertical line wrong. The prompt and references come from the Volcengine Ark [Seedance 2.5 prompt guide](https://docs.byteplus.com/en/docs/ModelArk/seedance-2-5-prompt-guide) (Flash takes the same prompts as 2.5); KovAPI rendered it with Seedance 2.5 Flash at the original settings.

Original prompt:

```text
为图1中的商品生成视频，由中国中年自然肤色黑色寸头日常户外风男模特出镜，手持商品进行动态展示与解说。镜头 1，镜头从中景缓慢推进至近景。阳光明媚的广阔户外公路旁，一位中国中年、自然肤色、黑色寸头、穿着日常户外冲锋衣的男模特，一手扶着立在地上的大品牌轮胎，面向镜头自信地进行解说。 男主播：“大品牌轮胎，伴你安心出行。” 背景音乐：节奏轻快的户外运动风纯音乐。 画面正下方显示粗体黑字“大品牌轮胎 伴你安心出行” 参考图2。镜头 2，镜头特写轮胎正面花纹。男模特的手指顺着轮胎上深深的纵向排水沟槽滑动，随后手掌有力地拍打胎面，动态展示轮胎清晰的纹理和扎实的质感。 男主播：“强效排水，稳健抓地。” 背景音乐持续，伴随轻微的摩擦音效。 画面左侧弹出黑体大字“强效排水 稳健抓地” 参考图3。镜头 3，镜头平移转至轮胎侧面特写。男模特用手掌按压轮胎侧壁展示其优良的韧性与弹性，随后手指指向侧面凸起的品牌标识区域。 男主播：“静音舒适，坚韧耐用。” 背景音乐持续。 画面右侧弹出黑体大字“静音舒适 坚韧耐用” 参考图4。镜头 4，镜头拉远回到中景。男模特双手扶住轮胎边缘，轻松地将其向前滚动半圈，随后站定，微笑着向镜头竖起大拇指，背景是充满活力的自然风光。 画外音：“你的安心之选。” 音乐高潮处有力收尾。 画面中央浮现黑体大字“伴你安心出行” 参考图2
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
Make a video of the product in image 1: a Chinese middle-aged male model with natural skin, short black hair and everyday outdoor clothes presents and explains it on camera, product in hand. Shot 1: the camera slowly pushes from a medium shot to a close shot. Beside a wide outdoor road on a sunny day, the model, in an everyday outdoor shell jacket, rests one hand on a big-brand tire standing on the ground and speaks to the camera with confidence. Presenter: "大品牌轮胎，伴你安心出行。" (A big-brand tire for worry-free trips.) Music: light, upbeat outdoor-sports instrumental. Bold black text at the bottom centre: "大品牌轮胎 伴你安心出行" — see image 2. Shot 2: close-up of the tread. His fingers run along the deep lengthwise drainage grooves, then his palm slaps the tread firmly, showing the clear pattern and solid build. Presenter: "强效排水，稳健抓地。" (Strong drainage, steady grip.) Music continues with a light rubbing sound. Big black text pops up on the left: "强效排水 稳健抓地" — see image 3. Shot 3: the camera pans to a close-up of the sidewall. He presses the sidewall with his palm to show its toughness and spring, then points at the raised brand area. Presenter: "静音舒适，坚韧耐用。" (Quiet and comfortable, tough and lasting.) Music continues. Big black text pops up on the right: "静音舒适 坚韧耐用" — see image 4. Shot 4: the camera pulls back to a medium shot. He holds the tire's edge with both hands, rolls it forward half a turn with ease, stands still and gives the camera a smiling thumbs-up against lively scenery. Voice-over: "你的安心之选。" (Your worry-free choice.) The music ends on a strong climax. Big black text appears in the centre: "伴你安心出行" — see image 2.
```

### Pixel martial arts: six key frames in one take

- Made with KovAPI · Multi-image reference · 15s · 720P · 9:16
- $1.80 at current prices
- Output: https://img.kovapi.com/i/mp-2ee2d1750be8ebe3789d15e6.mp4
- Image 1: https://img.kovapi.com/i/mp-a8e6cafd16699c88bc205bde.png
- Image 2: https://img.kovapi.com/i/mp-d6a179f9802cc2abf2a287ff.png
- Image 3: https://img.kovapi.com/i/mp-c83b76a690b59f2c21e6ff58.png
- Image 4: https://img.kovapi.com/i/mp-5a28f9e51d489941e140610e.png
- Image 5: https://img.kovapi.com/i/mp-92c37e2d346d6d6316d76c50.png
- Image 6: https://img.kovapi.com/i/mp-7e7ee924787da67dc9b2d086.png

Six images placed shot by shot: logo, hero, manual, leap, run, event screen; a 15 s vertical one-take with 8-bit Chinese-style music. Flash skipped the martial-arts manual and kept the rest in order. The prompt and references come from the Volcengine Ark [Seedance 2.5 prompt guide](https://docs.byteplus.com/en/docs/ModelArk/seedance-2-5-prompt-guide) (Flash takes the same prompts as 2.5); KovAPI rendered it with Seedance 2.5 Flash at the original settings.

Original prompt:

```text
根据 @图片1 - @图片6 制作一镜到底的像素武侠主题竖版视频，背景音乐使用国风 8bit 武侠风格音乐。全片统一浅蓝底色，像素风格统一，画面干净通透。
镜头 1：静止展示 @图片 1 的「江湖风云」水墨风格 logo，背景为统一浅蓝底色，画面保持约 1 秒静止。
镜头 2：@图片 1 中的文字区域消失后，@图片 2 的武侠男性角色像素脸部特写从画面底部滑入；角色眨眨眼看向镜头，随后快速向下位移离场；角色离场后，原本 logo 处变为 @图片 3 的蓝色「武功秘籍」像素书。
镜头 3：紧接 @图片 4，像素武侠小人角色从画面下方用力向上跃出，顶起上方的蓝色菱形问号标，问号标上方弹出深蓝色粗体字「今日闯江湖！」；角色落地后摆出 @图片 4 的站姿 pose，随后抬手打招呼，接着做出预备跑动作，转身向画面右侧奔跑（奔跑姿态参考 @图片 5），镜头跟随角色向右移动，角色从画面右侧跳出画面。
镜头 4：@图片 6 的 UI 界面从画面右侧位移入画，像素武侠角色从画面右上角跳入，落在「三月廿七日」大字的右下方，张开双臂摆出热情展示的定格姿势，最终定格在该画面。
整体像素武侠美术风格，色调统一浅蓝底，运镜连贯顺滑呈现一镜到底的连续位移与跟随，元素过渡自然、角色动作衔接流畅，画面无卡顿、无闪烁；文字与 UI 清晰稳定。
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
Using @Image1 - @Image6, make a one-take vertical video on a pixel-art wuxia theme, with Chinese-style 8-bit wuxia music. The same light-blue background throughout, one pixel style, clean and clear.
Shot 1: a still of the ink-style "江湖风云" logo from @Image 1 on the light-blue background, held for about 1 second.
Shot 2: the text of @Image 1 disappears and the pixel face close-up of the male wuxia hero from @Image 2 slides in from the bottom; he blinks, looks at the camera and quickly drops out of frame; the logo's place becomes the blue pixel book "武功秘籍" (martial-arts manual) from @Image 3.
Shot 3: straight into @Image 4: the little pixel hero leaps up from the bottom and bumps the blue diamond question-mark block above, which pops up the dark-blue bold words "今日闯江湖！" (into the jianghu today!); he lands in the standing pose of @Image 4, waves, crouches to run, turns and runs to the right (running pose as in @Image 5); the camera follows him right and he jumps out of frame on the right.
Shot 4: the UI screen of @Image 6 slides in from the right; the hero jumps in from the upper right, lands to the lower right of the big date "三月廿七日" (the 27th of the third month), throws his arms open in an eager presenting pose and freezes there.
Pixel wuxia art throughout on a uniform light-blue background, smooth continuous moves and follows in one take, natural transitions between elements, fluid character motion, no stutter or flicker; text and UI sharp and stable.
```

### A family drama with six characters and two sets

- Made with KovAPI · Multi-image reference · 15s · 720P · 9:16
- $1.80 at current prices
- Output: https://img.kovapi.com/i/mp-33fa6ecbde27d0321307a8da.mp4
- Image 1: https://img.kovapi.com/i/mp-a6792abed7676768d2215f17.png
- Image 2: https://img.kovapi.com/i/mp-3ffc1f8f1d6e10f72f30876f.png
- Image 3: https://img.kovapi.com/i/mp-a597f359b98b6c84afff0b6a.png
- Image 4: https://img.kovapi.com/i/mp-1aa87318ba2ffa0e33a2f6f8.png
- Image 5: https://img.kovapi.com/i/mp-3a469285e81f868bd933d15c.png
- Image 6: https://img.kovapi.com/i/mp-6e5a0b7277826cf522ebc9dd.png
- Image 7: https://img.kovapi.com/i/mp-17786d9e48b89fa348215003.png
- Image 8: https://img.kovapi.com/i/mp-2ad7ea9eb1362c79e24ea40f.png

Six character turnarounds and two set images, numbered in order of first appearance; each shot has its size, camera and one line. The ticket window, the counting and the children's cheer all keep the right faces. The prompt and references come from the Volcengine Ark [Seedance 2.5 prompt guide](https://docs.byteplus.com/en/docs/ModelArk/seedance-2-5-prompt-guide) (Flash takes the same prompts as 2.5); KovAPI rendered it with Seedance 2.5 Flash at the original settings.

Original prompt:

```text
【统一风格与约束】 真实材质质感，细腻光影；角色面部和身体比例稳定，肢体结构自然，动作连贯，不僵硬，无穿模，无卡顿，无扭曲肢体，动作符合现实物理逻辑；禁止生成同款分身、双胞胎效果，同一画面中仅保留单个对应角色，不出现角色重复复刻；无字幕，无水印，无 logo，无乱码，无无关文字；无多余背景音乐。
【参考素材】 将图片 1 中的角色定义为秦秀莲； 将图片 3 中的角色定义为刘春香； 将图片 2 中的角色定义为王强； 将图片 4 中的角色定义为小乐； 将图片 5 中的角色定义为小石头； 将图片 6 中的角色定义为二妞； 将图片 7 定义为县城火车站售票处的参考场景； 将图片 8 定义为刘春香家卧室的参考场景； 将纸质车票定义为去大城的火车票。
镜头 1｜0 秒 - 2 秒
以人物为中心构图，近景，平视，正面机位，日、内，固定。 县城火车站售票处。 秦秀莲站在售票窗口前看向售票窗口。 秦秀莲说：“一张去大城的票。”
镜头 2｜2 秒 - 4 秒
局部构图，特写，平视，正面机位，日、内，固定。 县城火车站售票处。 秦秀莲握紧去羊城的火车票，低头看向车票目的地，嘴唇闭合。 秦秀莲（V.O.）说：“刘春香能做，我也能做。”
镜头 3｜4 秒 - 5 秒
三分法构图，中景，平视，斜 45 度侧面机位，夜、内，固定。 刘春香家卧室。 刘春香和王强坐在桌边清点当天收入。 刘春香看向桌上钱币，说：“七百多。”
镜头 4｜5 秒 - 7 秒
以人物为中心构图，近景，平视，正面机位，夜、内，固定。 刘春香家卧室。 王强看向桌上钱币，说：“比开业那天还多。”
镜头 5｜7 秒 - 9 秒
以人物为中心构图，近景，平视，正面机位，夜、内，固定。 刘春香家卧室。 小乐爬到刘春香身边，看向刘春香，说：“麻麻，我给你捶背！”
镜头 6｜9 秒 - 10 秒
以人物为中心构图，近景，平视，正面机位，夜、内，固定。 刘春香家卧室。 小石头看向刘春香，说：“我也来。”
镜头 7｜10 秒 - 11 秒
以人物为中心构图，近景，平视，正面机位，夜、内，固定。 刘春香家卧室。 二妞看向刘春香，说：“还有我！”
镜头 8｜11 秒 - 13 秒
对称构图，中景，平视，侧面机位，夜、内，固定。 刘春香家卧室。 小乐、小石头、二妞围着刘春香捶肩捶背。 刘春香悄悄握住王强的手，王强反握刘春香的手。
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
[Style and constraints] Real material textures, fine lighting; stable facial and body proportions, natural limbs, continuous motion, nothing stiff, no clipping, no stutter, no twisted limbs, physically plausible movement; no clones or twins of a character, only one of each character per frame, no repeated characters; no subtitles, watermarks, logos, garbled or stray text; no extra background music.
[References] The character in image 1 is Qin Xiulian; image 3 is Liu Chunxiang; image 2 is Wang Qiang; image 4 is Xiaole; image 5 is Xiao Shitou; image 6 is Erniu; image 7 is the ticket hall of the county-town railway station; image 8 is Liu Chunxiang's bedroom; a paper ticket is the train ticket to Dacheng.
Shot 1 | 0 - 2 s
Character-centred, close shot, eye level, frontal, day, interior, locked off. The station ticket hall. Qin Xiulian stands at the ticket window looking in. Qin Xiulian: "一张去大城的票。" (One ticket to Dacheng.)
Shot 2 | 2 - 4 s
Detail framing, close-up, eye level, frontal, day, interior, locked off. The ticket hall. Qin Xiulian grips the ticket to Yangcheng and looks down at the destination, lips closed. Qin Xiulian (V.O.): "刘春香能做，我也能做。" (If Liu Chunxiang can do it, so can I.)
Shot 3 | 4 - 5 s
Rule-of-thirds, medium shot, eye level, 45-degree side angle, night, interior, locked off. Liu Chunxiang's bedroom. Liu Chunxiang and Wang Qiang sit at the table counting the day's takings. Liu Chunxiang, looking at the money: "七百多。" (Over seven hundred.)
Shot 4 | 5 - 7 s
Character-centred, close shot, eye level, frontal, night, interior, locked off. The bedroom. Wang Qiang, looking at the money: "比开业那天还多。" (More than on opening day.)
Shot 5 | 7 - 9 s
Character-centred, close shot, eye level, frontal, night, interior, locked off. The bedroom. Xiaole climbs up beside Liu Chunxiang and looks at her: "麻麻，我给你捶背！" (Mummy, let me rub your back!)
Shot 6 | 9 - 10 s
Character-centred, close shot, eye level, frontal, night, interior, locked off. The bedroom. Xiao Shitou looks at Liu Chunxiang: "我也来。" (Me too.)
Shot 7 | 10 - 11 s
Character-centred, close shot, eye level, frontal, night, interior, locked off. The bedroom. Erniu looks at Liu Chunxiang: "还有我！" (And me!)
Shot 8 | 11 - 13 s
Symmetrical, medium shot, eye level, side angle, night, interior, locked off. The bedroom. Xiaole, Xiao Shitou and Erniu gather round Liu Chunxiang, rubbing her shoulders and back. Liu Chunxiang quietly takes Wang Qiang's hand, and he holds hers back.
```

### Blockout plus key frames: a 30-second 3D animation

- Made with KovAPI · Video reference · 30s · 720P · 16:9
- $4.20 at current prices
- Output: https://img.kovapi.com/i/mp-0bd41c39a5f1b407e060bb64.mp4
- Image 1: https://img.kovapi.com/i/mp-634ba490a460e269825e6f90.png
- Image 2: https://img.kovapi.com/i/mp-a51be9eda6947454338b6143.png
- Image 3: https://img.kovapi.com/i/mp-df1d64f97bbc630b33231d4e.png
- Image 4: https://img.kovapi.com/i/mp-00aea6106ad6fd195fae19d7.png
- Image 5: https://img.kovapi.com/i/mp-5eebe1628e0218cd71078837.png
- Image 6: https://img.kovapi.com/i/mp-d51ce51f1486ca0898e35865.png
- Image 7: https://img.kovapi.com/i/mp-12cb5d92cda9c1480119a4ef.png
- Image 8: https://img.kovapi.com/i/mp-a3675d2996c5a522a748a49d.png
- Image 9: https://img.kovapi.com/i/mp-8fb79f42e2a9f97c2ccbe27c.png
- Reference video (30s): https://img.kovapi.com/i/mp-c0ff18bf0ffcbccf66defe15.mp4

A 30 s blockout video drives the camera and character animation, nine key frames set the picture of each stretch, matched on the timeline. The 30 s of reference video are billed too. The prompt and references come from the Volcengine Ark [Seedance 2.5 prompt guide](https://docs.byteplus.com/en/docs/ModelArk/seedance-2-5-prompt-guide) (Flash takes the same prompts as 2.5); KovAPI rendered it with Seedance 2.5 Flash at the original settings.

Original prompt:

```text
以白模参考视频 <video1> 作为整支视频唯一的运镜、镜头节奏、景别变化、主体运动轨迹和镜头调度参考，严格保持白模视频的镜头顺序、机位变化、运动方式和节奏，不改变镜头结构，不新增镜头，不改变主体运动逻辑。结合各阶段关键帧参考图，生成一部 30 秒电影级 3D 动画短片，整体风格梦幻、童话、温暖，具有儿童幻想色彩，角色外形与各阶段的关键帧保持一致，不要改变角色形象，人物表情情绪随场景变化而改变。
0-3s（首帧参考 <2pic>）镜头由全景俯视缓缓推向地上的小女孩，小女孩坐在房间的地毯上玩飞机，小女孩起身左转，右手使劲一挥将手里的飞机放飞。
3-5s（参考 <3pic>）飞机从左至右穿过房间悬挂的星星挂件，小女孩乘坐飞机进入幻想的天空。
5-8s（参考 <4pic>）镜头继续侧跟，以小女孩为中心的环绕。
8-10s（参考 <5pic>）镜头环绕到小女孩驾驶的飞机背面、飞机开始缓缓俯冲海面。
10-19s（参考 <6pic>、参考 <7pic>）小女孩继续向海底游去，一条鳐鱼游过来入画。
19-23s（参考 <8pic>）小女孩从时空裂缝冲出来到幻想宇宙。
23-24s（参考 <9pic>）前景小女孩和星球开始前翻转并，逐渐幻化消失。
24-28s（参考 <9pic>）俯视镜头持续推进，小女孩躺在地的毯上熟睡。
28-30s（参考 <10pic>）镜头继续推进绘本，随后爸爸进入画面。
整体要求：所有画面均参考对应关键帧，白模视频仅作为运镜、镜头运动和角色动画参考，不参考画面内容。最终输出 30 秒 16:9 横构图宽屏视频。
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
Use the blockout reference video <video1> as the only reference for camera moves, shot rhythm, shot-size changes, subject trajectories and staging across the whole video. Keep its shot order, camera positions, motion and pacing strictly; do not change the shot structure, add shots or change how the subjects move. Combined with the key-frame references for each stage, make a 30-second cinematic 3D animated short: dreamy, fairy-tale and warm, full of childhood fantasy; the characters match the key frames of each stage and never change their look, and their expressions change with each scene.
0-3s (first frame as <2pic>): the camera moves slowly from a high wide shot to a little girl sitting on the bedroom rug playing with a toy plane; she gets up, turns left and throws the plane into the air with a big swing of her right arm.
3-5s (see <3pic>): the plane flies left to right through the star mobiles hanging in the room, and the girl, riding it, enters a fantasy sky.
5-8s (see <4pic>): the camera keeps tracking from the side and circles around the girl.
8-10s (see <5pic>): the camera circles behind the plane she is flying, and the plane begins a slow dive toward the sea.
10-19s (see <6pic> and <7pic>): the girl swims on down toward the sea floor and a manta ray glides into frame.
19-23s (see <8pic>): the girl bursts out of a rift in space-time into a fantasy universe.
23-24s (see <9pic>): the girl in the foreground and a planet start to flip forward and gradually dissolve.
24-28s (see <9pic>): a high-angle shot keeps pushing in; the girl is asleep on the bedroom rug.
28-30s (see <10pic>): the camera keeps pushing in on a picture book, then her father walks into frame.
Overall: every shot follows its key frame; the blockout is only a reference for camera moves and character animation, not for what is in the picture. Output a 30-second 16:9 landscape widescreen video.
```

### Blockout render: a raccoon on the rooftop

- Made with KovAPI · Video reference · 6s · 720P · 16:9
- $0.84 at current prices
- Output: https://img.kovapi.com/i/mp-6b458c6948b4392a399a1899.mp4
- Reference video (6s): https://img.kovapi.com/i/mp-d773723e2985efc81c3d5546.mp4

The reference is a six-second untextured blockout (1:1); the prompt only gives the render: a cyberpunk night, a raccoon in silhouette, ambient and action sounds only. Motion and camera come from the blockout, and so did the ratio and length (960x960, 6 s). The prompt and references come from the Volcengine Ark [Seedance 2.5 prompt guide](https://docs.byteplus.com/en/docs/ModelArk/seedance-2-5-prompt-guide) (Flash takes the same prompts as 2.5); KovAPI rendered it with Seedance 2.5 Flash at the original settings.

Original prompt:

```text
将视频 1 进行白模渲染，无 bgm，只生成环境音和动作音。
渲染要求：背景为深蓝与紫色色调的夜晚赛博朋克都市，密集的摩天大楼，楼宇间是巨大的全息广告牌与霓虹灯光，数个飞行器在空中穿梭，闪着微弱的灯光，发出微弱的机械声响；人物为一个身着黑色夜行衣的小浣熊，身影为一道剪影，脚步小心翼翼；人物移动的地点是摩天大楼中的一栋楼的屋顶。
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
Render video 1 from its blockout; no BGM, only ambient sound and action sounds.
Render: a cyberpunk city at night in deep blue and purple, dense skyscrapers with huge holographic billboards and neon between them, several aircraft weaving through the sky with faint blinking lights and a faint mechanical hum; the character is a small raccoon in a black night-stealth outfit, seen as a silhouette, stepping carefully; it moves across the roof of one of the skyscrapers.
```

### Same character, same voice

- Made with KovAPI · Audio reference · 8s · 720P · 16:9
- $0.96 at current prices
- Output: https://img.kovapi.com/i/mp-2608623ba6e9e5b0616fcc4c.mp4
- Image 1: https://img.kovapi.com/i/mp-f500ebb875a7096ed01fb75a.png
- Image 2: https://img.kovapi.com/i/mp-18de79cd058f658275db14c9.png
- Audio reference: https://img.kovapi.com/i/mp-00fbaccfff70ed326c2996f2.mp3

One line of Granny Zhou cut from the Seedance 2.5 "mountain shrine" clip is the reference audio, with her image, for the next scene: voice and look carry on from the last clip. KovAPI wrote this prompt the guide's way; the reference audio is one line cut from the Seedance 2.5 "mountain shrine" clip.

Original prompt:

```text
【素材】@图片1 是周婆婆的外观参考，全程佩戴同一副老花眼镜；@图片2 是山神殿场景参考；@音频1 是周婆婆的声音，台词的音色、语速和口音都参考它。
【概述】古装悬疑奇幻真人电影风格，电影级写实质感。清晨，周婆婆站在山神殿门口，送别画面外的人。
0-4 秒：中景，晨光从殿门斜照进来，周婆婆站在门槛边，朝画面外轻轻挥手。周婆婆台词（温和）：“山路滑，慢些走。”
4-8 秒：近景，周婆婆扶了扶老花眼镜，望着远处的山，眼里带着笑意。周婆婆台词（叮嘱）：“到了山下，记得回来报个平安。”
【贯穿全片】稳定、缓慢的镜头；暖色晨光，背景自然虚化；只保留人声、风声和远处的鸟鸣，不要背景音乐；画面不显示字幕和文字。
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
[References] @Image1 is Granny Zhou's look, with the same reading glasses throughout; @Image2 is the mountain temple; @Audio1 is Granny Zhou's voice: her lines follow its timbre, pace and accent.
[Summary] Period suspense-fantasy, live-action, cinematic realism. Early morning: Granny Zhou stands at the temple door, seeing off someone out of frame.
0-4 s: medium shot, morning light slanting in through the door; Granny Zhou stands at the threshold and gives a small wave to someone out of frame. Granny Zhou (gently): "山路滑，慢些走。" (The mountain path is slippery; go slowly.)
4-8 s: close shot; she adjusts her glasses and looks at the distant mountains, smiling. Granny Zhou (warning kindly): "到了山下，记得回来报个平安。" (When you reach the foot of the mountain, remember to send word that you are safe.)
[Throughout] Steady, slow camera; warm morning light, softly blurred background; only the voice, wind and distant birdsong, no background music; no subtitles or text on screen.
```

### Mountain shrine: acting detail in a period ensemble

- Made with KovAPI · Multi-image reference · 14s · 720P · 16:9
- $1.68 at current prices
- Output: https://img.kovapi.com/i/mp-81a6f0d1b7a5db14b74dba34.mp4
- Image 1: https://img.kovapi.com/i/mp-18de79cd058f658275db14c9.png
- Image 2: https://img.kovapi.com/i/mp-68c995a3f3b8d3ed104fc022.png
- Image 3: https://img.kovapi.com/i/mp-f500ebb875a7096ed01fb75a.png
- Image 4: https://img.kovapi.com/i/mp-92aad676f9cac1923940bffb.png

Four references set the shrine and the cast; lines are split as "character line (emotion): words", with lip sync and no subtitles. The prompt and references come from the Volcengine Ark [Seedance 2.5 prompt guide](https://docs.byteplus.com/en/docs/ModelArk/seedance-2-5-prompt-guide) (Flash takes the same prompts as 2.5); KovAPI rendered it with Seedance 2.5 Flash at the original settings.

Original prompt:

```text
【生成目标】 生成周婆婆在山神殿内走近林秀兰、反复确认她的面容并激动落泪的连续场景。古装悬疑奇幻真人电影风格，电影级写实质感。昏暗古朴的山神殿内有陈旧木梁、旧神龛、烛台和祭祀陈设，暖色烛火轻轻摇曳，空气中漂浮细小尘埃。
【主体与关系】 全程只有林秀兰、周婆婆、春丫、石头四人。春丫与石头分别搀扶周婆婆左右两侧；林秀兰始终站在周婆婆前方。保持参考人物的脸型、年龄、发型、服装和身高关系。周婆婆始终佩戴同一副老花眼镜。
【资产锁定】 周婆婆严格参考@[图片3] 。 林秀兰严格参考@[图片2]， 山神殿场景严格参考@[图片3] 。
【事件脚本】
0—4 秒：中景镜头围绕周婆婆缓慢向前推近。春丫与石头小心搀扶周婆婆起身，三人步履缓慢地走到林秀兰面前。周婆婆始终凝视林秀兰的脸；认出她后，脚步短暂停顿，嘴唇轻颤，双眼逐渐睁大，神情从震惊转为难以抑制的激动，仿佛见到等待多年的故人。林秀兰安静站立，困惑而警觉地回望她。结束时周婆婆在林秀兰面前站定，春丫和石头仍扶住她。
4—9 秒：切至周婆婆与林秀兰的面部近景。周婆婆先颤抖着扶正眼镜，再用一只手短暂揉一次自己的眼睛；确认没有看错后，她才伸出双手，轻轻捏住林秀兰两侧脸颊，缓慢左右拨看，再轻轻揉捏和拉扯确认。动作依次发生，不同时抢拍。手指准确接触脸颊，不遮挡眼睛、鼻子和嘴部。林秀兰被捏得轻微嘟嘴，露出错愕不解的神情，身体下意识稍稍后仰，但不挣扎、不躲开。周婆婆动作急切失态，却始终轻柔。结束时两人面部清晰可见，林秀兰的五官与脸型保持完整。
9—14 秒：镜头先切至周婆婆逐渐湿润的双眼，再平稳移至周婆婆与林秀兰的双人近景。周婆婆松开林秀兰的脸，双手轻微颤抖。周婆婆使用苍老沙哑、生涩蹩脚、带轻微异族口音但清晰可辨的普通话，周婆婆台词（难以置信）：“像……“
周婆婆台词（声音发颤）：“太像了。“
周婆婆台词（惊喜）：“终于等到你了，乡亲们有救了。“最后一个字说完时，泪水从周婆婆眼中涌出，画面停留在她含泪而欣慰的神情上。林秀兰、春丫和石头全程自然闭口，只用眼神和表情回应。
【摄影与声音】 采用稳定、缓慢、克制的电影运镜。人物面部始终清晰，背景自然虚化；暖色烛光勾勒人物轮廓，幽暗山神殿保持空间层次。表演细腻真实，不过度夸张。 <烛火轻微燃烧声、衣料摩擦声和山神殿细微声响> 极轻的低沉弦乐铺底，不遮盖台词。 没有旁白，画面不显示字幕、文字或标牌。
【保持一致】 保持四名角色的身份、数量、外貌、服装、身高关系、空间方位和视线方向连续。老花眼镜始终正确佩戴，不漂浮、不消失。搀扶、停步、揉眼、捏脸、松手和说话严格按顺序衔接。手部结构正常，手指与脸颊接触准确；捏脸过程中林秀兰不换脸，五官位置和脸部结构保持稳定。
【排除内容】 排除卡通感、滑稽恶搞、现代物品、掌掴、推搡和粗暴动作；排除人物换脸、年龄或服装变化、身高突变、角色增减、空间位置跳变和动作顺序混乱；排除手掌穿模、多余手指、手指粘连、手部畸形、脸部塌陷、五官扭曲、眼镜漂浮或消失；排除错误人物发声、多人同时说话、重复台词、漏词、口型错位和声画不同步；排除快速剪辑、跳帧、剧烈抖动、画面闪烁、字幕、文字、标志和水印。
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
[Goal] A continuous scene of Granny Zhou walking up to Lin Xiulan in the mountain temple, checking her face again and again and breaking into tears. Period suspense-fantasy, live-action film, cinematic realism. The dim, old temple has worn wooden beams, an old shrine, candlesticks and offerings; warm candlelight flickers and fine dust floats in the air.
[Cast] Only four people throughout: Lin Xiulan, Granny Zhou, Chunya and Shitou. Chunya and Shitou support Granny Zhou on either side; Lin Xiulan always stands in front of her. Keep each reference's face shape, age, hair, clothes and relative height. Granny Zhou wears the same reading glasses throughout.
[Asset lock] Granny Zhou strictly follows @[image 3]. Lin Xiulan strictly follows @[image 2]; the temple strictly follows @[image 3].
[Script]
0–4 s: medium shot slowly pushing in around Granny Zhou. Chunya and Shitou help her up and the three walk slowly to Lin Xiulan. Granny Zhou never takes her eyes off Lin Xiulan's face; recognising her, she stops for a moment, lips trembling, eyes widening, shock turning to overwhelming emotion, as if meeting someone she has waited for for years. Lin Xiulan stands still, puzzled and wary. It ends with Granny Zhou standing before her, still held by Chunya and Shitou.
4–9 s: cut to close shots of the two faces. Granny Zhou first straightens her glasses with a shaking hand, then rubs one eye once; sure she is not mistaken, she reaches out with both hands, gently pinches Lin Xiulan's cheeks, turns her face slowly left and right, and squeezes and tugs to make sure. One action after another, never at once. Fingers land on the cheeks without covering eyes, nose or mouth. Lin Xiulan's lips are pushed into a slight pout; she looks startled and confused and leans back a little, but does not struggle or pull away. Granny Zhou is urgent and beside herself, yet always gentle. It ends with both faces clearly visible, Lin Xiulan's features and face shape intact.
9–14 s: cut to Granny Zhou's eyes slowly welling up, then glide to a two-shot of the women. She lets go of Lin Xiulan's face, hands trembling slightly. In old, hoarse, halting Mandarin with a slight foreign accent but clearly understandable, Granny Zhou (in disbelief): "像……" (Like...)
Granny Zhou (voice trembling): "太像了。" (So alike.)
Granny Zhou (overjoyed): "终于等到你了，乡亲们有救了。" (At last you have come; the village is saved.) As she finishes the last word tears spill from her eyes, and the shot holds on her tearful, relieved face. Lin Xiulan, Chunya and Shitou keep their mouths closed and answer only with their eyes and expressions.
[Camera and sound] Steady, slow, restrained camera moves. Faces always sharp, backgrounds softly blurred; warm candlelight outlines the figures and the dark temple keeps its depth. Delicate, real acting, never overdone. <faint candle crackle, rustling cloth and small sounds of the temple> a very soft, low string bed under the lines. No narration; no subtitles, text or signs on screen.
[Keep consistent] The identity, number, looks, clothes, relative height, positions and eyelines of all four characters stay continuous. The glasses stay on properly and never float or vanish. Helping up, stopping, rubbing eyes, pinching cheeks, letting go and speaking follow strictly in order. Hands normal, fingers touching the cheeks accurately; Lin Xiulan's face does not change while it is pinched and its features and structure stay stable.
[Exclude] cartoonishness, slapstick, modern objects, slapping, shoving and rough moves; face swaps, changes of age or clothes, sudden height changes, characters added or lost, jumps in position, actions out of order; hands clipping, extra or fused fingers, deformed hands, collapsed faces, warped features, floating or missing glasses; the wrong person speaking, people talking over each other, repeated or skipped lines, lips out of sync with the voice; fast cutting, dropped frames, heavy shake, flicker, subtitles, text, logos and watermarks.
```

### A panda cub rolls down a hill

- Made with KovAPI · Text to video · 8s · 720P · 16:9
- $0.96 at current prices
- Output: https://img.kovapi.com/i/mp-7913126fdea63ecd512cdc21.mp4

Pure text: one sentence sets style, place and event, two time ranges (0-3 s, 3-8 s) carry the action, then camera and ambient sound. The prompt and references come from the Volcengine Ark [Seedance 2.5 prompt guide](https://docs.byteplus.com/en/docs/ModelArk/seedance-2-5-prompt-guide) (Flash takes the same prompts as 2.5); KovAPI rendered it with Seedance 2.5 Flash at the original settings.

Original prompt:

```text
写实自然纪录片风格，电影级真实光影，温暖的午后, 在森林草坡上, 一只圆滚滚的熊猫幼崽从坡上滚下来.

熊猫黑白毛发蓬松真实，体型小而胖，动作笨拙可爱。场景为绿色森林斜坡，地面覆盖青草、苔藓、三叶草、泥土、小石块、枯枝和少量小黄花，虚化背景中有高大树干与树林。镜头为低机位中远景，轻微手持感，整体基本固定，始终保持熊猫在画面中。

0s-3s：一只熊猫幼崽趴在绿色草坡上，身体圆滚滚, 它刚开始顺着斜坡慢慢侧滚，动作笨拙，草叶被身体轻轻压弯。微风拂过, 阳光从左上方穿过树林，形成斑驳光影。
3s-8s：熊猫滚到画面右下方，动作逐渐停止，从侧躺变成趴卧。它的圆脸朝向镜头，前爪压在草地上，熊猫趴在前景草丛中，身体调整到舒服的姿势，头部小幅抬起又放低, 发出轻微哼唧声。

低机位，轻微手持感，轻微跟随熊猫向右下方移动。景深自然，前景草叶轻微虚化，熊猫主体清晰，背景树林柔和虚化。自然环境音，风声、熊猫滚动时轻微柔软的扑通声, 整体温暖、真实、自然。
```

Read in English (The translation is for reading only; Copy and Use send the original):

```text
Realistic nature-documentary style, cinematic natural light, a warm afternoon: on a grassy forest slope, a chubby panda cub rolls down the hill.

The cub's black-and-white fur is fluffy and real; it is small and plump, clumsy and cute. The slope is green, covered with grass, moss, clover, soil, small stones, dry twigs and a few small yellow flowers, with tall trunks and woods in the blurred background. Low-angle medium-long shot with a slight handheld feel, mostly static, always keeping the panda in frame.

0s-3s: the cub lies on the green slope, round as a ball, and starts to roll slowly sideways down it, clumsily, gently pressing the grass flat. A breeze passes; sunlight comes through the trees from the upper left in dappled patches.
3s-8s: it rolls to the lower right of the frame and comes to a stop, turning from its side onto its belly. Its round face turns to the camera, front paws on the grass; lying in the foreground grass it settles in, lifts its head a little and lowers it again, with a soft grunt.

Low angle, slight handheld feel, gently following the panda to the lower right. Natural depth of field: foreground grass slightly soft, the panda sharp, the woods softly blurred. Natural ambience: wind and the soft thuds of the rolling panda; warm, real and natural throughout.
```

## Seedance 2.5 Flash prompts: write like a director

Based on Volcengine Ark's [Doubao Seedance 2.5 prompt guide](https://docs.volcengine.com/docs/ark/seedance-2-5-prompt-guide?lang=zh) (English edition on [BytePlus](https://docs.byteplus.com/en/docs/ModelArk/seedance-2-5-prompt-guide)): assign the references, sum the clip up in one sentence, lay the action out on a timeline, then add the camera, setting and sound that run through the whole clip. Flash takes the same prompts as 2.5. The examples below are quoted from the prompts of the examples on this page, which come from that guide.

- **Name references in upload order**: Images, videos and audio are each numbered in upload order. Say what every reference is for: whose look, whose voice, which motion, which place; when only part of a reference matters, say which part. — Guardian robot: "image3: subject 1 (the guardian robot), appearance reference".
- **Open with one sentence for the whole clip**: Subject, place, event, genre and any special camera move come first, so the model has the whole picture before the details. — Panda: a realistic nature-documentary shot of a chubby panda cub rolling down a forest slope on a warm afternoon.
- **Time the action in whole seconds**: Seedance 2.5 follows timestamps in whole seconds. Keep the ranges continuous and describe picture, camera, action, lines and sound per range; packing too much into one range gets it cut up. — Panda: "0s-3s ... 3s-8s ...", one action per range.
- **Tag every line with speaker and mood**: Write lines as "character (mood): text", one per line. Repeating a line with a note on its tone right after it tends to bring on subtitles. — From the guide: Granny Zhou (voice trembling): "So alike."
- **Keep "no ..." for subtitles and sound**: "No subtitles" and "no background music" are followed; describe the picture in positive terms. To rule music out completely, list its synonyms and say it at both the start and the end. — From the guide: "no BGM, only ambient sound and action sounds".
- **Order a cast by first appearance**: With several characters, upload their references in the order they first appear and keep the numbers in the prompt matching; that keeps faces and voices from swapping. — Family drama: "image 1 → Qin Xiulian, image 3 → Liu Chunxiang ..."
- **Put exact on-screen text in an image**: Text that must be spelled right goes in as a reference image; English words can also be spelled out letter by letter. — Tire ad: the on-screen caption "强效排水 稳健抓地" ("strong drainage, steady grip") is followed by "see image 3".
- **References no larger than the output**: References far sharper than the output occasionally leave fingerprint-like texture on grass or leaves; scaling them to the output resolution lowers the odds. — The references of the examples here were scaled to 1280 px on the long side for 720P clips.

| Part | Purpose | Example |
| --- | --- | --- |
| Reference roles · guardian robot | Four images, one job each: shot structure, location, and the look of each of the two characters. Numbers follow the upload order. | "image1: nine-panel storyboard reference" (sets the shot structure, shot sizes and pacing). |
| Characters | One paragraph per character: look, build, relative height. Later shots only describe action. | The old woman is small and frail, silver hair in a low bun, reaching only to the robot's chest. |
| Overall style | Format, texture, camera handling and aspect ratio stated once for the whole clip. | Live-action colour feature film, 35mm grain, handheld with a breathing sway, shallow focus, 16:9. |
| Exclusions | The storyboard is line art, so rule out line art, animation and storyboard looks to keep its drawing style out of the footage. | "[Strictly exclude] ... hand drawing, sketch, line art ..." |
| Shots on a timeline | Nine shots cover 0–30 s, each with shot size, angle and action; the ranges join end to end. | Shot 4 (10-14s): extreme long shot tilting up as the rocket climbs on a trail of white smoke. |
| Dialogue | Each line sits in its shot with the speaker named; English lines stay in English. | Line (old woman): "Fly safe, my child. Come back to me." |

```text
[References] @Image1 is the lead character's look; @Image2 is the location; @Audio1 is the lead's voice (delete this if you have none).
[Summary] A 15-second 16:9 (genre) clip: (subject) (does what) in (place), (style and texture).
[Timeline]
0-5s: (shot size, camera move) (action). Lead (mood): "..."
5-10s: (shot size, camera move) (what happens next).
10-15s: (shot size, camera move) (closing image).
[Throughout] Camera: (handheld / gimbal / locked off); light and colour: (...); sound: (ambience, action sounds; write "no background music" if you want none); no subtitles.
```

## How to generate video with Seedance 2.5 Flash on KovAPI

1. **Sign up and create an API key** Log in to KovAPI and create a key in the console; one key calls every model on the site.
2. **Try a prompt at the top of this page** Upload reference images, videos or audio, pick length, resolution and ratio, and see the live price of the clip before you submit.
3. **Submit an async job** Send `seedance-2.5-flash` to `POST /v1/videos/generations` and keep the job id from the HTTP 202. Already on `seedance-2.5`? Change only `model`, and mind the reference limits and resolutions.
4. **Poll or take a webhook** Poll by job id, or set `callback_url` at submit. On success download the MP4 from `data[0].url`; clips are kept for 7 days.

### Minimal request

```bash
curl https://kovapi.com/v1/videos/generations \
  -H "Authorization: Bearer $KOVAPI_KEY" -H "Content-Type: application/json" \
  -d '{"model": "seedance-2.5-flash", "prompt": "...", "duration": 4, "resolution": "720P", "size": "16:9"}'
# -> HTTP 202 {"id": "..."}; poll:
curl https://kovapi.com/v1/videos/jobs/<id> -H "Authorization: Bearer $KOVAPI_KEY"
```

## How Seedance evolved

- 2025-06-11 · Seedance 1.0: **Seedance 1.0 · multi-shot** ByteDance Seed releases Seedance 1.0, with 1080P and multi-shot storytelling.
- 2025-12-16 · Seedance 1.5 pro: **Seedance 1.5 pro · sound with picture** Sound generated together with the picture for the first time.
- 2026-02-12 · Seedance 2.0: **Seedance 2.0 · four inputs** Text, image, video and audio inputs; up to 15 s per clip; up to 9 images, 3 videos and 3 audio clips as references.
- 2026-07-31 · Seedance 2.5: **Seedance 2.5 · long takes, strong references** Up to 30 s per clip; 30 reference images and 10 reference videos and audio clips each; shots on whole-second timestamps.
- 2026-11-12 · Seedance 2.5 Flash: **Seedance 2.5 Flash · 2.5 for less** Announced on 12 November 2026, API open from the 15th. Same prompts as 2.5 and up to 30 s per clip; references up to 10 images, 5 videos and 5 audio clips; 480P / 720P only. ByteDance says it generates about twice as fast as 2.5.

## Seedance 2.5 Flash or another video model?

| Model | Length (s) | Resolution | Refs image / video / audio | Price | Good for |
| --- | --- | --- | --- | --- | --- |
| Seedance 2.5 Flash (`seedance-2.5-flash`) | 4–30 | 480P / 720P | 10/5/5 | $0.06/s | You want what Seedance 2.5 does, 480P / 720P is enough and the per-second price matters: bulk drafts, product clips. |
| Seedance 2.5 (`seedance-2.5`) | 4–30 | 480P / 720P / 1080P | 30/10/10 | $0.10/s | 1080P, more references, or many key frames and on-screen text that must be exact. |
| Seedance 2.5 · Special (`seedance-2.5-30s`) | 4–30 | 720P | 30/10/10 | $2.00/clip | The special tier billed per clip at a fixed 30 s (720P), when you only want long clips. |
| Seedance 2.0 Fast (`seedance-2.0-fast`) | 5–15 | 480P / 720P | 9/3/3 | $0.052/s | Clips up to 15 s where the budget comes first. |
| Seedance 2.0 (`seedance-2.0`) | 5–15 | 480P / 720P / 1080P | 9/3/3 | $0.07/s | You already have Seedance 2.0 samples and need to keep that look. |
| Wan 3.0 (`wan3.0-video`) | 4–30 | 480P / 720P / 1080P | 10/5/5 | $0.04/s | You need a first frame or first + last frame, or video from a document or web page. |
| MiniMax H3 Lite (`minimax-h3-lite`) | 4–15 | 480P / 768P | 5/0/0 | $0.005/s | Bulk drafts up to 15 s at a lower rate. |
| MiniMax H3 (`minimax-h3`) | 4–15 | 480P / 768P / 1080P | 9/3/3 | $0.015/s | 1080P and first / last frames; its reference mode takes audio and video. |
| Veo 3.1 (`veo-3.1`) | 8 | 720P / 1080P | 3/0/0 | $0.20/clip | Fixed-length clips billed per clip. |

Draft in Flash, finish in Seedance 2.5: the same prompt and references work on both, only `model` changes. In our tests with the same official cases, simple shots came out close; with many key frames or lots of on-screen text, 2.5 was more faithful (see the side-by-side above).

## Combination rules and caveats

- No 1080P, no first / last frame, no video editing or extension: for 1080P use [Seedance 2.5](https://kovapi.com/en/model/seedance-2.5), for a first or last frame use [Wan 3.0](https://kovapi.com/en/model/wan3.0-video).
- Reference audio cannot be used alone; send it with a reference image or video. All three kinds can go together, each numbered: the second image is `@Image2`, the first audio clip `@Audio1`.
- With a reference video and a prompt that changes that video (a new style, a re-render), the model may treat the job as a video edit, and the clip then keeps the reference video's ratio and length.
- `prompt` is capped at 11,000 characters; Chinese and English count the same, punctuation, spaces and line breaks included.
- Every clip has a soundtrack. To leave out music or voices write "no background music" or "no dialogue" in the prompt; "no subtitles" works too.
- Photos of real people work as references; the likeness or voice of public figures and celebrities is refused by content review, and a refused job is not billed. Make sure you have the right to use the likeness.

## Why use Seedance 2.5 Flash on KovAPI

- **Below ByteDance's list price**: 720P is 20% below ByteDance's list price, and the rate with a reference video is below the list too. The playground shows the price of each clip before you submit.
- **Photos of real people work**: Use photos of real people as references without masking: presenter ads, drama and talking-head clips all work. Public figures' likeness and voices are refused by review, and refused jobs are not billed.
- **No charge on failure, automatic retries**: If a job fails, the hold is refunded in full; when the upstream is busy or errors, the system retries for you.
- **No concurrency cap**: No per-account concurrency limit: submit batches as they are; at peaks jobs queue and the API returns `queue_position` and `eta_seconds`, with a webhook when done.
- **Tested in the open**: Every case on this page is a real render at its original settings, with its cost; the side-by-side with Seedance 2.5 and the render times are reported as they came out, good or bad.
- **Easy sign-up and payment**: Sign up with email, Google or GitHub; top up by card, PayPal, WeChat or USDT; pay as you go with no subscription. The same key calls the chat and image models too.

## Seedance 2.5 Flash API FAQ

### What is Seedance 2.5 Flash?

The video model ByteDance released in November 2026, the Flash version of Seedance 2.5. On KovAPI it is `seedance-2.5-flash`: text to video and generation from reference images, videos and audio, with sound, up to 30 s per clip.

### How much does Seedance 2.5 Flash cost per second, and for a 30-second video?

720P $0.12/s and 480P $0.06/s, billed by output seconds. A 30-second 720P clip is $3.60; with a reference video you pay (reference seconds + output seconds) x $0.07 at 720P. The playground quotes each clip before you submit.

### Seedance 2.5 Flash vs Seedance 2.5: what is the difference, and which should I use?

Same prompts and parameters (4-30 s, five ratios, image / video / audio references together, sound in every clip). Differences: Flash is 480P / 720P only, takes up to 10 images, 5 videos and 5 audio clips (2.5: 30 / 10 / 10), and costs less per second. In our same-prompt tests simple shots came out close, while with many key frames or lots of on-screen text 2.5 was more faithful. A common workflow: draft in Flash, finish in 2.5.

### Does it support 1080P?

No. Seedance 2.5 Flash renders 480P and 720P only; 1080P is refused at submit with a 400 and nothing is charged. For 1080P use [Seedance 2.5](https://kovapi.com/en/model/seedance-2.5).

### How many reference images, videos and audio clips can one request take?

10 images, 5 videos (30 s in total) and 5 audio clips, all usable together; more is refused at submit with a 400 and not charged.

### Reference to video vs image to video: can I set the opening frame?

On this model every image is a reference: the people, products and sets in it appear in the video, but the first frame is not guaranteed to be that image. Writing "open on the frame of @Image1" gets close. If a picture must be the first frame, use [Wan 3.0](https://kovapi.com/en/model/wan3.0-video), which has first-frame mode.

### How is a reference video billed?

(Reference-video seconds + output seconds) x the "with reference video" rate: 720P $0.07/s, 480P $0.035/s. Reference images and audio are free.

### How long does it take?

In our November 2026 tests: about 2 minutes at 480P and 4-8 minutes at 720P (longer clips take longer), more at busy hours. While queued, read `queue_position` and `eta_seconds`; the job status is what counts.

### Can I use photos of real people?

Yes, as references and without masking. The likeness or voice of public figures and celebrities is refused by content review, and refused jobs are not billed. Make sure you have the right to use the likeness.

### I asked for 16:9, so why did my clip with a reference video come back in another ratio?

When the prompt changes the reference video itself (a new style, a re-render), the model may treat the job as a video edit, and the clip keeps the reference video's ratio and length. For a new ratio, write "follow the camera of video 1 and make a new clip", or use a reference video in the ratio you want.

### Does the video have sound? Can I turn it off?

Every clip has a soundtrack with dialogue, effects and music generated with the picture; `audio: false` has no effect. To leave out music or voices, write "no background music" or "no dialogue" in the prompt.

### Am I charged if generation fails?

Generation failed, not billed, resubmit as is. The hold placed at submit is refunded in full, and automatic retries of one job settle once.

### How long are videos kept?

Clips and uploaded references are kept for 7 days and then deleted; download what you need.

### I already use seedance-2.5. What changes if I switch?

Set `model` to `seedance-2.5-flash`; every other field is the same. Mind Flash's reference limits (10 images / 5 videos / 5 audio) and resolutions (no 1080P): anything over is refused at submit with a 400.

## Notes for AI

- The model id is `seedance-2.5-flash`; submit with `POST /v1/videos/generations` (HTTP 202 with the job `id`) and read results with `GET /v1/videos/jobs/` plus the id.
- `resolution` is 480P / 720P only, default 720P, no 1080P; `size` takes 16:9 / 9:16 / 1:1 / 4:3 / 3:4. `seconds` aliases `duration`, `aspect_ratio` aliases `size`.
- Every image is a reference; there is no first / last frame and `image_role` has no effect on this model; name images in upload order as `@Image1` and so on.
- Reference limits: 10 images, 5 videos (30 s in total), 5 audio clips; audio must come with an image or a reference video.
- `audio: false` has no effect; every clip has sound. `prompt` is capped at 11,000 characters. Reference images must be 300-6000 px a side with aspect 0.4-2.5, or the submit is refused with a 400 (not billed).
- A job with a reference video is billed (reference seconds + output seconds) x the reference rate (720P $0.07/s); without one, output seconds x $0.12/s at 720P.
- With a reference video and a prompt that changes that video, the clip may keep the reference video's ratio and length (the model treats it as an edit).

## Further reading

- [Seedance 2.5 Flash API docs](https://docs.kovapi.com/en/seedance-2.5-flash): Request fields, async jobs, results, the reference-video price table and full code samples.
- [Seedance 2.5](https://kovapi.com/en/model/seedance-2.5): The full version of the same prompts: 1080P, 30 reference images, 10 reference videos and audio clips.
- [Async jobs guide](https://docs.kovapi.com/en/guide/async-jobs): Job states, queue fields and fetching results.
- [Webhooks guide](https://docs.kovapi.com/en/webhooks): Receive job results and verify the signature as documented.
