superouter Video Gen
Overview
This skill is for the superouter platform API, not for talking to Jimeng upstream directly.
Default model: seedance-2.0-v1,you can get model list from the server.
Use this skill when the user wants to:
- - upload ordered reference assets
- submit an async video generation task
- query status with platform INLINECODE1
- download the final video from the real URL returned by the platform
The platform currently exposes:
- - INLINECODE2
- INLINECODE3
- INLINECODE4
- INLINECODE5
- INLINECODE6
- INLINECODE7
Preconditions
- -
SUPER_KEY must be a superouter client API key - If the platform returns
401 or 403, stop and tell the user the client key is invalid or disabled - Default ratio for short drama generation is
9:16 only when the user did not give a ratio - Default model is
seedance-2.0-v1; do not invent other fields that affect generation
Current Platform Rules
Prompt
- -
prompt max length: 2000 characters
File types
- - Images:
jpg jpeg png webp gif INLINECODE20 - Video:
mp4 mov INLINECODE23 - Audio:
mp3 INLINECODE25
File limits
- - Single file max: INLINECODE26
- Total referenced asset size per task max: INLINECODE27
- Image shortest side min: INLINECODE28
- Image long/short side ratio max: INLINECODE29
- Video duration max: INLINECODE30
- Audio duration max: INLINECODE31
Seedance submit limits
- -
aspectRatio: 1:1 4:3 3:4 16:9 9:16 INLINECODE38 - INLINECODE39 : INLINECODE40
- INLINECODE41 :
4 to INLINECODE43 - INLINECODE44 : currently only INLINECODE45
Pricing
For seedance-2.0-v1:
- - no video reference: INLINECODE47
- any video reference present: INLINECODE48
- if
duration is omitted, use 4 seconds for estimation
INLINECODE51 and query currently do not deduct credits.
Asset selection principles
- - Only upload the people or scene references that actually matter for the current shot
- Keep the asset list as small and ordered as possible
- If clothing continuity is critical, add clothing references explicitly instead of assuming them
- Video generation is sensitive to reference order; keep
references[] in the exact business order
Workflow
1. Check balance before submit
Use platform balance, not upstream quota:
CODEBLOCK0
Estimate credits before submit:
- - use
duration if provided, otherwise INLINECODE55 - if any reference asset is video, multiply by INLINECODE56
- otherwise multiply by INLINECODE57
If balance is insufficient, stop before upload/submit and explain the estimated cost.
2. Validate local inputs
Before any upload:
- - confirm every referenced local file exists
- confirm prompt length is INLINECODE58
- do not invent file order
- keep placeholder names aligned with reference order
If the prompt uses named placeholders such as <<<Image1>>>, make sure the corresponding references[].name values match exactly.
3. Upload assets first
Upload each local file to the platform and collect fileId.
CODEBLOCK1
Response example:
CODEBLOCK2
Rules:
- - always save the returned INLINECODE62
- same-client duplicate upload may return an existing INLINECODE63
- do not assume one upload call accepts multiple files unless the user specifically wants to script multiple sequential uploads
- you can save
fileId for next submit if file is same.
4. Submit the task
Submit to the platform, not to Jimeng directly.
CODEBLOCK3
Rules:
- - keep
references[] in the exact intended order - first reference is the first logical asset, second reference is the second logical asset, and so on
- do not invent
aspectRatio, resolution, duration, or INLINECODE69 - default to async submit behavior; the platform returns INLINECODE70
- after success, report
taskId and the estimated credit cost
Successful response shape:
CODEBLOCK4
5. Query task status
Use platform taskId.
CODEBLOCK5
Possible status values you should expect:
- - INLINECODE73
- INLINECODE74
- INLINECODE75
- INLINECODE76
Completed response example:
CODEBLOCK6
Rules:
- - the returned
videos[].url is the real download URL; give it directly to the user - do not proxy-download unless the user explicitly asks you to save the file locally
- if status is
processing, report that clearly and stop unless the user asked you to keep polling - if status is
failed, return errorMessage; the platform may refund automatically
6. Optional task list
If the user wants recent tasks instead of querying a single task:
CODEBLOCK7
Use this to inspect recent platform tasks. The platform may refresh incomplete tasks when listing them.
Prompt / ordering rules
- - Keep the reference list in the same order the user intends the model to read it
- Use
@1 placeholders only when the request actually names references - If the user gives unnamed files only, keep order and do not invent semantic names
- If the user asks for a shot prompt rewrite, keep the asset references consistent with the final INLINECODE82
Fixed rules
- - Use the superouter platform API only
- Default only the model; ask for missing generation inputs that materially affect output
- Do not claim success before you have a platform INLINECODE83
- Do not wait for completion by default after submit
- If the service is unreachable, tell the user to check whether the superouter API is running
superouter 视频生成
概述
此技能适用于 superouter 平台 API,而非直接对接 Jimeng 上游。
默认模型:seedance-2.0-v1,可通过服务器获取模型列表。
当用户需要以下操作时使用此技能:
- - 上传已排序的参考素材
- 提交异步视频生成任务
- 使用平台 taskId 查询状态
- 从平台返回的真实 URL 下载最终视频
平台当前提供的接口:
- - POST /v1/video/assets/upload
- POST /v1/video/omni-reference/task/submit
- GET /v1/video/omni-reference/task/query
- GET /v1/video/models
- GET /me/balance
- GET /me/tasks
前置条件
- - SUPER_KEY 必须是 superouter 客户端 API 密钥
- 如果平台返回 401 或 403,停止操作并告知用户客户端密钥无效或已被禁用
- 短剧生成的默认宽高比为 9:16,仅当用户未指定宽高比时生效
- 默认模型为 seedance-2.0-v1;请勿自行添加其他影响生成的字段
当前平台规则
提示词
文件类型
- - 图片:jpg jpeg png webp gif bmp
- 视频:mp4 mov m4v
- 音频:mp3 wav
文件限制
- - 单个文件最大:30MB
- 每个任务引用的素材总大小上限:100MB
- 图片最短边最小值:320px
- 图片长边/短边比例最大值:3:1
- 视频时长上限:15s
- 音频时长上限:15s
Seedance 提交限制
- - aspectRatio:1:1 4:3 3:4 16:9 9:16 21:9
- resolution:720p
- duration:4 到 15
- generatingCount:当前仅支持 1
定价
对于 seedance-2.0-v1:
- - 无视频参考:5 积分 / 秒
- 包含任何视频参考:10 积分 / 秒
- 如果未提供 duration,按 4 秒估算
upload 和 query 当前不扣除积分。
素材选择原则
- - 仅上传对当前镜头实际重要的人物或场景参考
- 尽可能保持素材列表精简且有序
- 如果服装连续性至关重要,请明确添加服装参考,而非依赖模型推断
- 视频生成对参考顺序敏感;请确保 references[] 严格按照业务顺序排列
工作流程
1. 提交前检查余额
使用平台余额,而非上游配额:
bash
curl http://superouter.nesports.top/me/balance \
-H Authorization: Bearer ${SUPER_KEY}
提交前估算积分:
- - 使用提供的 duration,否则使用 4
- 如果任何参考素材为视频,乘以 10
- 否则乘以 5
如果余额不足,在上传/提交前停止操作,并说明预估费用。
2. 验证本地输入
在任何上传操作之前:
- - 确认每个引用的本地文件存在
- 确认提示词长度 <= 2000
- 不要自行调整文件顺序
- 保持占位符名称与参考顺序一致
如果提示词使用了命名占位符(如 <<>>),请确保对应的 references[].name 值完全匹配。
3. 先上传素材
将每个本地文件上传至平台,并收集 fileId。
bash
curl --request POST http://superouter.nesports.top/v1/video/assets/upload \
--header Authorization: Bearer ${SUPER_KEY} \
--form file=@/absolute/path/to/reference-1.png
响应示例:
json
{
code: 200,
message: Success,
result: {
fileId: asset_xxx,
fileName: reference-1.png,
contentType: image/png,
mediaType: image,
sizeBytes: 12345,
status: uploaded
}
}
规则:
- - 始终保存返回的 fileId
- 同一客户端重复上传可能返回已有的 fileId
- 除非用户明确要求编写脚本进行多次顺序上传,否则不要假设一次上传调用可接受多个文件
- 如果文件相同,可保存 fileId 供下次提交使用
4. 提交任务
向平台提交,而非直接提交至 Jimeng。
bash
curl --request POST http://superouter.nesports.top/v1/video/omni-reference/task/submit \
--header Authorization: Bearer ${SUPER_KEY} \
--header Content-Type: application/json \
--data {
model: seedance-2.0-v1,
prompt: 让 @1 中的人物根据 @2 的内容说话。,
references: [
{ fileId: assetimage1, name: Image1 },
{ fileId: assetaudio1, name: Audio1 }
],
aspectRatio: 9:16,
resolution: 720,
duration: 4,
generatingCount: 1
}
规则:
- - 保持 references[] 严格按照预期顺序排列
- 第一个引用对应第一个逻辑素材,第二个引用对应第二个逻辑素材,以此类推
- 不要自行添加 aspectRatio、resolution、duration 或 references[].name
- 默认采用异步提交行为;平台返回 taskId
- 成功后,报告 taskId 和预估积分费用
成功响应格式:
json
{
code: 200,
message: Success,
result: {
taskId: task_xxx,
status: submitted
}
}
5. 查询任务状态
使用平台 taskId。
bash
curl --get http://superouter.nesports.top/v1/video/omni-reference/task/query \
--header Authorization: Bearer ${SUPER_KEY} \
--data-urlencode taskId=task_xxx
您应预期的可能状态值:
- - submitted
- processing
- completed
- failed
完成响应示例:
json
{
code: 200,
message: Success,
result: {
taskId: task_xxx,
status: completed,
errorMessage: null,
videos: [
{
url: https://...
}
]
}
}
规则:
- - 返回的 videos[].url 是真实的下载 URL;直接提供给用户
- 除非用户明确要求将文件保存到本地,否则不要代理下载
- 如果状态为 processing,清晰报告并停止,除非用户要求持续轮询
- 如果状态为 failed,返回 errorMessage;平台可能自动退款
6. 可选的任务列表
如果用户想查看最近的任务而非查询单个任务:
bash
curl http://superouter.nesports.top/me/tasks?limit=50 \
-H Authorization: Bearer ${SUPER_KEY}
使用此接口查看最近的平台任务。平台在列出任务时可能会刷新未完成的任务。
提示词/排序规则
- - 保持参考列表的顺序与用户期望模型读取的顺序一致
- 仅当请求中实际命名了参考时才使用 @1 占位符
- 如果用户仅提供未命名的文件,保持顺序,不要自行添加语义名称
- 如果用户要求重写镜头提示词,保持素材引用与最终 references[] 一致
固定规则
- - 仅使用 superouter 平台 API
- 仅默认设置模型;对实质性影响输出的缺失生成输入进行询问
- 在获得平台 taskId 之前,不要声称成功
- 提交后默认不等待完成
- 如果服务无法访问,告知用户检查 superouter API 是否正在运行