跳到主要内容

MiniMaxVideo model检查中

MiniMax H3 API

模型 IDminimax-h3-regeneration-image-to-video

MiniMax H3 video generation with text, image, multimodal reference and regeneration variants.

  • 图生视频
  • 文生视频
  • Video to video
  • Reference to video
  • Up to 9 input images
  • Up to 15 seconds
  • 480p or 768p output

在 Playground 中运行 MiniMax H3

The form is the request body: every field shows its key. Sign in to run it; a run is billed like an API task.

参数

模型
model

MiniMax H3Regeneration image to videominimax-h3-regeneration-image-to-video

必填
input.prompt

The final prompt actually sent to the model when generating the 768P source video. 1–40000 chars

input.firstFrameUrl

first frame image (same as generation) HTTPS image URL or media:<uuid>

input.baseVideoUrl

MiniMax-H3 768P source video (with audio track, 24fps, dims divisible by 32, area <=768x1344, 107-362 frames). ≤ 1 items · HTTPS video URL or media:<uuid>

input.resolution

Target resolution (2K only).

常用参数
input.lastFrameUrl

last frame image (same as generation) HTTPS image URL or media:<uuid>

预估费用
正在加载价格…

任务接收时预留额度,仅成功后扣费。失败任务释放预留额度;结果未确认时会保留额度,直到核实。

请求预览

{"model": "minimax-h3-regeneration-image-to-video","input": {"resolution": "2K"}}

网页 Playground 使用你的登录会话发送此请求体。在代码中请使用 API Key 发送相同内容。

结果

运行请求后,结果会显示在这里,并附上状态和费用。

MiniMax H3 价格

各可用版本和规格的价格以美元显示,计费单位随费率注明;只有找到可匹配的官方标价时才显示参考价格。

正在加载价格…
  • Charged only on success仅对成功任务扣费。失败任务释放预留额度;结果尚未确认的任务会保留额度,直到结果核实。
  • Safe retries使用相同幂等密钥重试不会重复扣费。
  • Price changes发送 expectedPriceMicros:若展示后价格发生变化,任务会以 PRICE_CHANGED 被拒绝。

对比 MiniMax H3 各版本

每个变体都有独立的模型 ID 和输入参数。请先选择任务和版本;变体存在差异时会显示额外字段。

  • Max turbo image to video

    minimax-h3-max-turbo-image-to-video

    图生视频; 文本, 图像; 5–15 s; 480p, 768p

  • Max turbo text to video默认值

    minimax-h3-max-turbo-text-to-video

    文生视频; 文本; 5–15 s; 480p, 768p; 21:9, 16:9, 4:3, 1:1, 3:4, 9:16

    aspectRatio
  • Max image to video

    minimax-h3-max-image-to-video

    图生视频; 文本, 图像; 5–15 s; 480p, 768p

  • Max text to video

    minimax-h3-max-text-to-video

    文生视频; 文本; 5–15 s; 480p, 768p; 21:9, 16:9, 4:3, 1:1, 3:4, 9:16

    ratio
  • 图生视频

    minimax-h3-image-to-video

    图生视频; 文本, 图像; 5–15 s; 2K, 768p

    aigc_watermark
  • 文生视频

    minimax-h3-text-to-video

    文生视频; 文本; 5–15 s; 2K, 768p; 21:9, 16:9, 4:3, 1:1, 3:4, 9:16

    ratioaigc_watermark
  • Regeneration text to video

    minimax-h3-regeneration-text-to-video

    Video to video; 文本, 视频; 2K

  • Regeneration image to video

    minimax-h3-regeneration-image-to-video

    Video to video; 文本, 图像, 视频; 2K

    firstFrameUrllastFrameUrl
  • Multimodal reference to video

    minimax-h3-multimodal-reference-to-video

    Reference to video; 文本, 图像, 视频, audio; 5–15 s; 2K, 768p; 21:9, 16:9, 4:3, 1:1, 3:4, 9:16

MiniMax H3 能做什么

MiniMax describes H3 as a multimodal video model that can use text, images, video and audio and generate video with native stereo sound. MiniMax's model documentation describes up to 2K and 15 seconds; actual options depend on the Hermes variant.

The Hermes catalog separates H3, Max and Max Turbo text-to-video and image-to-video, multimodal reference-to-video, and regeneration tasks. The multimodal reference schema lists up to nine images, three videos and three audio inputs. Regeneration variants take a base video; the other variants use different fields. Check the selected variant's status, valid options and price before running.

能力

  • Up to 15 seconds

    Choose a length from 5 to 15 seconds.

    duration

    Every variant except Regeneration text to video, Regeneration image to video

  • Start and end frames

    Set the first frame and, optionally, the last frame the clip moves to.

    firstFrameUrllastFrameUrl

    版本: Max turbo image to video, Max image to video, 图生视频, Regeneration image to video

  • 480p or 768p output

    Pick the output resolution: 480p or 768p.

    resolution

    Every variant except Regeneration text to video, Regeneration image to video

  • 6 aspect ratios

    21:9, 16:9, 4:3, 1:1, 3:4 or 9:16.

    aspectRatioratio

    版本: Max turbo text to video, Max text to video, 文生视频, Multimodal reference to video

  • Up to 9 input images

    Send up to 9 images in one request as HTTPS URLs or uploaded media references.

    imageUrls

    版本: Multimodal reference to video

  • Video input

    Accepts a source video as input.

    baseVideoUrlvideoUrls

    版本: Regeneration text to video, Regeneration image to video, Multimodal reference to video

  • Audio input

    Accepts reference audio as input.

    audioUrls

    版本: Multimodal reference to video

使用场景

  1. Generate a scene

    Describe the visual action and sound for a text-to-video task.

  2. Create from references

    Use the multimodal variant to supply supported reference media.

  3. Regenerate a clip

    Select a regeneration variant when you have a base video and follow its input requirements.

用代码调用 MiniMax H3

Every image, video and music model uses the same asynchronous job API: submit a task, then poll for the result.

  1. 1

    创建 API Key

    In the dashboard, with the jobs:create and jobs:read scopes. API 密钥
  2. 2

    提交任务

    POST /api/v1/jobs with the model ID and an input object. The response is 202 with a task ID.
  3. 3

    Poll for the result

    GET /api/v1/jobs/{id} until the status is succeeded, then download the output URLs.

最小请求

Required fields only, generated from the schema of Regeneration image to video. Set HERMES_API_KEY in your environment.

模型 ID
minimax-h3-regeneration-image-to-video
端点
https://api.hermes-ai.net/api/v1/jobs
Full task API reference
# 1. Submit the task (asynchronous; returns 202 with a task id)
curl -X POST https://api.hermes-ai.net/api/v1/jobs \
  -H "Authorization: Bearer $HERMES_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: $(uuidgen)" \
  -d '{
  "model": "minimax-h3-regeneration-image-to-video",
  "input": {
    "prompt": "A lighthouse on a rocky coast at sunrise, soft light",
    "firstFrameUrl": "https://example.com/input.png",
    "baseVideoUrl": "https://example.com/input.mp4",
    "resolution": "2K"
  }
}'

# 2. Poll until status is succeeded, failed or reconciliation_required
curl https://api.hermes-ai.net/api/v1/jobs/TASK_ID \
  -H "Authorization: Bearer $HERMES_API_KEY"

输入参数

Fields of Regeneration image to video (minimax-h3-regeneration-image-to-video). 未知字段会被拒绝并返回 INVALID_MODEL_INPUT。

字段类型必填默认值可选值与约束
prompt
string必填—

The final prompt actually sent to the model when generating the 768P source video.

1–40000 chars

firstFrameUrl
string必填—

first frame image (same as generation)

HTTPS image URL or media:<uuid>

lastFrameUrl
string可选—

last frame image (same as generation)

HTTPS image URL or media:<uuid>

baseVideoUrl
string必填—

MiniMax-H3 768P source video (with audio track, 24fps, dims divisible by 32, area <=768x1344, 107-362 frames).

≤ 1 items, HTTPS video URL or media:<uuid>

resolution
string必填"2K"

Target resolution (2K only).

可选值: "2K"

先上传本地文件

用返回的媒体引用替换示例 URL。

POST /api/v1/media
curl -X POST https://api.hermes-ai.net/api/v1/media \
  -H "Authorization: Bearer $HERMES_API_KEY" \
  -H "Content-Type: image/png" \
  --data-binary @input.png
# → { "reference": "media:<uuid>", ... }  Use the reference in an image, video or audio field.
Headers, response, task statuses and errors
请求头值说明
AuthorizationBearer <API key>也可以通过 x-api-key 发送密钥。
Content-Typeapplication/json
Idempotency-Key<unique string ≤128>必填。仅在重试完全相同的请求时复用。
字段类型说明
idstringHermes AI 任务 ID,用于轮询。
modelstring模型 slug。
statusstring参见任务状态。
errorCodestring | null任务失败或需要核对时设置。
priceMicrosinteger接收时锁定的报价,美元微单位。
currency"USD"
createdAt / updatedAtstringISO 8601 时间戳。
assets[]array{ id, type, url?, text? }。url 为同源 /api/v1/results/<uuid> 链接。
replayedboolean仅提交时返回。幂等重试命中时为 true。
状态终态含义
queued否已接收,等待执行。
submitting否正在提交给模型提供方。
running否模型生成中。
succeeded是已完成,结果位于 assets。
failed是未产生结果,预留额度已释放。
reconciliation_required是结果正在核实。请保留任务 ID,不要自动重新提交。
HTTP代码含义与处理方式
401AUTH_REQUIRED缺少或无效的 API Key 或会话。
403INSUFFICIENT_SCOPEAPI Key 缺少此端点所需的权限。
403PERMISSION_DENIED此凭据无权执行该操作。
429API_KEY_RATE_LIMITED已达到 API Key 的频率限制,请遵循 Retry-After。
429API_KEY_USAGE_EXCEEDEDAPI Key 已达到使用上限,请创建新密钥。
400INVALID_REQUEST请求体不是 JSON 对象,或缺少必需值。
400INVALID_IDEMPOTENCY_KEYIdempotency-Key 必须是 1–128 个可打印 ASCII 字符。
400INVALID_MODEL_INPUTinput 与模型 schema 不匹配(未知字段、缺少必填字段或值无效)。
400 / 413INPUT_TOO_LARGE请求体超过 256 KB。请先上传媒体并发送媒体引用。
404REQUEST_NOT_FOUNDNo chat request with this ID belongs to your account.
404MODEL_NOT_FOUND没有此 slug 对应的模型。
404JOB_NOT_FOUND你的账户中没有此 ID 的任务。
409IDEMPOTENCY_CONFLICT此 Idempotency-Key 已用于不同的输入。新请求请使用新密钥。
409PRICE_CHANGED价格与 expectedPriceMicros 不一致。请重新获取价格并确认。
409MODEL_PRICE_UNAVAILABLE此模型尚未配置价格,暂不可运行。
409PAYMENT_REVIEW_REQUIRED退款后余额不足。请联系支持或补足余额后再提交任务。
409INSUFFICIENT_CREDITS可用额度不足以支付此任务。
429RATE_LIMITED每分钟提交次数过多,请稍后重试。
429EXPORT_RATE_LIMITED本分钟内导出 CSV 次数过多,请按 Retry-After 等待后重试。
429CONCURRENCY_LIMITED未完成任务过多,请等待任务完成。
503EXECUTION_UNAVAILABLE此模型当前未配置执行能力。
429UPSTREAM_RATE_LIMITEDThe model provider is rate limiting chat requests. Retry with backoff; nothing was charged.
502UPSTREAM_ERRORThe model provider failed the chat request. Nothing was charged; retry later.
503PLATFORM_BUDGET_EXCEEDED此模型暂时不可用,请稍后重试。
503EXECUTION_TEMPORARILY_UNAVAILABLE执行暂时不可用,请使用相同的 Idempotency-Key 重试。
503AUTH_UNAVAILABLE身份验证暂时不可用,请稍后重试。
503API_UNAVAILABLEAPI 暂时不可用,请稍后重试。

MiniMax H3 API 常见问题

Answers built from the model's schema and the Hermes AI API contract.

Can all H3 variants accept audio references?

No. Audio reference fields are listed for the multimodal reference-to-video variant; the other tasks have different input schemas.

What can the MiniMax H3 API do?

MiniMax 的 MiniMax H3 支持图生视频, 文生视频, 视频生视频 和 参考生视频。目录列有9 variants, each,每个变体都有自己的模型 ID 和输入参数。提交请求前请先确认当前可用状态。

Which durations, resolutions and aspect ratios does MiniMax H3 support?

Across its variants, MiniMax H3 supports durations from 5 to 15 seconds, 480p, 768p 和 2K resolution 和 aspect ratios 21:9, 16:9, 4:3, 1:1, 3:4 和 9:16. Each variant accepts its own subset; the variants table on this page lists them.

What input does MiniMax H3 accept?

A text prompt of up to 40,000 characters, plus 图像, 视频 或 audio input (up to 9 images per request). Media fields take a public HTTPS URL or a media:<uuid> reference returned by POST /api/v1/media, so local files can be uploaded first. 未知字段会被拒绝并返回 INVALID_MODEL_INPUT。

How is MiniMax H3 priced and billed?

You pay per successful task from prepaid credits, with no subscription. The price depends on the variant and the specification you choose, such as resolution or duration; see the pricing table on this page or the pricing page. Credits are reserved when a task is accepted and charged only when it succeeds; a failed task returns the reservation to your balance.

Which MiniMax H3 variant should I choose?

For 图生视频, Max turbo image to video, Max image to video 和 图生视频 differ in fields: 图生视频 adds aigc_watermark; for 文生视频, Max turbo text to video, Max text to video 和 文生视频 differ in fields: Max turbo text to video adds aspectRatio; Max text to video adds ratio; 文生视频 adds ratio, aigc_watermark; for video to video, Regeneration text to video 和 Regeneration image to video differ in fields: Regeneration image to video adds firstFrameUrl, lastFrameUrl; for reference to video, use Multimodal reference to video. Pick the task first, then compare the tiers in the variants table.

How do I call the MiniMax H3 API from code?

Send POST https://api.hermes-ai.net/api/v1/jobs with the variant's model ID and an input object, then poll GET /api/v1/jobs/{id} until the status is succeeded or failed. The API section has cURL, JavaScript and Python examples generated from the schema.

Can I try MiniMax H3 without writing code?

Yes. Sign in and use the playground on this page: it sends the same request as the API with your session and charges the same price. Code needs an API key from the dashboard.

What happens when a task fails or I hit a rate limit?

A failed task is not charged and its reserved credits are returned. Too many submissions per minute return 429 RATE_LIMITED, and too many unfinished tasks return 429 CONCURRENCY_LIMITED; retry later or wait for a task to finish.