Skip to content

H100 clusters now available — 30% off your first month, plus 300 bonus video creditsH100 集群现已开放 — 首月 7 折,另赠 300 视频积分

H100 available now · 6 regionsH100 现已开放 · 6 大区域

Enterprise AI compute.Cinematic video, in yourprivate cloud.企业级 AI 算力。电影级视频生成,就在你的私有云。

From model training and inference to batch AI video generation — one high-performance compute platform. Your data never leaves your domain. On-demand GPUs, billed by the second.从模型训练、推理到批量 AI 视频生成,一套高性能算力平台。数据不出域,GPU 按秒计费、随取随用。

GPUs online在线 GPU

12,400+

Instance spin-up实例启动

<60s

SLA uptimeSLA 可用性

99.99%

Total compute总算力

3.2 EFLOPS

01Platform平台

One platform for the full AI lifecycle一个平台,覆盖 AI 全生命周期

Train, fine-tune, deploy, scale — without stitching four vendors together.训练、微调、部署、扩容 —— 不必再把四家供应商拼在一起。

  • 01

    Spin up in seconds秒级启动

    Containerized GPU instances are ready in under 60 seconds, with cold starts as low as 200ms. No queue, no ticket, no waiting in line.容器化 GPU 实例 60 秒内就绪,冷启动低至 200 毫秒。不排队、不提单、不等审批。

  • 02

    Training clusters训练集群

    NVLink inside the node, 3.2T InfiniBand between them. Better than 95% linear scaling across 1,000+ GPUs, with checkpoint resume on interruption.节点内 NVLink,节点间 3.2T InfiniBand。千卡以上线性加速比优于 95%,中断可从检查点续跑。

  • 03

    Serverless inference无服务器推理

    Autoscale from zero to thousands of workers in seconds. Pay per request, and pay nothing while idle.数秒内从 0 弹性扩容到数千并发。按请求计费,闲置不产生费用。

  • 04

    Secure & compliant安全合规

    Tier-certified data centres, VPC isolation and dedicated lines — built to clear a finance-grade compliance review.等保认证机房、VPC 隔离与专线接入 —— 面向金融级合规审查设计。

Cluster集群Live实时
H100 queueH100 排队
4 min4 分钟
A100 queueA100 排队
instant无需等待
Inference workers推理工作节点
1,2481,248
Regions可用区域
66
H100 SXMHigh demand需求旺盛
A100 SXMAvailable现货充足
L40SAvailable现货充足
RTX 4090Available现货充足

Updated continuously持续更新

02Video Studio视频工作室

Your own video model, deployed in your private cloud属于你的视频模型,部署在你的私有云

The same studio experience your team already wants — running entirely inside your perimeter.团队想要的那套创作体验 —— 完整运行在你自己的网络边界之内。

Hanqing V3 (self-hosted)Balance余额 29,375 credits积分

Model模型

Hanqing V3 (self-hosted)瀚擎 V3(自托管)

Prompt提示词

A wuxia short film. Deep in a bamboo grove, morning mist still hanging, the camera follows a wandering swordsman…一部武侠短片。竹林深处,晨雾未散,镜头跟随一位独行剑客……

Reference images参考图 3/930 credits30 积分
Preview预览Key frame关键帧Done · image-to-video · 44 minutes ago已完成 · 图生视频 · 44 分钟前

A wuxia short film. Deep in a bamboo grove, morning mist still hanging, the camera follows a wandering swordsman…一部武侠短片。竹林深处,晨雾未散,镜头跟随一位独行剑客……

1080P · 30S
Storyboard分镜
12 shots · 30s12 镜 · 30 秒
Reference images参考图
3 / 33 / 3
Video segments视频片段
12 / 12
  • 01

    Cinematic text- & image-to-video电影级文生视频 / 图生视频

    Up to 30s of continuous narrative at 1080P, with multi-shot character and style consistency.1080P 下最长 30 秒连续叙事,跨镜头保持人物与风格一致。

  • 02

    Data never leaves your domain数据不出域

    Weights, prompts and renders stay on your hardware. Air-gapped deployment supported.权重、提示词与成片全部留在自有硬件上,支持完全离网部署。

  • 03

    Batch production pipeline批量生产流水线

    A first-class API and job queue — tens of thousands of marketing videos per day, unattended.一等公民的 API 与任务队列 —— 每天数万条营销视频,无人值守。

  • 04

    Brand-style fine-tuning品牌风格微调

    Train a custom LoRA on your own footage, products and talent so every render stays on brand.用自有素材、产品与出镜人训练专属 LoRA,每一条成片都守住品牌调性。

03Showcase案例

Made on Hanqing由瀚擎生成

One finished film and five frames from recent customer work. Every shot generated in a private deployment.一支完整成片,加上近期客户作品中的五帧。每一镜都产自客户自己的私有化部署。

Featured film精选成片

Orbit crew → brand planet太空舱 → 品牌星球

1080P · 21s · with sound1080P · 21 秒 · 含声音

01

Cyberpunk neon rain赛博霓虹雨夜

Vertical 1080P · 10s竖屏 1080P · 10 秒

02

Bamboo grove at dawn竹林晨雾

Vertical 1080P · 10s竖屏 1080P · 10 秒

03

Brand product close-up品牌产品特写

Vertical 1080P · 6s竖屏 1080P · 6 秒

04

City aerial timelapse城市航拍延时

Vertical 1080P · 15s竖屏 1080P · 15 秒

05

Studio portrait relight棚拍人像重打光

Vertical 1080P · 8s竖屏 1080P · 8 秒

Drag to explore拖动查看
01 / 05
04GPUs算力

On-demand GPUs. Pay as you go, billed per second.GPU 随取随用,按秒计费。

No minimum commitment. Reserved capacity available for sustained training runs.无最低消费门槛。长周期训练可预留专属容量。

Estimate your run估算你的用量

GPU countGPU 数量

8

Duration时长

H100 SXM × 8 GPUs for 24 hoursH100 SXM × 8 卡 · 24 小时

Estimated total预计总价

$536

Billed per second — stop the instance and billing stops with it. Prices exclude tax.按秒计费 —— 实例一停,计费同时停止。价格未含税。

05Pricing价格

Transparent pricing, your choice价格透明,任你选择

Subscribe for video generation, or go with pure compute on on-demand and reserved plans.视频生成可订阅,纯算力可按量或预留 —— 两条路都走得通。

Light轻量版

For individual creators and first pilots.面向独立创作者与初次试点的团队。

$49/mo/月

  • ~500s of 720P video generation约 500 秒 720P 视频生成
  • Text-to-video / image-to-video文生视频 / 图生视频
  • Standard queue标准队列
  • Community support社区支持
Start now立即开始
Most popular最受欢迎

Pro专业版

For in-house content teams shipping every week.面向每周都要出片的内容团队。

$299/mo/月

  • ~5,000s of 1080P video generation约 5,000 秒 1080P 视频生成
  • Batch API · priority queue批量 API · 优先队列
  • 1× brand-style LoRA fine-tune1 次品牌风格 LoRA 微调
  • Dedicated account manager · 5×8专属客户经理 · 5×8
Subscribe立即订阅

Enterprise企业版

For regulated industries and production at scale.面向强监管行业与规模化生产。

Custom quote定制报价

  • Private deployment, data stays in-domain私有化部署,数据不出域
  • Unlimited generation生成量不设上限
  • Dedicated GPU cluster + custom models专属 GPU 集群 + 定制模型
  • 24/7 SLA · on-site delivery7×24 SLA · 驻场交付
Contact sales联系销售
06Private Cloud私有云

Turnkey private AI infrastructure, from racks to models从机柜到模型,交钥匙的私有 AI 基础设施

End-to-end service: IDC hosting, GPU cluster build-out, Kubernetes scheduling, private video-model deployment and managed operations — delivered in 30 days.端到端交付:IDC 托管、GPU 集群搭建、Kubernetes 调度、私有视频模型部署与托管运维 —— 30 天上线。

Delivery交付周期
30 days30 天
Data centre数据中心
Tier-certified等保认证
Orchestration编排调度
KubernetesKubernetes
Operations运维
24/7 on-site7×24 驻场
  1. 01

    Survey & design勘测与设计

    Power, cooling and network audit, then a rack-level design signed off against your compliance requirements.电力、制冷与网络勘测,随后输出通过合规评审的机柜级设计方案。

  2. 02

    Build & burn-in搭建与烤机

    Hardware staged, cabled and burned in for 72 hours. InfiniBand fabric validated before handover.硬件上架、布线并连续烤机 72 小时,InfiniBand 网络验证通过后才移交。

  3. 03

    Deploy & tune部署与调优

    Kubernetes scheduling, your video model deployed in-domain, throughput tuned against your real workload.部署 Kubernetes 调度,模型在域内落地,并按你的真实负载压测调优。

  4. 04

    Operate持续运维

    24/7 monitoring, spare-parts SLA and quarterly capacity reviews. Your team keeps root.7×24 监控、备件 SLA 与季度容量复盘。root 权限始终在你手里。

Compute, quietlyextraordinary.算力,安静而非凡。

Free credits on sign-up. Private deployment, quoted in a day.注册即赠免费额度。私有化部署,一个工作日内给出报价。