AI Avatar Promo Studio
数字人口播视频的全链路生产
A talking-head video is fixed as a seven-step pipeline; each step records its own state, so a job can resume where it failed and be retried. Rewriting offers four styles (viral hook, professional, casual, custom); voice covers ten providers - three hosted, seven fully local - and the avatar is lip-synced with an open-source digital-human stack. Uploads accept files up to 500 MB. Publishing runs through local browser automation on four platforms (Douyin, Kuaishou, Channels, Xiaohongshu), leaving the session on the customer machine.
The backend is Flask + Celery over MySQL, Redis and RabbitMQ: 26 tables and 11 API modules. The frontend is Vue 3 + Element Plus with 15 views, plus an Electron desktop app and a uni-app mobile client (6 routes, buildable to APK). Multi-user accounts carry four tiers with quotas and a licence SDK.
Stated plainly: there is no public demo instance today - it is deployed per customer. And the rights to any avatar likeness or cloned voice must be secured by the party using them; we supply the technology and the deployment.
Core capabilities
- Seven-step pipeline with per-step state: resume and retry
- Four rewrite styles: viral, professional, casual, custom
- Ten voice providers: 3 hosted, 7 deployable fully local
- Open-source avatar lip-sync, optional voice cloning
- Publishing to Douyin, Kuaishou, Channels and Xiaohongshu
- Browser sessions stay on the local machine
- Web + Electron desktop + uni-app mobile from one codebase
- 26 tables, 11 API modules, 15 web views
- Multi-user with four quota tiers and a licence SDK
Who uses it
- Brand and product pitch channels
- Explainer and training content
- Store and regional marketing matrices
- High-volume e-commerce talking heads
Delivery models
On-premises · source licence · hybrid cloud
Tech stack
Python + Flask + Celery · Vue3 + Element Plus · Electron · uni-app · MySQL / Redis / RabbitMQ