AI Video Generator · All-in-One Video Production Platform
Three independent production lines: Product Intro / Short Drama Draft / Digital Human Narration — an end-to-end AI video factory from script to final cut
The AI Video Generator is a full-stack video factory covering three independent production lines: Product Intro Videos (Remotion deterministic template rendering), Low-cost Drama Drafts (script → storyboard → per-shot Wan generation → screening), and Digital Human Narration (TTS → face swap → lip sync → denoise subtitle → one-click export). Built on React + FastAPI with LLM copy generation, Edge-TTS speech synthesis, Remotion cinematic rendering, WanGP/face_swap/lip_sync digital human pipeline, and an in-browser refine editor. Supports Web / Electron desktop / PWA mobile, plus GPU cluster management and job queue monitoring backend.
Core Modules
Choose a Production Line
Drama Draft Studio
Digital Human Studio
Unified Works Management
In-Browser Refine Editor
GPU Cluster & Job Management
Design Philosophy
Split "making video" into three specialized production lines — each focused on one scenario, non-interfering; sharing GPU cluster and asset library at the bottom, optimizing UX independently on top.
Three Parallel Lines, Each Specialized
Product Intro uses Remotion deterministic templates (high-quality reusable); Drama uses script→storyboard→Wan per-shot generation (fast low-cost iteration); Digital Human uses TTS+face swap+lip sync (one-stop narration). All three share the same user system, asset library, and GPU cluster.
Copy → Video Full Pipeline Automation
LLM auto-generates narration copy → Edge-TTS synthesizes speech (free multi-voice) → ffmpeg/Remotion renders video → auto quality check (duration/coverage/consistency). User only inputs product description or uploads assets; everything else is automatic.
Web + Desktop + Mobile
React frontend runs simultaneously as Web browser app, Electron desktop app (native menus/file dialogs/auto-update), and PWA mobile app (add to home screen/offline cache/responsive) — one codebase, three platforms.
GPU Cluster + Task Queue
Server management dashboard monitors GPU cluster status and job queue in real time. Supports face_swap / lip_sync / wan_draft capability tags; tasks route to matching nodes; heartbeat auto-offlines faulty nodes.
Core Capabilities · Three Production Lines Explained
Each production line is deeply optimized for one video scenario, forming a closed loop from input to output.
Product Intro Video Line
Remotion deterministic template rendering, high-quality commercial-ready. Supports portrait 1080×1920 / landscape 1280×720, duration 15-90s configurable. LLM generates TTS copy → multi-voice synthesis → BGM auto-match (9 moods→5 tracks) → ASS subtitle burn-in → auto quality check.
Short Drama Draft Line
Script→storyboard→per-shot Wan generation→draft screening rapid iteration pipeline. Input premise (≤200 chars) → auto storyboard → Wan v12 / local WanGP per-shot render → draft layer fast output / backend refine / auto-filter shots. Ideal for early creative validation.
Digital Human Narration Line
Full digital human pipeline: TTS → face swap → lip_sync → denoise → BGM → subtitles → export 1080P MP4. Supports custom avatar training (upload reference images), green-screen keying for transparent WebP export, multi-voice switching (1080P/720P).
In-Browser Refine Editor
Remotion-based online video editor requiring zero software installation. Supports 15+ format media import, multi-track timeline editing, canvas property fine-tuning (position/size/rotation/opacity/anchor), precise frame-skip preview, MP4 export.
Unified Works Hub
All three lines' outputs managed centrally in "My Works", tabbed by production line. Each record shows thumbnail/title/status tags (Draft ready/Refined/Completed)/timestamp, with preview/refine/download/batch-template-refine actions.
Server Management & Ops
GPU cluster management (server registration/capability tags/heartbeat detection/online status/offline delete), job queue monitoring (job ID/capability type/linked tasks/queued/completed/error logs), user management, dashboard overview.
Product Intro Video Generation Pipeline
Using the Product Intro line as example: full pipeline from product description input to 1080P final video output.
Key Metrics
Core indicators from architecture design to user experience.
Technical Breakthroughs
Not just API calls — every step of video production is engineered, observable, and controllable.
Shared Bottom Layer, Independent Top Layers
Three lines share auth/asset library/GPU cluster/task queue infrastructure, but each has its own UI flow, render engine, and parameter config — adding new lines does not affect existing ones.
Self-Built Full Digital Human Pipeline
TTS → face swap → lip sync → denoise → subtitle → export fully self-integrated, no dependency on external SaaS platforms — data sovereignty, near-zero marginal cost (Edge-TTS is free).
Dual Render Engine Adaptation
Standard scenes use ffmpeg fast compositing (second-level output); premium scenes use Remotion cinematic template (high-fps/high-design/shot recipe cards) — choose by need, no wasted compute.
In-Browser Post-Production
Remotion-based online editor lets users complete import/edit/tune/export without leaving the browser — eliminating the "generate→download→open PR→re-edit→re-export" toolchain breakage for a closed-loop experience.
Business Value
Convert technical capability into quantifiable, reusable business value.
Content Under Active Update
Product content has been published based on internal materials. The following areas are planned for further development:
Other Products & Solutions
Explore Xianma AI solutions in other domains