SparkMV
From one sentence or one link to a finished short video
An end-to-end production line for MCN agencies, e-commerce and short-drama teams. Pick a structure from 31 proven formats (20 general plus 11 drama/comic/novel scenarios), or paste a video link and let the system analyse its hook, structure and pacing before reusing it. Work advances through eleven stages - analyse, script, storyboard, voice, music, visual, generation, compose, polish, QC, export - each individually rerunnable with variant lineage. Characters, scenes and props live in an asset library; replacing one image rewrites the storyboards bound to it. Neural TTS produces real Chinese voices with per-character timbres. The finished video is visually reviewed: frames are scored by a multimodal model and an all-zero score triggers a retry on another model. When no external engine is configured, a local engine still delivers a video.
Batch mode splits a whole novel into episodes by chapter (up to 100 per task) and renders differentiated copy for the same theme; jobs run concurrently in a queue and worker restarts reclaim orphaned tasks.
Core capabilities
- 31 proven formats with hook, structure and pacing templates
- 11-stage pipeline, rerunnable per stage with variant lineage
- Link cloning: analyses structure from major short-video platforms
- Asset library: character/scene/prop settings, auto image generation, rewrite on image swap
- Real Chinese narration with per-character voices
- Visual QC: frame-level scoring, automatic model fallback on anomaly
- Batch mode: split a novel into episodes by chapter, concurrent rendering
- Local fallback engine when no external key is configured
Who uses it
- Matrix production for MCN accounts
- E-commerce selling videos and copy rewriting
- Novel or script adapted into serialized drama episodes
- Talking-head and illustrated videos for educators
Delivery models
SaaS · on-premises · source licence · white label
Tech stack
Go（API + Worker）· Vue3 · Electron · uni-app · MySQL / Redis / MongoDB / RabbitMQ