sjtuplayer / showvi

What is showvi?

Showvi is an end-to-end AI video production system written in Python that transforms a simple idea, novel excerpt, or reference video into a finished film with minimal human intervention. The system chains together multiple specialized AI agents: one generates the screenplay, another tracks character consistency across scenes, a third handles storyboarding, and a VLM-based critic reviews each generated clip before it passes or gets automatically rewritten. Video output runs on Seedance 2.0 through the Jimeng web interface, with the agent handling browser automation, queue management, and retry logic entirely on its own. A web dashboard lets you monitor production, tweak prompts, and manage assets without touching the command line.

Why is it gaining traction?

The standout feature is the self-healing production loop. Every generated clip gets scored by Gemini across six dimensions, and failures automatically trigger prompt rewriting and regeneration until quality thresholds are met. For longer videos, the system maintains character and scene continuity using a 16-grid keyframe passing technique and cross-segment state tracking, which solves the "character face changes between shots" problem that plagues most AI video tools. The checkpoint-and-resume system means a crash does not wipe progress mid-generation. Developers are drawn to the declarative skill pipeline, which lets you add new video models by dropping in a YAML file rather than rewriting core logic.

Who should use this?

Content creators who want to turn scripts or story chapters into video without managing a complex toolchain. Indie game studios prototyping narrative cutscenes from text descriptions. Developers building automated video pipelines who need a reference architecture for multi-agent orchestration. It is less suited for one-off video edits or users without access to a Jimeng account and Gemini API keys.

Verdict

Showvi is technically impressive and well-architected, but with only 35 stars and a credibility score of 0.8999999761581421%, it is early-stage software with limited community validation. The documentation is solid and the web dashboard works, but test coverage and production hardening are unclear. Worth exploring as a proof-of-concept for AI-driven video pipelines, but do not bet a production workflow on it without running your own evaluation first.