taruma/SceneFlow

Sync screenplays with video. Evaluate AI-generated footage. Analyze script-to-screen fidelity.

What it solves

SceneFlow is designed for AI filmmakers to synchronize screenplays with generated video content. It solves the problem of evaluating prompt adherence—specifically, determining how accurately an AI video model visualized specific instructions from a script, such as camera movements, VFX, and character actions.

How it works

The tool provides a specialized workstation that links specific script segments (cues) to video timestamps. Users can author these cues manually in a 3-panel edit studio or generate them using multimodal AI models like Google Gemini via a built-in JSON schema and system prompt. Once synced, the script highlights in real-time during playback, allowing filmmakers to pinpoint missed prompt elements or continuity deviations.

Who it’s for

It is primarily for AI cinema creators and prompt engineers who need to assess and iterate on AI-generated footage against a written screenplay.

Highlights

  • Dual Script Workflows: Supports both traditional screenplay formatting and high-precision "Auteur Scripts" with staging metadata.
  • Multi-Track Sync Timeline: A visual, NLE-inspired timeline with color-coded lanes for different cue types (Dialogue, Action, Camera, VFX, etc.).
  • LLM Sync Setup: Includes a built-in Gemini system prompt and JSON schema to automate the creation of structured sync cues.
  • 3-Panel Edit Workstation: An integrated environment featuring a media preview, a reading canvas, and a draggable cue inspector with micro-nudge timing controls.
  • Accessibility: Includes a Protanopia/Deuteranopia safe mode for color-coded cues.

Related

  • Project
    web-infra-dev/midscene-skills

    A vision-driven automation library that enables natural-language control of UI across web, desktop, and mobile platforms using screenshots.

  • Project
    pwilkin/trellis.cpp

    A standalone C++ implementation of the TRELLIS image-to-3D pipeline using GGML, enabling the generation of textured 3D models from images without Python at runtime.

  • Project
    Aperivue/medsci-skills

    MedSci Skills is an open‑source collection of 59 AI‑agent “skills” that automate the full clinical‑research pipeline – literature search, study design, data cleaning, statistical analysis, manuscript drafting, compliance checking, and, for AI studies, reproducible model‑scaffolding and validation. Designed for Claude Code, Copilot, Cursor, etc., it installs via `npx`, GitHub CLI, or a Claude plugin, and provides end‑to‑end demos (diagnostic accuracy, meta‑analysis, epidemiology, CNN segmentation, external validation) that output manuscripts, figures, and guideline‑compliant audit reports.

  • Project
    SamurAIGPT/AI-Influencer-Generator

    An open-source pipeline that uses Stable Diffusion, gTTS, and SadTalker to create and animate consistent virtual AI influencers for social media.