From Static Frames to Final Cut via AI Interpolation

What it does StoryGen Atelier is a full-stack tool that generates storyboard scripts and images with Gemini, then bridges each adjacent frame into a motion clip using Vertex AI Veo. A backend Node.js service coordinates the whole chain and losslessly stitches the results into a final MP4 with ffmpeg. It also persists logs and manages a local gallery of generated storyboards and videos.

The interesting bit The project treats video generation as an interpolation problem rather than a prompt-engineering marathon. It uses a sliding window to feed pairs of storyboard frames to Gemini, which writes specific transition prompts and durations; Veo then generates the connecting clips in parallel, and ffmpeg concatenates them without re-encoding. The result is a coherent narrative from static keyframes without manual timeline editing.

Key highlights

• Uses an “Interpolation Chain” strategy: Gemini analyzes shot pairs to generate transition prompts and durations automatically.
• Calls Vertex AI Veo in parallel to generate intermediate clips between storyboard frames, plus a dedicated closing shot.
• Stitches everything via ffmpeg concat with stream copy (-c copy) to avoid generational quality loss.
• Supports custom visual styles (Anime, Cyberpunk, Ghibli, Realism, Chinese Ink) and configurable shot counts.
• Full-stack React frontend with a SQLite-backed dashboard for logs and gallery management.

Caveats

• Hard dependency on Google Cloud (Vertex AI API) and Gemini API keys; no fallback providers.
• Built around specific preview model versions (gemini-3-pro-preview, veo-3.1-generate-preview) that may shift or break without warning.
• Requires local ffmpeg installation alongside the Node.js backend.

Verdict Worth a look if you want to experiment with automated narrative video pipelines using Google’s generative stack. Skip it if you need provider flexibility or are trying to avoid preview API dependencies.