ve7 — Dialogue Manga Shorts Mass Production
A pipeline that, given just a dialogue script, automates panel image generation, line-by-line TTS, vertical video rendering with Remotion, and posting to YouTube/TikTok.
Duration
In-house development
Team
1 person
01 // OVERVIEW
Project Overview
A pipeline that mass-produces manga short videos in the "smart girl x dumb guy" two-character format popularized by Fermi Manga University-style channels. Aside from a single Claude operation to create the dialogue script (panels.json), everything else runs automatically via scripts. Panel images are generated with gpt-image-2, TTS is handled by Fish Audio, and Remotion assembles the final 1080x1920 vertical video for posting to YouTube/TikTok.
02 // KEY FEATURES
Key Features
Panel Image Generation with gpt-image-2
Based on each panel definition in panels.json (speaker, emotion, pose, line), gpt-image-2 generates two panels per image in a single batch, which are then automatically split at detected white borders into individual panel images. Speech-bubble text is baked directly into the image.
Character TTS with Fish Audio
Each line is converted into a character-specific voice using Fish Audio (s2-pro). Panel duration is driven by audio length, with timestamps recorded in durations.json and used to build the timeline.
Vertical Video Rendering and Posting with Remotion
The MangaShort component reads timeline.json and renders a vertical video (1080x1920) with pop-in, slow-zoom, and emotion-specific shake effects. Supports posting via the official YouTube Data API and TikTok (via an unofficial cookie-based method).
03 // TECHNICAL DEEP-DIVE
Technical Highlights
Solution
Maintaining Per-Cast Art Style Consistency with gpt-image-2
Manages multiple casts - chibi, ink, suisai, and jito - and, using cast settings and reference images in config.json, ensures gpt-image-2 generates the same art style every time. Different casts are assigned to different accounts, keeping each channel's visual world consistent.
04 // IMPACT & RESULTS
Results
Fully automated
One script file to finished vertical video
1080x1920
Vertical video formatted for YouTube/TikTok
Multiple casts
Manages character sets by art style