LiveAIAI Creative Studio
BLACKBOLT
AI Comic Adaptation Studio
Turns static comic pages into voice-cast, scored audio experiences through a six-layer AI pipeline with human review gates.
The problem
Comics are silent by default, and turning one into a directed, voice-cast audio experience by hand takes hours — or days — per issue.
The approach
- Built a 6-layer pipeline: ingestion → vision → comic understanding → production setup → audio production → delivery.
- Inserted two human review gates: Character Review lets the user fix casting/speaker attribution; Scene Review lets them adjust mood and pacing before audio is generated.
- Progressively generates scenes so users see and hear results as they go.
Architecture
- Ingestion: CBZ/CBR/PDF/images → normalized, ordered pages
- Vision: panel, bubble, reading-order, and text detection
- Comic Understanding: character ID, speaker attribution, emotion, scene definition
- Production Setup: voice casting per character, music per scene
- Audio Production: TTS, ambience, stitching
- Delivery: progressive scene generation → shareable Experience
Impact
The most ambitious side project in the portfolio
Sits at the intersection of AI agent orchestration, Marvel/comics fandom, and creative tooling
What I learned
- Multi-stage AI pipelines need explicit review gates; full automation degrades where attribution is ambiguous.
- Voice casting and music selection are editorial decisions; humans should stay the director.
Tech
TypeScriptAI/ML APIsAudio ProcessingComic Parsing