Comic Generation Pipeline

My senior capstone: an agentic pipeline that turns a story prompt into a finished multi-page comic. A LangChain agent plans panels and characters, generates the art with Leonardo AI while keeping characters consistent across panels, composes pages with automatic layouts, and adds speech bubbles, narration, and sound effects. A human stays in the loop through a chat interface at every step. Before choosing a generator I benchmarked SDXL, PixArt-Σ, Flux Pro, and Nano Banana on prompt adherence, style referencing, and context-image guidance. Batch generation cut manual iteration time by more than 80% for four-page comics.

LangChain agent with tool calls for planning, characters, and imagesLeonardo AI Phoenix 1.0 image generationGoogle Gemini for narrative and dialogueCloudinary upload and CDN display

View Code

Note: Documentation is available; a live demo is not hosted for this project.

Project Features

🗂️

Prompt to Panels

Describe a story and the agent breaks it into panels with shot types (establishing, medium, close-up) and scene descriptions before any image is generated.

🧑‍🎤

Character Consistency

Characters are generated once, then referenced across every panel so faces, outfits, and proportions hold from page one to the last frame.

📐

Smart Page Layouts

Automatic composition with multiple layout options per page, plus speech bubbles, narration boxes, and sound effects placed by the pipeline.

🧪

Model Bake-off

Tested SDXL with LoRA, PixArt-Σ, Flux Pro, and Nano Banana for sequential art. Flux Pro with context images held anatomy and lighting best across panels.

Project Info

Duration

10/2025 – 12/2025

Role

Pipeline & Agent Developer

Team Size

Capstone Team

Status

Completed

Technologies Used

• LangChain agent with tool calls for planning, characters, and images
• Leonardo AI Phoenix 1.0 image generation
• Google Gemini for narrative and dialogue
• Cloudinary upload and CDN display
• React + TypeScript + Tailwind CSS front end
• Framer Motion for the comic grid
• Sharp and Canvas for compositing

Project Gallery

Void's Echo – Cover

Generated cover page for a sci-fi short. Title lettering and composition come from the layout step.

Void's Echo – Page 2

Same astronaut, same suit, across four panels. This is the character-consistency pass at work.

Desert Encounter – Page 1

A darker, painterly style selected by the agent for a western-noir prompt.

Desert Encounter – Page 2

Four-panel grid with alternating close-ups and wide shots.

Liam's Promise – Cover

A romance prompt produces a warm anime-inspired palette, chosen without manual style tuning.

Liam's Promise – Page 2

Dialogue bubbles and narration placed by the pipeline over the generated panels.

Coffee & Second Chances – Cover

A slice-of-life prompt, showing the range of styles a single pipeline can produce.

Interested in This Project?

I'm always excited to discuss my projects and share insights about the development process. Feel free to reach out if you'd like to know more!

Get In Touch View More Projects