Back to Briefings AI Comic Studio - Project Briefing

AI Comic Studio - Project Briefing

Project Type: Research Collaboration / Technical Build Website: aicomicstudio.app Backup URL: graphic-novel-maker.web.app (used when school firewalls block primary domain) Lead Researcher: Cathy Chase (Senior Research Scholar, SAL Digital Learning Initiative) Reuben’s Role: Technical build + logging infrastructure Status: Active - Building


1. Project Overview

What It Is

AI Comic Studio is an AI-powered comic/graphic novel creation tool designed for classroom research. Students use generative AI to create comic panels while the tool logs detailed usage data for Cathy Chase’s Learning Sciences research. The tool runs primarily on Chromebooks in school environments.

The project originated in December 2024 when Dan Schwartz caught Reuben in the hallway during a trip to campus and asked if he could build something — details were sparse. Reuben took the loose brief and built a prototype of what he thought would be cool: a story-driven tool where students chose from the seven story archetypes and the tool generated 5 panels following that arc, which students could then customize. That approach was eventually dropped in favor of the current design: students choose an art style and build their comic one panel at a time, up to 16 panels. Cathy Chase took over as the research lead and the tool has been through multiple iterations of both the research design and the underlying technology stack since then.

The Research Question

Cathy’s research focuses on how students interact with AI-supported creative tools — specifically, how they use resources (PDFs, videos, web searches) during the creative process, how long they engage with different materials, and what patterns emerge in their usage of AI-generated imagery for storytelling.

Reuben’s Contribution

Reuben serves as the technical builder, responsible for:

  • Building and maintaining the web application
  • Designing and implementing the logging/tracking infrastructure
  • Managing the image generation stack (model selection, visual consistency)
  • Deploying updates based on Cathy’s research requirements
  • Exporting and preparing usage data for analysis
  • Supporting school testing sessions (troubleshooting firewalls, Chromebook issues)

This is classified as a Builder operating mode (medium defensibility). The value isn’t just in the code — it’s in translating research requirements into technical instrumentation that produces usable data.


2. Key People

Person Role Context
Cathy Chase PI / Research Lead Senior Research Scholar at SAL. Stanford GSE alum (PhD Learning Sciences & Technology Design, 2011). Former Assistant Professor at Columbia Teachers College. 15+ years in Learning Sciences. Notable for “Teachable Agents and the Protege Effect” with Dan Schwartz. Contact: cchase@stanford.edu
Marisol Liu Research Collaborator Analyzing February log data. Joining Mar 18 meeting
Josue Vazquez Jr School Contact Participated in Jan 22 tech test

3. Technical Architecture

Image Generation Stack

The image generation backend has evolved through multiple providers:

  1. Gemini (initial attempt) — tried first but didn’t produce reliable results
  2. GPT-4 Image (fallback) — switched to OpenAI’s GPT-4 image generation when Gemini failed; used during the 2025 study
  3. Gemini “Nano Banana” (current) — migrated back to Gemini’s improved image generation model; now the active provider

Visual consistency approach: Previous comic frames are fed back into the image generation prompt so new panels maintain consistent character appearances, settings, and art style across the story.

Application Features

  • 6-digit student ID system — tracks individual students across sessions; returning students resume previous work
  • Resource library logging — open/close times, idle tracking (5-second threshold), scroll measurement
  • Scene setting system — simplified to two options: “Same setting as last scene” / “New setting”
  • Save triggers — fires on panel reorder and login session persistence
  • Resource tracking — PDFs, video, web searches; modal close detection; blocks resource access during comic creation
  • Moderator bot — AI moderation layer; code later shared with Sarah Williams-Habibi for VFT project
  • Chromebook-first — designed for school Chromebook environments
  • Firebase hosting — primary domain aicomicstudio.app; backup graphic-novel-maker.web.app

4. Development Timeline

Date Event
Dec 2024 Dan Schwartz catches Reuben in a hallway, asks if he can build something (sparse details). Reuben builds a prototype on his own initiative — story-driven (7 archetypes → 5 pre-generated panels to customize). Gemini image generation tried first but unreliable; switches to GPT-4 Image
Early 2025 Prototype evolves: story archetype approach dropped in favor of art style selection + free-form panel-by-panel creation (up to 16 panels). Cathy runs first study
May 2025 Referenced in Josh 1:1 — “Comic-book maker - liase”
Q4 2025 Ongoing tool building and refinement; listed alongside VFT as active technical build
Jan 13, 2026 New phase: logging infrastructure requirements meeting with Cathy. Scoped: student ID system, Chromebook-first, resource library logging. Migrated image generation to Gemini “Nano Banana” with frame-to-frame visual consistency
Jan 21 Test plan created and shared (test plan doc). Save process bug fixes ready
Jan 22 Tech test at school — successful. School firewall blocked primary URL; used backup. Three fixes identified: (1) resume previous work on same student ID, (2) save on panel reorder, (3) track time on resource search results
Jan 26 Style consistency update pushed to production. Dropdown simplified per Cathy’s feedback
Week of Jan 26 All three fixes deployed (dropdown simplification, login persistence, panel reorder save)
Jan 31, Feb 3, Feb 5 School testing sessions with students
Mar 13 Cathy requests February data export via Slack. Moderator bot code shared with Sarah Williams-Habibi for VFT
Mar 18 Upcoming: Meeting with Cathy + Marisol Liu to analyze February log data (Wed 2pm)

5. Current Status & Next Steps

The project is transitioning from build/test to data analysis. School testing is complete.

Immediate

  • Export February usage data for Cathy (action item — before Mar 18 meeting)
  • Mar 18 meeting with Cathy + Marisol Liu to review log data and determine analysis approach

Longer-term

  • Research publication from log data findings
  • Follow-up study design
  • Potential new tracking: student usage of research/web searches
  • Tool updates based on research findings

6. Strategic Value

Cross-Project Reuse

Moderator bot code was adapted for the VFT project. Sarah Williams-Habibi’s response: “Thank you so much digging into AI moderation and sending this our way!! I’ve learned so much already!!” (Mar 13, 2026) — a concrete example of [TOOLBOX] and cross-project knowledge transfer.

Re-invigorating Stalled Research

Cathy told Reuben something to the effect of: “You really re-invigorated the project with this. It had stalled.” (paraphrased — exact quote not captured). The hallway prototype from a vague Dan Schwartz ask turned into an active research program. That’s a strong [STUDIO-BUILD] story.

Research Infrastructure

Demonstrates the Accelerator Studio’s ability to build real logging and tracking directly into tools — not just prototypes, but research-grade instrumentation. This is a story worth telling in end-of-month meetings.

Publication Pathway

Cathy expects research findings from the log data, with potential for published papers and follow-up studies. A publication co-authored with a Stanford research scholar is high-value evidence for the Studio’s impact.

Relationship Value

The collaboration with Cathy is classified as “high-value” in capacity analysis. She also connects to Build-a-Bot research (pre/post survey methodology) and LabGPT. Maintaining this relationship deepens the Studio’s research credibility.

Tags

[STUDIO-BUILD] [CO-CREATE] — supports Goal 2 (Strategic Influence in AI+Education) via direct research collaboration.


7. Connection to Other Projects

Project Connection
LabGPT Also built for Cathy; migrated from Vercel to Firebase at labgpt.web.app. She may ask about condition bleed status
VFT (Sarah Williams-Habibi) Moderator bot code shared as a template for AI feedback implementation
Build-a-Bot Cathy interested in research components; connected via pre/post survey methodology

Last updated: March 2026

Source: projects/briefings/ai-comic-studio.md