Comic Builder · AI Science-Popularization Comic Generator
Turn "Quantum Entanglement / Relativity / Evolution" into 6 readable comic panels with one sentence
Comic Builder is an AI tool that industrializes the production of multi-panel science-popularization comics from a one-sentence topic. Built on the baoyu-comic skill, it first uses an LLM to parse the requirement into content analysis + 6-segment storyboard, then generates an image prompt per panel, calls the image-generation API panel-by-panel, and finally stitches all panels into a single comic. The full progress is pushed to the frontend via SSE in real time, with a 6-panel typical task completing in 1-3 minutes. Designed to turn abstract science / engineering / business concepts into publishable visual content.
Content Industrialization Matrix
Comic Builder focuses on "6-panel visual science comics"; yanxuan-story focuses on "9-step short-story pipeline"; novel-writing focuses on "30-step long-form creation" — the three form a "long-form → short-form → visual" content industrialization matrix.
Core Modules
Generation Progress (Content Analysis → Storyboard → Prompt → Incremental Panel N → Stitched Comic)
6-Panel Comic Output · Brief History of Evolution (XiaoAi & Guangsu)
Design Philosophy
Decompose "generate comic" into 9 independently observable, rerunnable, debuggable engineering steps, so "one sentence to comic" is not a black box.
Content Analysis Before Drawing
On receiving a requirement, the first step is not to draw but to let the LLM distill the topic, knowledge-point structure, concept relations, and narrative rhythm — only after content analysis passes do we move to storyboard, avoiding "fix the setup after drawing".
Storyboard & Image Prompt Decoupled
The storyboard only describes "what this panel says / who does what / camera / dialogue", and image prompts are derived independently by the prompt generator from storyboard + characters + style — readable storyboards, reproducible prompts, unified style.
Incremental Panel Generation
Call the image-generation API panel-by-panel from 1 to N, push each completed panel back to the frontend showing "Panel N complete (XXX KB)"; any failed panel can be retried independently without restart.
Auto-Stitch into a Complete Comic
Once all panels are generated, PIL auto-stitches them into a single complete comic with unified font size, panel borders, and style — the 6-panel deliverable is a single publishable PNG/JPG, no external layout needed.
Industry Status & Pain Points
Translating "expert knowledge" into "visual content the public can read" has long suffered 6 common pain points, each lowering reach and conversion.
Long & High-Threshold Science Copy
Traditional science articles run 3000-5000 words — unfriendly to the public; short videos are short but low-density and rewatch-costly; comics are the recognized best compromise yet hard to mass-produce.
Illustrator Capacity Limited & Costly
Hiring an illustrator for 6 panels takes 1-2 weeks and ¥1000+ per piece; revision communication cost is high, style is hard to unify, multi-topic serialization is nearly impossible.
Generic Image Gen "AI Flavor"
Using generic image models directly for comics, the character's look / clothing / pose drift each panel, dialogue misaligns with the picture, style is inconsistent — finished work is hard to publish.
Character & Style Hard to Unify
The same character must keep consistent look / clothing / props across 6 panels (XiaoAi + Guangsu the tabby with a pocket watch); generic models can't be panel-faithful, only single-panel compromises.
Generation Process Unobservable
Generating 6 panels typically takes 1-3 minutes; users can't see progress, don't know which panel is stuck, can't rerun a single panel — "click and wait 3 minutes" is a bad experience, users won't retry.
High Post-Layout Cost
After generating single panels, manual stitching, font-size adjustment, and unified panel borders take 1-2 hours; in batch production this step is the biggest bottleneck.
Core Capabilities · 9-Step Engineering Pipeline
From "one-sentence requirement" to "6-panel publishable comic", the full 9-step chain runs independently, observably, and rerunnably per step.
OpenAI-Compatible API Key
Support any OpenAI-compatible LLM + image-generation API (default agnes-ai agnes-2.5-pro + agnes-image-2.1-flash). Key stays in browser local storage, never on disk. Switch models by changing Base.
Natural-Language "Comic Requirement" Input
"Turn "Quantum Entanglement" into a 6-panel science comic. Characters: 14-year-old middle-schooler + tabby with pocket watch. Style: fresh." The more specific the requirement, the more stable the output. Two-segment "Elements + Requirements" prompting supported.
Configurable Panel Count (default 6)
Default 6 panels fits science / concept explanation; configurable 2 / 4 / 8 / 12 to adapt different topics and platform tones, with auto layout adjustment.
Multiple Styles (ligne-claire / default fresh / etc.)
Built-in multiple style enums (ligne-claire clean lines / children-book / flat illustration / retro watercolor / etc.) — pick style by topic, no per-panel drift.
Content Analysis + Storyboard
LLM first outputs "content analysis" (topic decomposition + knowledge-point list + character inventory + narrative arc), then breaks it into N "storyboard" segments (per-panel scene / dialogue / camera), with structured output before prompts.
Auto Image Prompt Generation
Auto-derive 6 high-quality image prompts from "storyboard + character profiles + style" (including character-consistency constraints + style keywords + camera language); each can be independently copied to any image model for verification.
Panel-by-Panel Incremental + Real-Time Progress
Generate 1→N panels sequentially; each completion pushes "Incremental Panel N complete (XXX KB)" to frontend via SSE. Any failed panel can be retried independently.
Auto-Stitch + Unified Layout
PIL stitches all panels into one complete comic with unified canvas size, font size, panel borders, and whitespace. Output is directly publishable, downloadable, and second-editable.
Quick Examples + Evolution History
Built-in 3 quick examples ("Core Concepts of Relativity" / "Quantum Entanglement 101" / "Brief History of Evolution") — one click fills the "Comic Requirement" and triggers generation, lowering first-use barrier.
9-Step Generation Flow · From Requirement to Comic
The 9-step full-chain engineering flow from "one-sentence requirement" to "6-panel complete comic": each step has independent prompt / log / rerun capability.
Key Outcomes · Industrial-Grade Comic Generation Capability
Upgrade "AI-generated comics" from one-off demo to engineered, reusable visual-science production line.
Key Breakthroughs
The difficulty is not "single panel drawing" but connecting "content → storyboard → prompt → incremental generation → stitching" into an observable, rerunnable, unified-style industrial production line.
Storyboard & Image Prompt Decoupled
The "storyboard" only describes "what this panel says / who does what / dialogue", and image prompts are derived independently by a generator from the storyboard — readable storyboards, reproducible prompts, unified style, avoiding the "AI flavor" and dialogue misalignment of generic models.
Incremental Panels + SSE Real-Time Progress
Generate 6 panels 1→N sequentially, each independently pushing "Incremental Panel N complete (XXX KB)" to the frontend; any failed panel can be retried independently (no full retry), and the entire process is fully observable.
Character Profiles + Style Unified Constraints
Extract "character profiles" (XiaoAi + Guangsu etc.) and "style" (ligne-claire etc.) from the requirement and inject as unified constraints into every prompt, keeping the same character consistent across 6 panels — solving the generic-model "AI flavor + character drift" twin pain points.
Auto-Stitch + Second-Editable
PIL auto-stitches into one complete comic with unified canvas / font size / panel border; output saves to outputs/<task_id>/, downloadable, second-editable — upgrading "6 separate AI images" to "1 publishable comic".
Business Value
Convert technical capability into quantifiable, reusable business value.
Content Under Active Update
Product content has been published based on internal materials. The following areas are planned for further development:
Other Products & Solutions
Explore Xianma AI solutions in other domains